<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:wfw="http://wellformedweb.org/CommentAPI/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom"
	xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
	xmlns:slash="http://purl.org/rss/1.0/modules/slash/"
	>

<channel>
	<title>Running Costs Archives - Puffin Notes</title>
	<atom:link href="https://thepuffingroup.com/category/running-costs/feed/" rel="self" type="application/rss+xml" />
	<link>https://thepuffingroup.com/category/running-costs/</link>
	<description>Open source, Linux and the real cost of running software</description>
	<lastBuildDate>Wed, 18 Nov 2026 09:32:22 +0000</lastBuildDate>
	<language>en-US</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	<generator>https://wordpress.org/?v=7.1.4</generator>

<image>
	<url>https://thepuffingroup.com/wp-content/uploads/2026/04/android-chrome-512x512-1-150x150.png</url>
	<title>Running Costs Archives - Puffin Notes</title>
	<link>https://thepuffingroup.com/category/running-costs/</link>
	<width>32</width>
	<height>32</height>
</image> 
	<item>
		<title>Azure Startup Credits for OpenAI GPT-6 Models: Costs Compared</title>
		<link>https://thepuffingroup.com/azure-startup-credits-for-openai-gpt/</link>
		
		<dc:creator><![CDATA[PuffingGroup]]></dc:creator>
		<pubDate>Wed, 18 Nov 2026 09:31:55 +0000</pubDate>
				<category><![CDATA[Running Costs]]></category>
		<guid isPermaLink="false">https://thepuffingroup.com/?p=1236</guid>

					<description><![CDATA[<p>Running Costs &#183; Cloud &#38; Infrastructure &#183; ~18 min read Microsoft will give an early-stage startup up to $150,000 in Azure credits, and those credits can be spent on OpenAI&#8217;s newest models. That turns a model choice that normally costs real money into one that costs, at least for a while, nothing out of pocket. [&#8230;]</p>
<p>The post <a rel="nofollow" href="https://thepuffingroup.com/azure-startup-credits-for-openai-gpt/">Azure Startup Credits for OpenAI GPT-6 Models: Costs Compared</a> appeared first on <a rel="nofollow" href="https://thepuffingroup.com">Puffin Notes</a>.</p>
<p>The post <a href="https://thepuffingroup.com/azure-startup-credits-for-openai-gpt/">Azure Startup Credits for OpenAI GPT-6 Models: Costs Compared</a> appeared first on <a href="https://thepuffingroup.com">Puffin Notes</a>.</p>
]]></description>
										<content:encoded><![CDATA[<!-- ============================================================ -->
<!-- PUFFIN NOTES — POST: AZURE STARTUP CREDITS & GPT-6 MODELS     -->
<!-- Format: Q&A with the editor. No H1 (theme renders the title) -->
<!-- ============================================================ -->

<style>
.az { font-family:'IBM Plex Sans', 'Segoe UI', Helvetica, Arial, sans-serif; }
.az-mono { font-family:'IBM Plex Mono', Consolas, 'Courier New', monospace; }
.az-label { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; color:#2a9d8f; margin:0 0 14px 0; }
.az-h2 { font-size:31px; font-weight:700; color:#16181d; line-height:1.25; margin:0 0 22px 0; letter-spacing:-0.6px; }
.az-p { font-size:17px; line-height:1.85; color:#2e333b; margin:0 0 20px 0; }
.az-p strong { color:#16181d; }
.az-p code, .az-a code { font-family:'IBM Plex Mono', Consolas, monospace; font-size:14px; background:#eef1f4; padding:1px 6px; border-radius:3px; color:#16181d; }
.az-note { font-size:13px; line-height:1.65; color:#8a929e; margin:12px 0 0 0; font-style:italic; }
.az-stats { display:grid; grid-template-columns:repeat(4, 1fr); gap:16px; }
.az-stat { background:#0d0f12; border:1px solid #2a2e36; border-radius:6px; padding:24px 20px; }
.az-stat-n { font-family:'IBM Plex Mono', Consolas, monospace; font-size:30px; font-weight:600; color:#e9a23b; margin:0 0 10px 0; line-height:1; }
.az-stat-l { font-size:13.5px; line-height:1.55; color:#a7afba; margin:0; }
.az-qa { padding:34px 0; border-top:1px solid #dfe3e8; }
.az-q { display:flex; gap:14px; align-items:flex-start; margin:0 0 18px 0; }
.az-q span { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; font-weight:600; color:#ffffff; background:#e9a23b; border-radius:4px; padding:5px 9px; line-height:1.2; flex-shrink:0; margin-top:3px; }
.az-q p { font-size:21px; font-weight:700; color:#16181d; line-height:1.4; margin:0; }
.az-ans { display:flex; gap:14px; align-items:flex-start; }
.az-ans > span { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; font-weight:600; color:#ffffff; background:#2a9d8f; border-radius:4px; padding:5px 9px; line-height:1.2; flex-shrink:0; margin-top:4px; }
.az-ans > div { flex:1; min-width:0; }
.az-quote { border-left:3px solid #e9a23b; background:#fbf6ec; padding:16px 22px; margin:0 0 20px 0; font-size:17px; line-height:1.65; color:#16181d; font-style:italic; }
.az-quote cite { display:block; font-family:'IBM Plex Mono', Consolas, monospace; font-style:normal; font-size:12px; color:#a8681a; margin-top:10px; }
.az-term { background:#0d0f12; border:1px solid #2a2e36; border-radius:6px; overflow:hidden; margin:4px 0 20px 0; }
.az-term-bar { background:#1f232a; padding:9px 14px; font-family:'IBM Plex Mono', Consolas, monospace; font-size:12px; color:#8a929e; }
.az-term pre { margin:0; padding:18px 20px; font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; line-height:1.75; color:#c9d1d9; overflow-x:auto; background:transparent; border:none; white-space:pre; }
.az-term .c { color:#6b7480; }
.az-term .p { color:#2a9d8f; }
.az-tbl { width:100%; border-collapse:collapse; background:#ffffff; border:1px solid #dfe3e8; font-size:14px; margin:8px 0; }
.az-tbl caption { text-align:left; font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; color:#2a9d8f; padding-bottom:12px; }
.az-tbl th { background:#16181d; color:#e6e8eb; padding:13px 14px; text-align:left; font-family:'IBM Plex Mono', Consolas, monospace; font-size:11.5px; font-weight:600; }
.az-tbl td { padding:12px 14px; border-bottom:1px solid #eceff2; color:#525a66; vertical-align:top; line-height:1.55; }
.az-tbl tr:last-child td { border-bottom:none; }
.az-tbl .b { font-weight:700; color:#16181d; }
.az-tbl .m { font-family:'IBM Plex Mono', Consolas, monospace; font-size:12.5px; color:#16181d; }
.az-tbl .num { text-align:right; font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; white-space:nowrap; }
.az-tbl .tot td { background:#fbf6ec; font-weight:700; color:#16181d; }
.az-tbl .good { color:#1f7a6f; font-weight:700; }
.az-tbl .bad { color:#b03a2e; font-weight:700; }
.az-gantt { background:#ffffff; border:1px solid #dfe3e8; border-radius:6px; padding:24px 24px 16px 24px; }
.az-g-row { display:grid; grid-template-columns:190px 1fr; gap:14px; align-items:center; margin:0 0 11px 0; }
.az-g-l { font-family:'IBM Plex Mono', Consolas, monospace; font-size:12.5px; color:#16181d; }
.az-g-t { position:relative; height:22px; background:#f5f7f8; border-radius:3px; }
.az-g-f { position:absolute; top:0; height:22px; border-radius:3px; font-family:'IBM Plex Mono', Consolas, monospace; font-size:11px; color:#ffffff; line-height:22px; padding-left:7px; white-space:nowrap; overflow:hidden; }
.az-g-axis { display:grid; grid-template-columns:190px 1fr; gap:14px; margin-top:6px; }
.az-g-axis div:last-child { display:flex; justify-content:space-between; font-family:'IBM Plex Mono', Consolas, monospace; font-size:11px; color:#8a929e; }
.az-bars { background:#ffffff; border:1px solid #dfe3e8; border-radius:6px; padding:24px 24px 12px 24px; }
.az-bar { display:grid; grid-template-columns:200px 1fr 80px; gap:14px; align-items:center; margin:0 0 12px 0; }
.az-bar-l { font-family:'IBM Plex Mono', Consolas, monospace; font-size:12.5px; color:#16181d; }
.az-bar-t { background:#eef1f4; height:22px; border-radius:3px; }
.az-bar-f { height:22px; border-radius:3px; background:#b03a2e; }
.az-bar-v { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13.5px; font-weight:600; color:#16181d; text-align:right; }
.az-check { background:#0d0f12; border-radius:8px; overflow:hidden; border:1px solid #2a2e36; }
.az-check-bar { background:#1f232a; padding:10px 16px; font-family:'IBM Plex Mono', Consolas, monospace; font-size:12px; color:#8a929e; }
.az-check-cols { display:grid; grid-template-columns:1fr 1fr; gap:0 30px; padding:18px 24px; }
.az-check h5 { font-family:'IBM Plex Mono', Consolas, monospace; font-size:12px; color:#e9a23b; margin:12px 0 6px 0; font-weight:600; }
.az-check ul { list-style:none; margin:0; padding:0; }
.az-check li { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; line-height:1.6; color:#c9d1d9; padding:4px 0 4px 30px; position:relative; }
.az-check li:before { content:'[ ]'; position:absolute; left:0; color:#2a9d8f; }
.az-faq { display:grid; grid-template-columns:1fr 1fr; gap:0 50px; }
.az-faq-item { padding:22px 0; border-top:1px solid #dfe3e8; }
.az-faq-q { font-family:'IBM Plex Mono', Consolas, monospace; font-size:15px; font-weight:600; color:#16181d; margin:0 0 10px 0; line-height:1.5; }
.az-a { font-size:15px; line-height:1.75; color:#525a66; margin:0; }
/* Override theme heading styles */
.az h1, .az h2, .az h3, .az h4, .az h5, .az p { text-transform:none !important; }
.az .az-h2 { color:#16181d !important; }
.stk-az-pq h2 { color:#ffffff !important; }
@media screen and (max-width:900px) { .az-stats { grid-template-columns:1fr 1fr; } .az-check-cols { grid-template-columns:1fr; } }
@media screen and (max-width:768px) {
  .az-faq { grid-template-columns:1fr; }
  .az-g-row, .az-g-axis { grid-template-columns:1fr; gap:4px; }
  .az-bar { grid-template-columns:1fr; gap:4px; }
  .az-bar-v { text-align:left; }
  .az-tbl th, .az-tbl td { padding:8px 8px; font-size:12px; }
  .az-ans > span, .az-q span { display:none; }
}
@media screen and (max-width:520px) { .az-stats { grid-template-columns:1fr; } }
</style>


