-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
498 lines (493 loc) · 29.4 KB
/
Copy pathindex.html
File metadata and controls
498 lines (493 loc) · 29.4 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Tokenmaxxing: Is AI a Bubble? Why It's Both a Revolution and a Bubble</title>
<meta name="description" content="A company accidentally spent $500M on AI in one month. Inside 'tokenmaxxing,' OpenAI's billions in losses, the circular hyperscaler money, and why AI is both the most transformative technology of our lifetime and a historic bubble.">
<meta name="author" content="Jason Colapietro">
<meta name="robots" content="index,follow,max-image-preview:large,max-snippet:-1,max-video-preview:-1">
<link rel="canonical" href="https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in">
<link rel="author" href="https://suedeai.ai/founder">
<link rel="me" href="https://x.com/johnnysuede">
<link rel="me" href="https://github.com/JasonColapietro">
<link rel="me" href="https://www.linkedin.com/in/jasoncolapietro">
<link rel="me" href="https://jasoncolapietro.substack.com">
<meta property="og:type" content="article">
<meta property="og:locale" content="en_US">
<meta property="article:modified_time" content="2026-06-11">
<meta property="article:tag" content="AI bubble">
<meta property="article:tag" content="tokenmaxxing">
<meta property="article:tag" content="OpenAI">
<meta property="article:tag" content="artificial intelligence">
<meta property="article:tag" content="Anthropic">
<meta property="article:tag" content="AI spending">
<meta property="article:section" content="Technology">
<meta property="og:title" content="A Company Spent $500,000,000 on AI in One Month — By Accident. They Call It "Tokenmaxxing."">
<meta property="og:description" content="AI is the biggest breakthrough of our lives. It's also a historic bubble. A half-billion-dollar accident shows why both are true at once.">
<meta property="og:image" content="https://jasoncolapietro.github.io/tokenmaxxing/assets/tokenmaxxing-card.png">
<meta property="og:url" content="https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in">
<meta property="og:site_name" content="Jason Colapietro">
<meta property="article:published_time" content="2026-05-31">
<meta property="article:author" content="Jason Colapietro">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:site" content="@AISUEDE">
<meta name="twitter:creator" content="@johnnysuede">
<meta name="twitter:title" content="A Company Spent $500,000,000 on AI in One