<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-hero stk-block-background" data-block-id="az-hero"><style>.stk-az-hero {background-color:#16181d !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:920px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<p class="az-label">Running Costs &middot; Cloud &amp; Infrastructure &middot; ~18 min read</p>
<p style="font-size:21px; line-height:1.7; color:#c9d1d9; margin:0 0 26px 0;">Microsoft will give an early-stage startup up to $150,000 in Azure credits, and those credits can be spent on OpenAI&rsquo;s newest models. That turns a model choice that normally costs real money into one that costs, at least for a while, nothing out of pocket. We asked Puffin Notes editor Owen Calder how the credits actually work, which GPT-6 model to point them at, and where founders quietly waste them.</p>
<p class="az-mono" style="font-size:12.5px; color:#6b7480; margin:0;">Q&amp;A WITH <span style="color:#f2f3f5;">OWEN CALDER</span> &middot; NOVEMBER 2026 &middot; PRICES CHECKED THIS MONTH</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-stats stk-block-background" data-block-id="az-stats"><style>.stk-az-stats {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1100px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<p class="az-label">$ cat summary.txt</p>
<div class="az-stats">
  <div class="az-stat"><p class="az-stat-n">$150K</p><p class="az-stat-l">Maximum Azure credits through Microsoft for Startups, unlocked in stages</p></div>
  <div class="az-stat"><p class="az-stat-n">$5,000</p><p class="az-stat-l">Unlocked after business verification &mdash; the amount most founders actually start with</p></div>
  <div class="az-stat"><p class="az-stat-n">80&times;</p><p class="az-stat-l">Output price gap between GPT-6.2 Astra ($40) and GPT-6 Luna ($0.50) per million tokens</p></div>
  <div class="az-stat"><p class="az-stat-n">$0</p><p class="az-stat-l">Cash cost of running GPT-6 models on Azure while credits last</p></div>
</div>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p1 stk-block-background" data-block-id="az-p1"><style>.stk-az-p1 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<p class="az-label">// part one &middot; how the money works</p><h2 class="az-h2">Credits are not cash. They behave like a budget with an expiry date.</h2><div class="az-qa"><div class="az-q"><span>Q01</span><p>Let&rsquo;s start simple. What do startups actually get from Microsoft?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">Under the current Microsoft for Startups programme, you get approved, you get a starter amount of credit to build and test with, and then you unlock up to <strong>$5,000</strong> once Microsoft has verified that your business is real. After that, more credits unlock as your Azure usage grows and stays steady, up to a ceiling of <strong>$150,000</strong>. Microsoft describes it as one programme that grows with you, and that&rsquo;s a fair summary.</p><p class="az-p">The part people skim past is the shape of it. The $150,000 is a ceiling, not a deposit. Most teams I talk to are living on the first few thousand for months, and the larger tranches only show up once you&rsquo;re running a real workload on Azure. If you plan your first year around the headline figure, you&rsquo;re planning around money you haven&rsquo;t earned yet.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q02</span><p>And those credits really cover OpenAI models?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">Yes &mdash; that&rsquo;s the bit that makes the programme interesting in 2026. Microsoft lists Azure OpenAI alongside Foundry Tools and Azure Machine Learning as eligible AI services, next to the boring infrastructure you&rsquo;d expect: virtual machines, storage, networking, App Service and Kubernetes. So the same credit balance that pays for your database can pay for GPT-6 tokens.</p><p class="az-p">That&rsquo;s a genuinely different deal from going direct. On OpenAI&rsquo;s own platform you pay by card, by the token, from day one. On Azure, inside the credit window, the token bill is real but it&rsquo;s deducted from credits, not from your bank account.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q03</span><p>What happens when the credits run out?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">Nothing dramatic, which is exactly the danger. Microsoft is clear that your workloads keep running and you move to normal pay-as-you-go pricing. There&rsquo;s no cliff, no shutdown, no migration. There&rsquo;s also no moment that forces you to look at the bill. I&rsquo;ve seen teams discover a five-figure invoice because the credits ran out mid-month and the model they&rsquo;d been happily burning on was the most expensive one in the catalogue.</p><p class="az-p">My advice: set a budget alert in Azure Cost Management at 50%, 75% and 90% of your credit balance on day one. It takes five minutes and it&rsquo;s the cheapest insurance you&rsquo;ll buy all year.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t0 stk-block-background" data-block-id="az-t0"><style>.stk-az-t0 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 1 &mdash; How Microsoft for Startups credits unlock</caption>
<thead><tr><th>Stage</th><th>What triggers it</th><th class="num">Credits</th><th>What it realistically pays for</th></tr></thead>
<tbody>
<tr><td class="b">Approval</td><td>Accepted into the programme</td><td class="num">starter amount</td><td>Prototyping, a dev VM, first API calls</td></tr>
<tr><td class="b">Verified</td><td>Business verification</td><td class="num">up to $5,000</td><td>An MVP with modest model traffic</td></tr>
<tr><td class="b">Growing</td><td>More Azure services, sustained usage</td><td class="num">additional tranches</td><td>Early production, real users</td></tr>
<tr class="tot"><td>Ceiling</td><td>Scaling on Azure over time</td><td class="num">up to $150,000</td><td>A serious production workload &mdash; for a while</td></tr>
</tbody>
</table>
<p class="az-note">Source: Microsoft for Startups programme FAQ, November 2026. Exact tranche sizes depend on eligibility and usage; check your portal.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p2 stk-block-background" data-block-id="az-p2"><style>.stk-az-p2 {background-color:#f5f7f8 !important; border-top:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<p class="az-label">// part two &middot; the model menu</p><h2 class="az-h2">Four models, one family, very different bills</h2><div class="az-qa"><div class="az-q"><span>Q04</span><p>OpenAI&rsquo;s lineup has changed a lot this year. What&rsquo;s actually on the menu?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">The current generation is easy to read once you ignore the naming. There are four general-purpose models, and they&rsquo;re named after a size ladder rather than a version number. <strong>GPT-6.2 Astra</strong> is the flagship for the hardest work. <strong>GPT-6.2 Sol</strong> is the everyday workhorse. <strong>GPT-6.1 Terra</strong> is the fast, efficient one with a strong line in visual reasoning and computer use. <strong>GPT-6 Luna</strong> is the cheap, high-volume option.</p><p class="az-p">What surprised me is how much they share. All four have the same 1.05 million token context window, the same 128K maximum output, the same tool support &mdash; functions, web search, file search, computer use &mdash; and all of them do text and image input. The difference is how hard they think and what they charge for it.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t1 stk-block-background" data-block-id="az-t1"><style>.stk-az-t1 {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1050px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 2 &mdash; The GPT-6 lineup at OpenAI&rsquo;s direct prices (per 1M tokens)</caption>
<thead><tr><th>Model</th><th>Positioning</th><th class="num">Input</th><th class="num">Cached input</th><th class="num">Output</th><th>Knowledge cutoff</th></tr></thead>
<tbody>
<tr><td class="b">GPT-6.2 Astra</td><td>Hardest projects, start to finish</td><td class="num">$8.00</td><td class="num">$0.40</td><td class="num">$40.00</td><td>18 May 2026</td></tr>
<tr><td class="b">GPT-6.2 Sol</td><td>Everyday work, agents at scale</td><td class="num">$2.00</td><td class="num">$0.10</td><td class="num">$10.00</td><td>1 Jun 2026</td></tr>
<tr><td class="b">GPT-6.1 Terra</td><td>Fast; visual reasoning, computer use, writing</td><td class="num">$1.00</td><td class="num">&mdash;</td><td class="num">$5.00</td><td>1 Jun 2026</td></tr>
<tr><td class="b">GPT-6 Luna</td><td>Focused, high-volume tasks</td><td class="num">$0.10</td><td class="num">$0.01</td><td class="num">$0.50</td><td>18 May 2026</td></tr>
</tbody>
</table>
<p class="az-note">Source: OpenAI API model catalogue, November 2026. All four: 1.05M context, 128K max output. Terra&rsquo;s cached rate wasn&rsquo;t in the catalogue view we used.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p2b stk-block-background" data-block-id="az-p2b"><style>.stk-az-p2b {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<div class="az-qa"><div class="az-q"><span>Q05</span><p>So when should someone actually reach for Astra?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">When a wrong answer costs more than the tokens. Long, multi-step engineering work where the model has to hold a plan for an hour. Legal or financial analysis where you&rsquo;d otherwise pay a human to check the output. Agentic tasks with computer use where a mistake halfway through means starting over. Astra&rsquo;s reasoning goes all the way up to a &ldquo;max&rdquo; setting, and on genuinely hard problems that extra thinking pays for itself.</p><p class="az-p">When <em>not</em> to: anything you&rsquo;d describe as &ldquo;just&rdquo;. Just summarise this. Just tag these tickets. Just rewrite the email. I see Astra used as a default far more than it should be, usually because someone picked it in a prototype and never revisited the choice. At $40 per million output tokens, defaults are expensive.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q06</span><p>And Sol and Terra? They sound close.</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">They are close, and that&rsquo;s where most of the interesting decisions are. <strong>Sol</strong> is what I&rsquo;d put behind most production features: a chat assistant, a coding helper, an agent that calls a handful of tools. It&rsquo;s a quarter of Astra&rsquo;s price and, in my experience, good enough for the large majority of real work.</p><p class="az-p"><strong>Terra</strong> is half the price of Sol again, and it&rsquo;s the one I&rsquo;d test first for anything visual &mdash; reading screenshots, filling in forms through computer use, extracting data from scanned documents &mdash; and for writing that needs to sound natural rather than clever. It also has a &ldquo;none&rdquo; reasoning setting, which Astra and Sol don&rsquo;t, so for quick interactive replies you can turn the thinking off entirely and keep latency low.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q07</span><p>Luna at ten cents per million input tokens sounds almost free. What&rsquo;s the catch?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">There isn&rsquo;t much of one, if you use it for what it&rsquo;s built for. Classification, routing, extraction into a fixed schema, moderation pre-filters, short summaries &mdash; tasks where the answer is narrow and you can check it automatically. Luna also gets much higher rate limits: 180 million tokens per minute at OpenAI&rsquo;s top tier, against 40 million for Astra and Sol. For batch jobs, that matters as much as price.</p><p class="az-p">Where it falls down is open-ended reasoning. Ask Luna to plan a migration or debug a race condition and you&rsquo;ll get a confident, plausible answer that&rsquo;s wrong in the third paragraph. The best architectures I see use Luna as the front door and only escalate to Sol or Astra when the task needs it.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t2 stk-block-background" data-block-id="az-t2"><style>.stk-az-t2 {background-color:#f5f7f8 !important; border-bottom:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1050px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 3 &mdash; When to use each model, and when not to</caption>
<thead><tr><th>Model</th><th>Use it for</th><th>Avoid it for</th><th>Reasoning settings</th></tr></thead>
<tbody>
<tr><td class="b">GPT-6.2 Astra</td><td>Long agentic tasks, hard code, high-stakes analysis</td><td class="bad">Summaries, tagging, chat replies, anything &ldquo;just&rdquo;</td><td class="m">low &rarr; max</td></tr>
<tr><td class="b">GPT-6.2 Sol</td><td>Production assistants, coding help, tool-calling agents</td><td>Bulk classification where Luna is enough</td><td class="m">low &rarr; max</td></tr>
<tr><td class="b">GPT-6.1 Terra</td><td>Screenshots, forms, computer use, natural writing, low latency</td><td>Deep multi-step planning</td><td class="m">none &rarr; max</td></tr>
<tr><td class="b">GPT-6 Luna</td><td>Routing, extraction, moderation, batch jobs</td><td class="bad">Open-ended reasoning, debugging, planning</td><td class="m">none &rarr; max</td></tr>
</tbody>
</table>
<p class="az-note">Puffin Notes guidance, based on OpenAI&rsquo;s positioning and what we see in practice. Test on your own data before committing.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p3 stk-block-background" data-block-id="az-p3"><style>.stk-az-p3 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<p class="az-label">// part three &middot; openai direct vs azure</p><h2 class="az-h2">Same models. Very different invoice.</h2><div class="az-qa"><div class="az-q"><span>Q08</span><p>Is running these models on Azure cheaper than going to OpenAI directly?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">At list price, no &mdash; and occasionally it&rsquo;s a bit dearer. Microsoft&rsquo;s published Foundry prices for the GPT-6 family on its Global Standard tier are $2 in and $10 out per million tokens for Sol, and $0.10 and $0.50 for Luna. That&rsquo;s the same as OpenAI. For the original GPT-6 Astra, Azure lists $10 and $50, which is higher than OpenAI&rsquo;s current $8 and $40 for GPT-6.2 Astra. Pin a US Data Zone deployment for data residency and Microsoft adds a 10% premium on top.</p><p class="az-p">But that&rsquo;s the wrong comparison for a startup with credits. The right comparison is cash leaving your account. Direct, you pay every token. On Azure, inside the credit window, you pay <strong>zero</strong>. The list price only decides how fast the credits burn.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t3 stk-block-background" data-block-id="az-t3"><style>.stk-az-t3 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1050px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 4 &mdash; Price per 1M tokens: OpenAI direct vs Azure list vs Azure on credits</caption>
<thead><tr><th>Model</th><th class="num">OpenAI direct<br>in / out</th><th class="num">Azure Global Standard<br>in / out</th><th class="num">Azure US Data Zone</th><th class="num">Your cash cost<br>on credits</th></tr></thead>
<tbody>
<tr><td class="b">Astra</td><td class="num">$8 / $40 <span class="m">(6.2)</span></td><td class="num">$10 / $50 <span class="m">(6)</span></td><td class="num">+10%</td><td class="num good">$0</td></tr>
<tr><td class="b">Sol</td><td class="num">$2 / $10</td><td class="num">$2 / $10</td><td class="num">+10%</td><td class="num good">$0</td></tr>
<tr><td class="b">Terra</td><td class="num">$1 / $5</td><td class="num">check catalogue</td><td class="num">+10%</td><td class="num good">$0</td></tr>
<tr><td class="b">Luna</td><td class="num">$0.10 / $0.50</td><td class="num">$0.10 / $0.50</td><td class="num">+10%</td><td class="num good">$0</td></tr>
</tbody>
</table>
<p class="az-note">Azure prices are Microsoft&rsquo;s published Global Standard short-context rates for the GPT-6 family (September 2026). Long-context requests above 272K input tokens cost more on both platforms. Model versions reach Foundry on their own schedule, so confirm which version &mdash; 6, 6.1 or 6.2 &mdash; is deployable in your region before you plan around it.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p3b stk-block-background" data-block-id="az-p3b"><style>.stk-az-p3b {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<div class="az-qa"><div class="az-q"><span>Q09</span><p>Put some numbers on it. What does a real workload cost?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">I modelled three workloads I see constantly, at OpenAI&rsquo;s direct prices with no caching, so the comparison is clean. A support assistant handling 50,000 conversations a month, each about 3,000 tokens in and 500 out. A coding agent doing 2,000 tasks a month at 200,000 tokens in and 20,000 out, reasoning included. And a classification job pushing two million documents a month at 1,500 tokens in and 50 out.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t4 stk-block-background" data-block-id="az-t4"><style>.stk-az-t4 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1050px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 5 &mdash; Monthly cost of three workloads by model (OpenAI direct prices)</caption>
<thead><tr><th>Workload</th><th class="num">Astra</th><th class="num">Sol</th><th class="num">Terra</th><th class="num">Luna</th></tr></thead>
<tbody>
<tr><td class="b">Support assistant<br><span class="m">50K chats &middot; 3K in / 500 out</span></td><td class="num">$2,200</td><td class="num">$550</td><td class="num">$275</td><td class="num">$27.50</td></tr>
<tr><td class="b">Coding agent<br><span class="m">2K tasks &middot; 200K in / 20K out</span></td><td class="num">$4,800</td><td class="num">$1,200</td><td class="num">$600</td><td class="num">$60</td></tr>
<tr><td class="b">Document classifier<br><span class="m">2M docs &middot; 1.5K in / 50 out</span></td><td class="num bad">$28,000</td><td class="num">$7,000</td><td class="num">$3,500</td><td class="num good">$350</td></tr>
</tbody>
</table>
<p class="az-note">Our calculation. Reasoning tokens are billed as output, so real coding-agent costs vary widely with the reasoning setting.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p3c stk-block-background" data-block-id="az-p3c"><style>.stk-az-p3c {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<p class="az-p" style="margin-top:14px;">Look at the classifier row. Running that job on Astra costs eighty times what it costs on Luna, and for a narrow labelling task you won&rsquo;t get eighty times the quality. You probably won&rsquo;t get twice the quality. That&rsquo;s the single most common waste I see, and credits hide it, because nobody feels a $28,000 month that&rsquo;s paid in credits.</p><div class="az-qa"><div class="az-q"><span>Q10</span><p>How far does a credit balance actually stretch?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">Less far than the headline suggests, if you choose the wrong model. Here&rsquo;s how many tokens $5,000 buys, assuming a typical mix of three input tokens for every output token.</p><p class="az-p">One thing to keep in mind as you read it: the credit balance isn&rsquo;t ring-fenced for AI. The same pot pays for your virtual machines, your managed database, your storage and your bandwidth. In the startups I&rsquo;ve looked at, ordinary infrastructure takes somewhere between a third and a half of the monthly credit draw before a single token is generated. So whatever runway the model numbers suggest, assume the real figure is shorter, and check Cost Management by service rather than by total.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-bars stk-block-background" data-block-id="az-bars"><style>.stk-az-bars {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<div class="az-bars"><p class="az-mono" style="font-size:12px; color:#2a9d8f; margin:0 0 16px 0;">tokens bought with $5,000 of credit (3:1 input:output mix, OpenAI direct prices)</p><div class="az-bar"><span class="az-bar-l">GPT-6.2 Astra</span><div class="az-bar-t"><div class="az-bar-f" style="width:1.25%; background:#b03a2e;"></div></div><span class="az-bar-v">312M</span></div><div class="az-bar"><span class="az-bar-l">GPT-6.2 Sol</span><div class="az-bar-t"><div class="az-bar-f" style="width:5%; background:#e9a23b;"></div></div><span class="az-bar-v">1.25B</span></div><div class="az-bar"><span class="az-bar-l">GPT-6.1 Terra</span><div class="az-bar-t"><div class="az-bar-f" style="width:10%; background:#2a9d8f;"></div></div><span class="az-bar-v">2.5B</span></div><div class="az-bar"><span class="az-bar-l">GPT-6 Luna</span><div class="az-bar-t"><div class="az-bar-f" style="width:100%; background:#1f7a6f;"></div></div><span class="az-bar-v">25B</span></div></div>
<p class="az-note">Our calculation: blended cost per million tokens is $16 for Astra, $4 for Sol, $2 for Terra and $0.20 for Luna. At the $150,000 ceiling, multiply by 30: about 9.4 billion Astra tokens, or 750 billion Luna tokens.</p>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t5 stk-block-background" data-block-id="az-t5"><style>.stk-az-t5 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 6 &mdash; How many months the credits last for the support assistant</caption>
<thead><tr><th>Model</th><th class="num">Monthly bill</th><th class="num">$5,000 lasts</th><th class="num">$150,000 lasts</th></tr></thead>
<tbody>
<tr><td class="b">GPT-6.2 Astra</td><td class="num">$2,200</td><td class="num bad">2.3 months</td><td class="num">5.7 years</td></tr>
<tr><td class="b">GPT-6.2 Sol</td><td class="num">$550</td><td class="num">9.1 months</td><td class="num">22.7 years</td></tr>
<tr><td class="b">GPT-6.2 Sol, prompt cached</td><td class="num">$360</td><td class="num">13.9 months</td><td class="num">34.7 years</td></tr>
<tr><td class="b">GPT-6 Luna</td><td class="num">$27.50</td><td class="num good">15 years</td><td class="num good">effectively forever</td></tr>
</tbody>
</table>
<p class="az-note">Our calculation, model token costs only &mdash; your VMs, storage and databases draw on the same balance, so real runway is shorter. Cached row assumes 2,000 of each 3,000 input tokens are a reused system prompt.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-pq stk-block-background" data-block-id="az-pq"><style>.stk-az-pq {background-color:#2a9d8f !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<p class="az-mono" style="font-size:13px; color:#d5f0ec; margin:0 0 16px 0;">$ echo $LESSON</p>
<h2 style="font-size:29px; font-weight:600; color:#ffffff; line-height:1.4; margin:0; letter-spacing:-0.4px;">&ldquo;Free tokens don&rsquo;t make a bad model choice cheap. They make it invisible until the day they run out.&rdquo;</h2>
<p class="az-mono" style="font-size:12px; color:#d5f0ec; margin:18px 0 0 0;">&mdash; Owen Calder</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p4 stk-block-background" data-block-id="az-p4"><style>.stk-az-p4 {background-color:#f5f7f8 !important; border-top:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<p class="az-label">// part four &middot; where credits leak</p><h2 class="az-h2">The quiet ways a credit balance disappears</h2><div class="az-qa"><div class="az-q"><span>Q11</span><p>You mentioned reasoning tokens. Why do they matter so much?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">Because they&rsquo;re billed as output, and output is the expensive side. When you turn Astra or Sol up to &ldquo;high&rdquo; or &ldquo;max&rdquo;, the model does its thinking in tokens you never see but absolutely pay for. On a hard coding task the hidden reasoning can be several times longer than the visible answer.</p><p class="az-p">My rule of thumb: start every new feature at the lowest reasoning setting that passes your tests, then raise it only where the tests fail. Most teams do the opposite &mdash; they start at max because the demo looked better, and never come back down.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q12</span><p>What about caching? Is it worth the effort?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">It&rsquo;s the best discount in the price list. Cached input on Astra and Sol costs 5% of the normal input price &mdash; $0.40 instead of $8 on Astra, $0.10 instead of $2 on Sol. If your app sends the same long system prompt, the same tool definitions or the same reference document on every call, put that stable part at the start of the prompt so it can be cached.</p><p class="az-p">There is a small catch: writing to the cache costs a little more than normal input &mdash; $10 instead of $8 per million on Astra. So caching pays off on prompts you reuse, not on one-offs. For the support assistant above, caching the system prompt cuts the Sol bill from $550 to $360 a month. That&rsquo;s a third off for an afternoon of work.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q13</span><p>Anything Azure-specific people get wrong?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">Three things. First, deployment type. Global Standard is the cheapest; Data Zone deployments keep processing inside the US or EU but cost about 10% more. If you don&rsquo;t need residency, don&rsquo;t pay for it with your credits. Second, long context. Both platforms charge more once input goes past 272,000 tokens &mdash; on Azure, Astra roughly doubles its input price. A 1.05 million token window is a capability, not a target.</p><p class="az-p">Third, versions. OpenAI ships a new point release and the Azure catalogue follows on its own timetable, region by region. Don&rsquo;t write &ldquo;we&rsquo;ll use 6.2 Sol on Azure&rdquo; into a plan until you&rsquo;ve actually deployed it in the region you need.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-t6 stk-block-background" data-block-id="az-t6"><style>.stk-az-t6 {background-color:#f5f7f8 !important; border-bottom:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<table class="az-tbl">
<caption>Table 7 &mdash; Credit leaks, ranked by how much they usually cost</caption>
<thead><tr><th>Leak</th><th>Typical overspend</th><th>Fix</th></tr></thead>
<tbody>
<tr><td class="b">Flagship model on a narrow task</td><td class="bad">up to 80&times;</td><td>Route simple work to Luna or Terra</td></tr>
<tr><td class="b">Reasoning left on max</td><td class="bad">2&ndash;5&times; on output</td><td>Start low, raise only where tests fail</td></tr>
<tr><td class="b">Long system prompt not cached</td><td>~35% on chat workloads</td><td>Put stable content first; reuse it</td></tr>
<tr><td class="b">Data Zone without a residency need</td><td>+10%</td><td>Use Global Standard</td></tr>
<tr><td class="b">Prompts over 272K tokens</td><td>~2&times; input price</td><td>Retrieve the relevant chunks instead</td></tr>
<tr><td class="b">Idle infrastructure on the same balance</td><td>varies</td><td>Budget alerts; delete what you don&rsquo;t use</td></tr>
</tbody>
</table>
<p class="az-note">Puffin Notes estimates from the published price lists. Your mileage will vary with prompt shape and traffic.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-p5 stk-block-background" data-block-id="az-p5"><style>.stk-az-p5 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">
<p class="az-label">// part five &middot; strategy</p><h2 class="az-h2">Using the window well</h2><div class="az-qa"><div class="az-q"><span>Q14</span><p>If you were a founder with fresh credits tomorrow, what would you do?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">I&rsquo;d treat the credit window as a research budget, not a subsidy. Use it to run the same real traffic through Astra, Sol, Terra and Luna, side by side, and measure which is the cheapest model that&rsquo;s good enough for each task. That experiment would cost real money on OpenAI direct. On credits it&rsquo;s free, and the answer is worth more than the credits themselves, because it sets your cost structure for the next five years.</p><p class="az-p">Then I&rsquo;d build routing in from day one: a cheap model classifies the request, and only the hard ones go up the ladder. Teams that leave this until after the credits run out have to do it in a panic, with real customers watching.</p></div></div></div><div class="az-qa"><div class="az-q"><span>Q15</span><p>And what if a company simply can&rsquo;t use all of them?</p></div><div class="az-ans"><span>OC</span><div><p class="az-p">It happens more than you&rsquo;d think. A team pivots away from AI, moves to another cloud for a customer contract, or just builds something leaner than they expected. The balance sits there until it expires. Before you write that off, check your programme terms and what your options are. Got unused credits? There are marketplaces where you can <a href="https://aicreditmart.com/sell-azure-credits/" style="color:#1f7a6f; text-decoration:underline;" target="_blank" rel="noopener">sell your Azure startup credits</a> rather than letting them lapse.</p><p class="az-p">Whatever you decide, decide it before the expiry date, not after. Every credit programme I&rsquo;ve seen is generous on the way in and completely unforgiving on the way out.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-faq stk-block-background" data-block-id="az-faq"><style>.stk-az-faq {background-color:#f5f7f8 !important; border-top:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1050px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<p class="az-label">$ man azure-credits</p>
<h2 class="az-h2" style="margin-bottom:24px;">Frequently asked questions</h2>
<div class="az-faq"><div><div class="az-faq-item"><p class="az-faq-q">How much Azure credit can a startup get from Microsoft?</p><p class="az-a">Up to $150,000 through Microsoft for Startups, unlocked over time. Founders start with a smaller amount and can unlock up to $5,000 after business verification; more becomes available with sustained Azure usage.</p></div><div class="az-faq-item"><p class="az-faq-q">Which OpenAI models are current in November 2026?</p><p class="az-a">GPT-6.2 Astra (flagship), GPT-6.2 Sol (everyday), GPT-6.1 Terra (fast, visual and computer use) and GPT-6 Luna (high-volume, low cost).</p></div><div class="az-faq-item"><p class="az-faq-q">Is Azure OpenAI cheaper than OpenAI direct?</p><p class="az-a">At list price, Sol and Luna cost the same. Azure&rsquo;s GPT-6 Astra is listed higher than OpenAI&rsquo;s GPT-6.2 Astra. With credits, your cash cost on Azure is zero until the balance runs out.</p></div><div class="az-faq-item"><p class="az-faq-q">What is the cheapest GPT-6 model?</p><p class="az-a">GPT-6 Luna, at $0.10 per million input tokens and $0.50 per million output tokens, with the highest rate limits.</p></div><div class="az-faq-item"><p class="az-faq-q">Are reasoning tokens charged?</p><p class="az-a">Yes, as output tokens. Higher reasoning settings can multiply the output bill, so start low and raise it only where needed.</p></div><div class="az-faq-item"><p class="az-faq-q">What is the context window of the GPT-6 models?</p><p class="az-a">1.05 million tokens for all four, with up to 128,000 output tokens. Requests above 272,000 input tokens are priced higher.</p></div><div class="az-faq-item"><p class="az-faq-q">What happens when Azure credits run out?</p><p class="az-a">Your workloads keep running and move to pay-as-you-go pricing. There is no shutdown, which is why budget alerts matter.</p></div><div class="az-faq-item"><p class="az-faq-q">Can I use credits if I already use another cloud?</p><p class="az-a">Yes. Microsoft says you can join while using another provider and move workloads to Azure gradually.</p></div></div><div><div class="az-faq-item"><p class="az-faq-q">Can Azure startup credits be used for OpenAI models?</p><p class="az-a">Yes. Microsoft lists Azure OpenAI among the eligible AI services, alongside Foundry Tools and Azure Machine Learning.</p></div><div class="az-faq-item"><p class="az-faq-q">Is GPT-6.2 available on Azure?</p><p class="az-a">Azure adds new OpenAI versions on its own timetable and region by region. Check the Microsoft Foundry model catalogue for your region before planning around a specific version.</p></div><div class="az-faq-item"><p class="az-faq-q">What does GPT-6.2 Astra cost?</p><p class="az-a">On OpenAI direct, $8 per million input tokens, $0.40 cached and $40 per million output tokens.</p></div><div class="az-faq-item"><p class="az-faq-q">When should I use GPT-6.1 Terra instead of Sol?</p><p class="az-a">For visual tasks, computer use, natural writing and low-latency replies. It is half Sol&rsquo;s price and can run with reasoning switched off.</p></div><div class="az-faq-item"><p class="az-faq-q">How much does prompt caching save?</p><p class="az-a">Cached input costs 5% of the normal input price on Astra and Sol. On a chat workload with a long, reused system prompt, that can cut the total bill by around a third.</p></div><div class="az-faq-item"><p class="az-faq-q">What is the Azure Data Zone premium?</p><p class="az-a">Microsoft prices US Data Zone deployments at a 10% premium to Global Standard, in exchange for keeping processing within the zone.</p></div><div class="az-faq-item"><p class="az-faq-q">Do Azure startup credits expire?</p><p class="az-a">Credit grants typically carry an expiry date. Check the date shown in your Microsoft for Startups portal and plan usage around it.</p></div></div></div>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-az-end stk-block-background" data-block-id="az-end"><style>.stk-az-end {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks az">

<p style="font-size:13px; line-height:1.75; color:#8a929e; margin:0; font-style:italic;">Sources: Microsoft for Startups programme page and FAQ; OpenAI API model catalogue and per-model pricing pages; Microsoft Azure blog on GPT-6 Astra, Sol and Luna in Microsoft Foundry (September 2026). Workload and runway figures are our own illustrative calculations. Prices change often &mdash; check the current price lists before you commit to a budget. Puffin Notes has no commercial relationship with Microsoft or OpenAI.</p>

</div></div></div>
</div></div>


<!-- END --><p>The post <a rel="nofollow" href="https://thepuffingroup.com/azure-startup-credits-for-openai-gpt/">Azure Startup Credits for OpenAI GPT-6 Models: Costs Compared</a> appeared first on <a rel="nofollow" href="https://thepuffingroup.com">Puffin Notes</a>.</p>
<p>The post <a href="https://thepuffingroup.com/azure-startup-credits-for-openai-gpt/">Azure Startup Credits for OpenAI GPT-6 Models: Costs Compared</a> appeared first on <a href="https://thepuffingroup.com">Puffin Notes</a>.</p>
]]></content:encoded>
					
		
		
			</item>
		<item>
		<title>Cloud Egress Costs 2026: What It Really Costs to Leave AWS, Azure &#038; GCP</title>
		<link>https://thepuffingroup.com/cloud-egress-costs-2026-what-it-really-costs-to-leave-aws-azure-gcp/</link>
		
		<dc:creator><![CDATA[PuffingGroup]]></dc:creator>
		<pubDate>Wed, 18 Nov 2026 09:08:51 +0000</pubDate>
				<category><![CDATA[Open Source & Linux]]></category>
		<category><![CDATA[Running Costs]]></category>
		<guid isPermaLink="false">https://thepuffingroup.com/?p=1227</guid>

					<description><![CDATA[<p>Cloud &#38; Infrastructure &#183; Running Costs &#183; ~18 min read We priced what it costs to move 10 TB out of the major clouds, added the charges that don&#8217;t appear on the pricing page, and checked what the EU Data Act really changes in January 2027. Short version: the law fixes one scenario. Your bill [&#8230;]</p>
<p>The post <a rel="nofollow" href="https://thepuffingroup.com/cloud-egress-costs-2026-what-it-really-costs-to-leave-aws-azure-gcp/">Cloud Egress Costs 2026: What It Really Costs to Leave AWS, Azure &#038; GCP</a> appeared first on <a rel="nofollow" href="https://thepuffingroup.com">Puffin Notes</a>.</p>
<p>The post <a href="https://thepuffingroup.com/cloud-egress-costs-2026-what-it-really-costs-to-leave-aws-azure-gcp/">Cloud Egress Costs 2026: What It Really Costs to Leave AWS, Azure &#038; GCP</a> appeared first on <a href="https://thepuffingroup.com">Puffin Notes</a>.</p>
]]></description>
										<content:encoded><![CDATA[<!-- ============================================================ -->
<!-- PUFFIN NOTES — POST: EGRESS IS THE SECOND LOCK-IN            -->
<!-- Only ONE outbound link: aws.amazon.com/ec2/pricing/          -->
<!-- ============================================================ -->

<style>
.eg { font-family:'IBM Plex Sans', 'Segoe UI', Helvetica, Arial, sans-serif; }
.eg-mono { font-family:'IBM Plex Mono', Consolas, 'Courier New', monospace; }
.eg-label { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; color:#2a9d8f; margin:0 0 14px 0; }
.eg-h2 { font-size:32px; font-weight:700; color:#16181d; line-height:1.25; margin:0 0 24px 0; letter-spacing:-0.6px; }
.eg-p { font-size:17px; line-height:1.85; color:#2e333b; margin:0 0 22px 0; }
.eg-p strong { color:#16181d; }
.eg-p a { color:#1f7a6f; font-weight:600; text-decoration:none; border-bottom:1px solid #2a9d8f; }
.eg-p code, .eg-faq-a code { font-family:'IBM Plex Mono', Consolas, monospace; font-size:14px; background:#eef1f4; padding:1px 6px; border-radius:3px; color:#16181d; }
.eg-note { font-size:13px; line-height:1.65; color:#8a929e; margin:12px 0 0 0; font-style:italic; }
.eg-stats { display:grid; grid-template-columns:repeat(4, 1fr); gap:16px; }
.eg-stat { background:#0d0f12; border:1px solid #2a2e36; border-radius:6px; padding:26px 22px; }
.eg-stat-n { font-family:'IBM Plex Mono', Consolas, monospace; font-size:30px; font-weight:600; color:#e9a23b; margin:0 0 10px 0; line-height:1; }
.eg-stat-l { font-size:13.5px; line-height:1.55; color:#a7afba; margin:0; }
.eg-qa { display:grid; grid-template-columns:64px 1fr; gap:26px; padding:36px 0; border-top:1px solid #dfe3e8; }
.eg-qn { font-family:'IBM Plex Mono', Consolas, monospace; font-size:15px; font-weight:600; color:#ffffff; background:#2a9d8f; border-radius:4px; padding:8px 0; text-align:center; margin:0; height:fit-content; }
.eg-q { font-size:22px; font-weight:700; color:#16181d; line-height:1.4; margin:0 0 18px 0; }
.eg-quote { border-left:3px solid #e9a23b; background:#fbf6ec; padding:16px 22px; margin:0 0 22px 0; font-size:17px; line-height:1.65; color:#16181d; font-style:italic; }
.eg-quote cite { display:block; font-family:'IBM Plex Mono', Consolas, monospace; font-style:normal; font-size:12px; color:#a8681a; margin-top:10px; }
.eg-take { background:#16181d; border-radius:6px; padding:22px 26px; margin:8px 0 0 0; }
.eg-take p { font-size:15px; line-height:1.75; color:#c9d1d9; margin:0; }
.eg-take b { font-family:'IBM Plex Mono', Consolas, monospace; color:#2a9d8f; font-size:13px; display:block; margin-bottom:8px; font-weight:600; }
.eg-tbl { width:100%; border-collapse:collapse; background:#ffffff; border:1px solid #dfe3e8; font-size:14px; margin:8px 0; }
.eg-tbl caption { text-align:left; font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; color:#2a9d8f; padding-bottom:12px; }
.eg-tbl th { background:#16181d; color:#e6e8eb; padding:13px 14px; text-align:left; font-family:'IBM Plex Mono', Consolas, monospace; font-size:11.5px; font-weight:600; }
.eg-tbl td { padding:12px 14px; border-bottom:1px solid #eceff2; color:#525a66; vertical-align:top; line-height:1.55; }
.eg-tbl tr:last-child td { border-bottom:none; }
.eg-tbl .b { font-weight:700; color:#16181d; }
.eg-tbl .num { text-align:right; font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; white-space:nowrap; }
.eg-tbl .tot td { background:#fbf6ec; font-weight:700; color:#16181d; }
.eg-tbl .good { color:#1f7a6f; font-weight:700; }
.eg-tbl .bad { color:#b03a2e; font-weight:700; }
.eg-bars { background:#ffffff; border:1px solid #dfe3e8; border-radius:6px; padding:26px 26px 14px 26px; }
.eg-bar { display:grid; grid-template-columns:160px 1fr 90px; gap:14px; align-items:center; margin:0 0 13px 0; }
.eg-bar-l { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; color:#16181d; }
.eg-bar-t { background:#eef1f4; height:24px; border-radius:3px; }
.eg-bar-f { height:24px; border-radius:3px; }
.eg-bar-v { font-family:'IBM Plex Mono', Consolas, monospace; font-size:14px; font-weight:600; color:#16181d; text-align:right; }
.eg-tl { border-left:2px solid #2a9d8f; margin:10px 0 0 8px; padding:0 0 0 28px; }
.eg-tl-item { position:relative; padding:0 0 28px 0; }
.eg-tl-item:before { content:''; position:absolute; left:-36px; top:4px; width:14px; height:14px; border-radius:50%; background:#2a9d8f; border:3px solid #ffffff; box-shadow:0 0 0 1px #2a9d8f; }
.eg-tl-date { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13px; color:#a8681a; margin:0 0 6px 0; font-weight:600; }
.eg-tl-item h4 { font-size:18px; font-weight:700; color:#16181d; margin:0 0 6px 0; }
.eg-tl-item p { font-size:15px; line-height:1.7; color:#525a66; margin:0; }
.eg-check { background:#0d0f12; border-radius:8px; overflow:hidden; border:1px solid #2a2e36; }
.eg-check-bar { background:#1f232a; padding:10px 16px; font-family:'IBM Plex Mono', Consolas, monospace; font-size:12px; color:#8a929e; }
.eg-check ul { list-style:none; margin:0; padding:20px 24px; }
.eg-check li { font-family:'IBM Plex Mono', Consolas, monospace; font-size:13.5px; line-height:1.7; color:#c9d1d9; padding:6px 0 6px 30px; position:relative; }
.eg-check li:before { content:'[ ]'; position:absolute; left:0; color:#2a9d8f; }
.eg-faq { display:grid; grid-template-columns:1fr 1fr; gap:0 50px; }
.eg-faq-item { padding:22px 0; border-top:1px solid #dfe3e8; }
.eg-faq-q { font-family:'IBM Plex Mono', Consolas, monospace; font-size:15px; font-weight:600; color:#16181d; margin:0 0 10px 0; line-height:1.5; }
.eg-faq-a { font-size:15px; line-height:1.75; color:#525a66; margin:0; }
/* Override theme heading styles */
.eg h1, .eg h2, .eg h3, .eg h4, .eg p { text-transform:none !important; }
.eg .eg-h2, .eg .eg-tl-item h4 { color:#16181d !important; }
.stk-eg-hero h1 { color:#f2f3f5 !important; }
.stk-eg-hero h1 span { color:#e9a23b !important; }
.stk-eg-pq h2 { color:#ffffff !important; }
@media screen and (max-width:900px) { .eg-stats { grid-template-columns:1fr 1fr; } }
@media screen and (max-width:768px) {
  .eg-qa { grid-template-columns:1fr; gap:10px; }
  .eg-qn { width:56px; }
  .eg-faq { grid-template-columns:1fr; }
  .eg-bar { grid-template-columns:1fr; gap:6px; }
  .eg-bar-v { text-align:left; }
  .eg-tbl th, .eg-tbl td { padding:8px 8px; font-size:12px; }
  .eg-h1 { font-size:36px !important; }
}
@media screen and (max-width:520px) { .eg-stats { grid-template-columns:1fr; } }
</style>


<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-hero stk-block-background" data-block-id="eg-hero"><style>.stk-eg-hero {background-color:#16181d !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:920px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p class="eg-label">Cloud &amp; Infrastructure &middot; Running Costs &middot; ~18 min read</p>
<p style="font-size:19px; line-height:1.75; color:#a7afba; margin:0 0 30px 0;">We priced what it costs to move 10 TB out of the major clouds, added the charges that don&rsquo;t appear on the pricing page, and checked what the EU Data Act really changes in January 2027. Short version: the law fixes one scenario. Your bill lives in the other three.</p>
<p class="eg-mono" style="font-size:12.5px; color:#6b7480; margin:0;">BY <span style="color:#f2f3f5;">OWEN CALDER</span> &middot; NOVEMBER 2026 &middot; PRICES CHECKED AGAINST PUBLISHED RATES AS OF SEPTEMBER 2026</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-intro stk-block-background" data-block-id="eg-intro"><style>.stk-eg-intro {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p class="eg-p">Every cloud migration plan I&rsquo;ve reviewed has a slide that says &ldquo;portable by design&rdquo;. The containers are standard, the database is open source, the infrastructure is in Terraform. Then someone runs the transfer estimate for the data itself, and the project quietly goes back for another budget round.</p>
<p class="eg-p">That&rsquo;s egress: what a provider charges for data leaving its network. Uploading is free everywhere. Downloading, serving, replicating and leaving are not. Nobody hides this &mdash; it&rsquo;s on every pricing page &mdash; but very few teams model it until they&rsquo;re already trying to move.</p>
<p class="eg-p">This post pulls together three things: published egress rates as compiled by the independent pricing index EgressCost (verified September 2026), a sharp recent analysis of exit costs by infrastructure architect Kees Baggerman on his blog <em>My Virtual Vision</em>, and our own arithmetic on top. I&rsquo;ve split it into the questions I&rsquo;d want answered before signing off any exit plan.</p>
<div style="background:#f5f7f8; border:1px solid #dfe3e8; border-left:3px solid #2a9d8f; border-radius:4px; padding:18px 22px; margin:6px 0 0 0;">
<p class="eg-mono" style="font-size:12px; color:#2a9d8f; margin:0 0 6px 0;">// how to read this</p>
<p style="font-size:14.5px; line-height:1.7; color:#525a66; margin:0;">Rates are US-region internet egress unless stated, and change often &mdash; check the provider&rsquo;s own page before you budget. Where a number is our calculation rather than a published figure, the table says so. Quotes are from published material and are attributed; nobody quoted has reviewed this post.</p>
</div>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-stats stk-block-background" data-block-id="eg-stats"><style>.stk-eg-stats {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1100px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p class="eg-label">$ cat summary.txt</p>
<div class="eg-stats">
  <div class="eg-stat"><p class="eg-stat-n">$0</p><p class="eg-stat-l">Cost to send 10 TB to the internet from Cloudflare R2, which charges no egress at all</p></div>
  <div class="eg-stat"><p class="eg-stat-n">$1,137</p><p class="eg-stat-l">The same 10 TB from Google Cloud on its Premium network tier</p></div>
  <div class="eg-stat"><p class="eg-stat-n">+73%</p><p class="eg-stat-l">How much NAT and cross-zone charges can add to a 10 TB AWS bill (our calculation)</p></div>
  <div class="eg-stat"><p class="eg-stat-n">12.01.27</p><p class="eg-stat-l">The date EU law bans switching charges &mdash; for one specific kind of move</p></div>
</div>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p1 stk-block-background" data-block-id="eg-p1"><style>.stk-eg-p1 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<p class="eg-label">// part one &middot; the price list</p><h2 class="eg-h2">What does leaving actually cost?</h2><div class="eg-qa"><p class="eg-qn">01</p><div><p class="eg-q">Why is egress priced so differently from ingress?</p><p class="eg-p">Because the asymmetry is the point. Free ingress removes every reason not to bring your data in. Metered egress means that the bigger your dataset grows, the more it costs to take it anywhere else. Providers will tell you it funds their global network &mdash; which is true &mdash; but it also works as a very effective retention mechanism.</p><p class="eg-p">The rates are published openly. AWS, for example, lists its data transfer tiers on the <a href="https://aws.amazon.com/ec2/pricing/" target="_blank" rel="noopener">Amazon EC2 pricing page</a>: the first 100 GB a month is free, the next 10 TB costs $0.09 per GB, and the rate steps down as volume grows.</p><div class="eg-take"><p><b>// owen&rsquo;s take</b>The tier boundaries are binary on AWS: &ldquo;first 10 TB&rdquo; means 10,240 GB, not 10,000. It&rsquo;s a small thing, but it&rsquo;s exactly the kind of detail that makes two estimates of the same move disagree by a few percent.</p></div></div></div><div class="eg-qa"><p class="eg-qn">02</p><div><p class="eg-q">So what does 10 TB cost, provider by provider?</p><p class="eg-p">Here&rsquo;s the entry-tier rate for internet egress, and what a single 10 TB month costs on each. The spread between cheapest and most expensive is the entire bill.</p></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-t1 stk-block-background" data-block-id="eg-t1"><style>.stk-eg-t1 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<table class="eg-tbl">
<caption>Table 1 &mdash; 10 TB of internet egress in one month</caption>
<thead><tr><th>Provider</th><th>Entry rate / GB</th><th>Free allowance</th><th class="num">10 TB / month</th></tr></thead>
<tbody>
<tr><td class="b">Cloudflare R2</td><td>$0.00</td><td>All volumes</td><td class="num good">$0</td></tr>
<tr><td class="b">Oracle Cloud</td><td>$0.0085 after allowance</td><td>10 TB / month</td><td class="num good">$0</td></tr>
<tr><td class="b">Backblaze B2</td><td>$0.01 (free to Cloudflare CDN)</td><td>Partner network</td><td class="num">~$100</td></tr>
<tr><td class="b">Microsoft Azure</td><td>$0.087</td><td>100 GB</td><td class="num">~$861</td></tr>
<tr><td class="b">Amazon Web Services</td><td>$0.09</td><td>100 GB</td><td class="num">~$913</td></tr>
<tr><td class="b">Google Cloud (Premium Tier)</td><td>$0.12 first TB, then $0.11</td><td>Minimal</td><td class="num bad">~$1,137</td></tr>
</tbody>
</table>
<p class="eg-note">Hyperscaler and R2 figures: published rates as compiled by the EgressCost index, US regions, verified September 2026. Oracle and Backblaze figures are our calculation from the same published entry rates. Google&rsquo;s Standard network tier is cheaper than Premium; price it in Google&rsquo;s own calculator for your route.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p1b stk-block-background" data-block-id="eg-p1b"><style>.stk-eg-p1b {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<div class="eg-qa"><p class="eg-qn">03</p><div><p class="eg-q">Does the gap shrink as you move more data?</p><p class="eg-p">No &mdash; it grows. Per-GB rates fall with volume, but they never approach zero, so the absolute difference between providers keeps widening. Baggerman makes the same point about the curve: at around 1 PB, an exit costs roughly 61 times a 10 TB exit &mdash; cheaper per gigabyte, but nowhere near free.</p><div class="eg-quote">&ldquo;A 100 TB exit is not 10&times; the cost of a 10 TB exit. A 1 PB exit works out at about 61&times;, still expensive enough to derail a project plan that assumed the EU Data Act made it free.&rdquo;<cite>&mdash; Kees Baggerman, My Virtual Vision</cite></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-t2 stk-block-background" data-block-id="eg-t2"><style>.stk-eg-t2 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<table class="eg-tbl">
<caption>Table 2 &mdash; Monthly internet egress cost by volume</caption>
<thead><tr><th>Monthly volume</th><th>Typical workload</th><th class="num">AWS</th><th class="num">Azure</th><th class="num">Google (Premium)</th><th class="num">R2</th></tr></thead>
<tbody>
<tr><td class="b">1 TB</td><td>Small SaaS</td><td class="num">$83</td><td class="num">$78</td><td class="num">$123</td><td class="num">$0</td></tr>
<tr><td class="b">10 TB</td><td>Mid-sized SaaS</td><td class="num">$913</td><td class="num">$861</td><td class="num">$1,137</td><td class="num">$0</td></tr>
<tr><td class="b">50 TB</td><td>Larger platform</td><td class="num">$4,395</td><td class="num">$4,182</td><td class="num">$4,413</td><td class="num">$0</td></tr>
<tr><td class="b">100 TB</td><td>Media / heavy API</td><td class="num">$7,980</td><td class="num">$7,683</td><td class="num">$8,509</td><td class="num">$0</td></tr>
</tbody>
</table>
<p class="eg-note">Source: EgressCost index, US regions, after free allowances, verified September 2026. We re-ran the AWS 100 TB line from the published tiers ($0.09 &rarr; $0.085 &rarr; $0.07) and got $7,978 &mdash; consistent.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p1c stk-block-background" data-block-id="eg-p1c"><style>.stk-eg-p1c {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<div class="eg-qa"><p class="eg-qn">04</p><div><p class="eg-q">Is that a lot, compared with what I already pay to store the data?</p><p class="eg-p">This is the comparison I find most useful, because it turns an abstract per-GB number into something a finance team instantly understands. If you keep data in standard object storage on AWS, taking it out once costs roughly <strong>three and a half to four months of storage rent</strong>.</p></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-t3 stk-block-background" data-block-id="eg-t3"><style>.stk-eg-t3 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<table class="eg-tbl">
<caption>Table 3 &mdash; The &ldquo;exit tax&rdquo; in months of storage (AWS, S3 Standard, us-east-1) &mdash; our calculation</caption>
<thead><tr><th>Dataset</th><th class="num">Storage / month</th><th class="num">Storage / year</th><th class="num">One-time egress to leave</th><th class="num">Exit = months of storage</th></tr></thead>
<tbody>
<tr><td class="b">10 TB</td><td class="num">~$236</td><td class="num">~$2,830</td><td class="num">~$913</td><td class="num">~3.9</td></tr>
<tr><td class="b">100 TB</td><td class="num">~$2,304</td><td class="num">~$27,650</td><td class="num">~$7,980</td><td class="num">~3.5</td></tr>
</tbody>
</table>
<p class="eg-note">Our calculation using S3 Standard list storage of $0.023/GB-month for the first 50 TB and $0.022 above, and the egress figures in Table 2. Excludes request charges, retrieval fees for colder storage classes, and the hidden meters below.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p2 stk-block-background" data-block-id="eg-p2"><style>.stk-eg-p2 {background-color:#f5f7f8 !important; border-top:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<p class="eg-label">// part two &middot; the meters you didn&rsquo;t see</p><h2 class="eg-h2">The pricing page shows one meter. Your bill has several.</h2><div class="eg-qa"><p class="eg-qn">05</p><div><p class="eg-q">What else gets charged when data moves?</p><p class="eg-p">Internet egress is the line everybody prices. The others sit inside the network path and are easy to miss in an estimate, because they don&rsquo;t look like &ldquo;egress&rdquo; on the invoice. On AWS, the main ones are a NAT Gateway processing fee for traffic leaving private subnets, a charge in each direction for traffic crossing availability zones, and an hourly fee for every public IPv4 address.</p></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-t4 stk-block-background" data-block-id="eg-t4"><style>.stk-eg-t4 {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<table class="eg-tbl">
<caption>Table 4 &mdash; The hidden meters, and what they add to a 10 TB month on AWS</caption>
<thead><tr><th>Charge</th><th>Published rate</th><th>When it applies</th><th class="num">On 10 TB</th></tr></thead>
<tbody>
<tr><td class="b">Internet egress (base)</td><td>$0.09/GB (first 10 TB)</td><td>Any data leaving AWS to the internet</td><td class="num">$913</td></tr>
<tr><td class="b">NAT Gateway processing</td><td>$0.045/GB</td><td>Traffic from private subnets routed through NAT</td><td class="num">+$461</td></tr>
<tr><td class="b">Cross-AZ transfer</td><td>$0.01/GB each way</td><td>Data crossing zones on its way out</td><td class="num">+$205</td></tr>
<tr><td class="b">Public IPv4 addresses</td><td>$0.005/hour each (~$43.80/yr)</td><td>Every public IPv4 you hold</td><td class="num">per IP</td></tr>
<tr class="tot"><td>Worst-case 10 TB month</td><td></td><td>NAT + cross-AZ on every byte</td><td class="num">~$1,579</td></tr>
</tbody>
</table>
<p class="eg-note">Rates as published by AWS and compiled by the EgressCost index. Totals are our calculation for a worst case where all 10 TB passes through a NAT Gateway and crosses one zone boundary each way. That&rsquo;s +73% on the base egress line &mdash; and it&rsquo;s more common than it sounds in default VPC designs.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p2b stk-block-background" data-block-id="eg-p2b"><style>.stk-eg-p2b {background-color:#f5f7f8 !important; border-bottom:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<div class="eg-qa"><p class="eg-qn">06</p><div><p class="eg-q">Are these prices going down?</p><p class="eg-p">Mostly they&rsquo;re flat, and in one notable case they went up. On 1 May 2026, Google raised its North American peering egress rates &mdash; on CDN Interconnect, Direct Peering and Carrier Peering &mdash; from $0.04 to $0.08 per GiB. Standard internet egress didn&rsquo;t change, so most workloads pay what they did before. But if you&rsquo;d built a cost model around cheap peering, it just doubled.</p><p class="eg-p">The downward pressure comes from the zero-egress providers. Cloudflare R2, Backblaze B2 and Wasabi have spent the past two years making &ldquo;store it somewhere that doesn&rsquo;t charge you to read it&rdquo; a credible default for media and backups.</p><div class="eg-take"><p><b>// owen&rsquo;s take</b>Run <code style="background:#2a2e36; color:#e9a23b;">aws ec2 describe-addresses</code> and count. I&rsquo;ve yet to audit an account that didn&rsquo;t have a few Elastic IPs attached to nothing, each quietly billing $43.80 a year. It&rsquo;s small money, but it&rsquo;s a perfect proxy for how closely anyone is watching the network bill.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p3 stk-block-background" data-block-id="eg-p3"><style>.stk-eg-p3 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<p class="eg-label">// part three &middot; the law</p><h2 class="eg-h2">Doesn&rsquo;t the EU Data Act make leaving free?</h2><div class="eg-qa"><p class="eg-qn">07</p><div><p class="eg-q">What exactly changes, and when?</p><p class="eg-p">Less than the headlines suggest, and with a specific date attached. Baggerman&rsquo;s reading of Regulation (EU) 2023/2854 is the clearest I&rsquo;ve seen, and he&rsquo;s careful to note he&rsquo;s an architect, not a lawyer &mdash; as am I. Check anything here with your own legal team.</p></div></div><div class="eg-tl">
<div class="eg-tl-item"><p class="eg-tl-date">11 January 2024</p><h4>Reduced switching charges</h4><p>Providers may only charge switching fees capped at the costs directly linked to the switch. Egress counts as a switching charge under Article 2(36).</p></div>
<div class="eg-tl-item"><p class="eg-tl-date">Since then</p><h4>Switching credits, on application</h4><p>AWS, Azure and Google all offer free egress for customers leaving &mdash; but as a process you apply for, document and complete in a fixed window (60 days on Azure; 180 days on AWS for customers outside the EU), not as a default price.</p></div>
<div class="eg-tl-item"><p class="eg-tl-date">September 2025</p><h4>Google Data Transfer Essentials</h4><p>Qualifying multicloud traffic for EU and UK customers is billed at zero once you opt in and declare it.</p></div>
<div class="eg-tl-item" style="padding-bottom:0;"><p class="eg-tl-date">12 January 2027</p><h4>Switching charges banned</h4><p>Providers may no longer charge for switching at all. A one-way, permanent move off a provider becomes free by law.</p></div>
</div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p3b stk-block-background" data-block-id="eg-p3b"><style>.stk-eg-p3b {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<div class="eg-qa"><p class="eg-qn">08</p><div><p class="eg-q">So what&rsquo;s still on the meter after January 2027?</p><p class="eg-p">Nearly everything that isn&rsquo;t a clean, one-way exit. The ban covers switching. It doesn&rsquo;t cover the traffic that most organisations actually generate.</p><div class="eg-quote">&ldquo;All of that is metered after January 2027 exactly as it is today.&rdquo;<cite>&mdash; Kees Baggerman, on routine egress, multicloud, BCDR replication, in-parallel use and internal transfer</cite></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-t5 stk-block-background" data-block-id="eg-t5"><style>.stk-eg-t5 {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<table class="eg-tbl">
<caption>Table 5 &mdash; Which moves the Data Act actually makes free</caption>
<thead><tr><th>Scenario</th><th>Free after 12 Jan 2027?</th><th>Why</th></tr></thead>
<tbody>
<tr><td class="b">Full one-way exit</td><td class="good">Yes</td><td>This is the &ldquo;switching&rdquo; the law covers</td></tr>
<tr><td class="b">Exit test / round trip</td><td class="bad">No</td><td>You extract, test and come back &mdash; that isn&rsquo;t switching</td></tr>
<tr><td class="b">Running two clouds in parallel</td><td class="bad">Not by default</td><td>Article 34(2) lets providers pass on their costs for in-parallel use</td></tr>
<tr><td class="b">Partial rebalance to another cloud</td><td class="bad">No</td><td>Standard egress, often at the most expensive tier because volumes are small</td></tr>
<tr><td class="b">Backup / DR replication</td><td class="bad">No</td><td>Routine traffic, metered as normal</td></tr>
<tr><td class="b">Cross-zone, NAT, inter-region</td><td class="bad">No</td><td>Internal transfer, outside the switching definition</td></tr>
<tr><td class="b">Serving your users</td><td class="bad">No</td><td>Ordinary internet egress</td></tr>
</tbody>
</table>
<p class="eg-note">Summary of Baggerman&rsquo;s reading of Articles 2(36), 29 and 34(2). Not legal advice.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p3c stk-block-background" data-block-id="eg-p3c"><style>.stk-eg-p3c {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<div class="eg-qa"><p class="eg-qn">09</p><div><p class="eg-q">Who should worry most?</p><p class="eg-p">Regulated financial firms in the EU. Under DORA, exit plans for critical outsourced services have to be tested, and a real exit test is a round trip &mdash; exactly the scenario the Data Act doesn&rsquo;t cover. Baggerman models a 5,000-seat virtual desktop estate and finds the data that actually has to move &mdash; user profiles, golden images, session recordings, snapshots &mdash; usually lands at 100 to 250 TB, not the 50 TB the project plan assumed. That makes each test a recurring budget line rather than a one-off.</p><div class="eg-take"><p><b>// owen&rsquo;s take</b>You don&rsquo;t need to be a bank for this to matter. Swap &ldquo;DORA exit test&rdquo; for &ldquo;we want to try a second provider for a quarter&rdquo; and the economics are identical. The partial rebalance is the most common real-world move and the one with the worst per-GB rate.</p></div></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-pq stk-block-background" data-block-id="eg-pq"><style>.stk-eg-pq {background-color:#2a9d8f !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p class="eg-mono" style="font-size:13px; color:#d5f0ec; margin:0 0 16px 0;">$ git commit -m</p>
<h2 style="font-size:30px; font-weight:600; color:#ffffff; line-height:1.4; margin:0; letter-spacing:-0.4px;">&ldquo;An exit plan that does not name the egress tariff, the volume estimate, and the mitigation path is documentation, not strategy.&rdquo;</h2>
<p class="eg-mono" style="font-size:12px; color:#d5f0ec; margin:18px 0 0 0;">&mdash; Kees Baggerman, My Virtual Vision</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p4 stk-block-background" data-block-id="eg-p4"><style>.stk-eg-p4 {background-color:#f5f7f8 !important; border-top:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<p class="eg-label">// part four &middot; what actually helps</p><h2 class="eg-h2">Making the meter run slower.</h2><div class="eg-qa"><p class="eg-qn">10</p><div><p class="eg-q">Is a dedicated connection worth it?</p><p class="eg-p">At volume, yes. AWS Direct Connect, Azure ExpressRoute and Google Cloud Interconnect carry much lower per-GB rates than internet egress. Baggerman&rsquo;s example: moving 250 TB over Direct Connect at $0.02/GB cuts the transfer line from $16,691 to $5,120 (using EU Ireland internet tiers). The catch is the port: a dedicated 10 Gbps connection runs about $2.25 an hour, roughly $1,640 a month, before cross-connect and carrier circuit costs. Since September 2026, AWS also offers flat-rate pricing on 10G and 100G dedicated connections.</p><p class="eg-p">So where&rsquo;s the break-even? We ran it against a monthly 10 G port.</p></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-bars stk-block-background" data-block-id="eg-bars"><style>.stk-eg-bars {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<div class="eg-bars">
<p class="eg-mono" style="font-size:12px; color:#2a9d8f; margin:0 0 18px 0;">monthly cost: internet egress vs Direct Connect (10 G port + $0.02/GB)</p>
<div class="eg-bar"><span class="eg-bar-l">10 TB internet</span><div class="eg-bar-t"><div class="eg-bar-f" style="width:26%; background:#525a66;"></div></div><span class="eg-bar-v">$913</span></div>
<div class="eg-bar"><span class="eg-bar-l">10 TB via DX</span><div class="eg-bar-t"><div class="eg-bar-f" style="width:53%; background:#b03a2e;"></div></div><span class="eg-bar-v">$1,845</span></div>
<div class="eg-bar"><span class="eg-bar-l">25 TB internet</span><div class="eg-bar-t"><div class="eg-bar-f" style="width:63%; background:#525a66;"></div></div><span class="eg-bar-v">$2,219</span></div>
<div class="eg-bar"><span class="eg-bar-l">25 TB via DX</span><div class="eg-bar-t"><div class="eg-bar-f" style="width:62%; background:#e9a23b;"></div></div><span class="eg-bar-v">$2,152</span></div>
<div class="eg-bar"><span class="eg-bar-l">40 TB internet</span><div class="eg-bar-t"><div class="eg-bar-f" style="width:100%; background:#525a66;"></div></div><span class="eg-bar-v">$3,524</span></div>
<div class="eg-bar"><span class="eg-bar-l">40 TB via DX</span><div class="eg-bar-t"><div class="eg-bar-f" style="width:72%; background:#2a9d8f;"></div></div><span class="eg-bar-v">$2,459</span></div>
</div>
<p class="eg-note">Our calculation: AWS internet tiers ($0.09 to 10 TB, $0.085 above), 1 TB = 1,024 GB, 100 GB free; Direct Connect at $0.02/GB plus a ~$1,640/month port. Ignores cross-connect, carrier circuit and any flat-rate option. Break-even lands around 24 TB a month &mdash; below that, the port costs more than it saves.</p>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p4b stk-block-background" data-block-id="eg-p4b"><style>.stk-eg-p4b {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<div class="eg-qa"><p class="eg-qn">11</p><div><p class="eg-q">What else works, roughly in order of effort?</p><p class="eg-p">Most of the savings come from boring decisions made early. Here&rsquo;s how I&rsquo;d rank the options for a typical small-to-mid team.</p></div></div>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-t6 stk-block-background" data-block-id="eg-t6"><style>.stk-eg-t6 {background-color:#f5f7f8 !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1000px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<table class="eg-tbl">
<caption>Table 6 &mdash; Egress reduction options, ranked</caption>
<thead><tr><th>Tactic</th><th>Effort</th><th>Best for</th><th>What it cuts</th></tr></thead>
<tbody>
<tr><td class="b">Delete what nobody needs</td><td>Low</td><td>Everyone</td><td>Stale snapshots, old logs and recordings &mdash; cuts storage now and egress later</td></tr>
<tr><td class="b">Compress before transfer</td><td>Low</td><td>APIs, logs, text-heavy data</td><td>Bytes on the wire; text often shrinks dramatically</td></tr>
<tr><td class="b">Put a CDN in front</td><td>Low&ndash;medium</td><td>Static assets, media, downloads</td><td>Origin egress, served from cache instead</td></tr>
<tr><td class="b">Gateway endpoints for object storage</td><td>Low</td><td>Private workloads talking to S3</td><td>NAT processing on that traffic</td></tr>
<tr><td class="b">Move to IPv6 where possible</td><td>Medium</td><td>New services</td><td>IPv4 charges and, in some designs, the need for NAT</td></tr>
<tr><td class="b">Zero-egress storage for media</td><td>Medium</td><td>Video, images, public datasets</td><td>Internet egress on read-heavy data</td></tr>
<tr><td class="b">Dedicated interconnect</td><td>High</td><td>Steady 25 TB+ a month, exit tests</td><td>Per-GB rate, at the cost of a port</td></tr>
<tr><td class="b">Separate &ldquo;must leave&rdquo; data</td><td>High (early)</td><td>Anyone with an exit plan</td><td>The volume you&rsquo;d actually have to move</td></tr>
</tbody>
</table>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-p4c stk-block-background" data-block-id="eg-p4c"><style>.stk-eg-p4c {background-color:#f5f7f8 !important; border-bottom:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">
<p class="eg-p" style="margin-top:10px;">That last row is the one that pays off most and gets done least. If the data you&rsquo;d need to take with you lives in the same buckets as telemetry, analytics and five years of logs, your exit project is much bigger than it needs to be. Keeping them separate costs nothing on day one.</p>
</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-check stk-block-background" data-block-id="eg-check"><style>.stk-eg-check {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p class="eg-label">$ cat exit-readiness.md</p>
<h2 class="eg-h2">The exit-readiness checklist</h2>
<p class="eg-p">If you can tick all of these, your exit plan is a plan. If you can&rsquo;t, it&rsquo;s a hope with a Terraform repo.</p>
<div class="eg-check">
<div class="eg-check-bar">exit-readiness.md</div>
<ul>
<li>A volume estimate of the data that physically has to move, split by type</li>
<li>The egress tariff that applies, with the source page and the date you checked it</li>
<li>Hidden meters included: NAT, cross-zone, inter-region, IPv4</li>
<li>The scenario named: one-way exit, round-trip test, parallel run or partial move</li>
<li>The mitigation chosen: switching credit, interconnect, or paid egress &mdash; with a cost</li>
<li>Lifecycle rules so you aren&rsquo;t paying to move data nobody wants</li>
<li>&ldquo;Must leave&rdquo; data stored separately from everything else</li>
<li>At least one test run, with what it actually cost written down</li>
</ul>
</div>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-faq stk-block-background" data-block-id="eg-faq"><style>.stk-eg-faq {background-color:#f5f7f8 !important; border-top:1px solid #dfe3e8;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:1050px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p class="eg-label">$ man egress</p>
<h2 class="eg-h2" style="margin-bottom:24px;">Questions people ask</h2>
<div class="eg-faq">
<div>
<div class="eg-faq-item"><p class="eg-faq-q">What is cloud egress?</p><p class="eg-faq-a">The charge for data leaving a provider&rsquo;s network &mdash; to the internet, to another cloud or to your own data centre. Data coming in (ingress) is free on every major provider.</p></div>
<div class="eg-faq-item"><p class="eg-faq-q">Which cloud has the cheapest egress?</p><p class="eg-faq-a">Cloudflare R2 charges nothing for egress, and Oracle includes 10 TB a month free. Among the three largest providers, Azure is slightly cheaper than AWS at entry level, and Google&rsquo;s Premium Tier is the most expensive.</p></div>
<div class="eg-faq-item"><p class="eg-faq-q">Does the EU Data Act make leaving my cloud free?</p><p class="eg-faq-a">From 12 January 2027, a full one-way switch can&rsquo;t be charged for. Round-trip tests, parallel running, partial moves, backups and internal transfer are still billed as normal.</p></div>
</div>
<div>
<div class="eg-faq-item"><p class="eg-faq-q">What is a NAT Gateway charge?</p><p class="eg-faq-a">On AWS, traffic from private subnets that goes out through a NAT Gateway is charged a processing fee of $0.045/GB on top of normal egress. Gateway endpoints for S3 avoid it for storage traffic.</p></div>
<div class="eg-faq-item"><p class="eg-faq-q">When is Direct Connect worth it?</p><p class="eg-faq-a">On our numbers, once you steadily move more than roughly 24 TB a month, or for large one-off exit tests. Below that, the monthly port fee costs more than the per-GB saving.</p></div>
<div class="eg-faq-item"><p class="eg-faq-q">How do I find egress on my bill?</p><p class="eg-faq-a">On AWS, filter Cost Explorer by usage types containing <code>DataTransfer-Out</code>, plus NAT Gateway and inter-AZ usage types. Azure and Google show it under bandwidth and network egress respectively.</p></div>
</div>
</div>

</div></div></div>
</div></div>



<div class="wp-block-stackable-columns alignfull stk-block-columns stk-block stk-eg-end stk-block-background" data-block-id="eg-end"><style>.stk-eg-end {background-color:#ffffff !important;}</style><div class="stk-row stk-inner-blocks stk-block-content stk-content-align">
<div class="wp-block-stackable-column stk-block-column stk-column stk-block"><div class="stk-column-wrapper stk-block-column__content stk-container stk--no-background stk--no-padding" style="max-width:900px; margin:auto;"><div class="stk-block-content stk-inner-blocks eg">

<p style="font-size:13px; line-height:1.75; color:#8a929e; margin:0 0 12px 0; font-style:italic;">Sources: published provider rates as compiled by the independent EgressCost index (verified September 2026); Kees Baggerman, &ldquo;Egress Is the Second Lock-In: Modeling the Real Cost of Leaving Your Cloud Provider&rdquo;, My Virtual Vision, November 2026; AWS published EC2 and S3 pricing. Calculations marked as ours are illustrative and use list prices only. Nothing here is legal advice on the EU Data Act or DORA.</p>
<p class="eg-mono" style="font-size:12px; color:#8a929e; margin:0;">Spotted a price that changed? Tell us &mdash; we update posts rather than delete them.</p>

</div></div></div>
</div></div>


<!-- END --><p>The post <a rel="nofollow" href="https://thepuffingroup.com/cloud-egress-costs-2026-what-it-really-costs-to-leave-aws-azure-gcp/">Cloud Egress Costs 2026: What It Really Costs to Leave AWS, Azure &#038; GCP</a> appeared first on <a rel="nofollow" href="https://thepuffingroup.com">Puffin Notes</a>.</p>
<p>The post <a href="https://thepuffingroup.com/cloud-egress-costs-2026-what-it-really-costs-to-leave-aws-azure-gcp/">Cloud Egress Costs 2026: What It Really Costs to Leave AWS, Azure &#038; GCP</a> appeared first on <a href="https://thepuffingroup.com">Puffin Notes</a>.</p>
]]></content:encoded>
					
		
		
			</item>
	</channel>
</rss>
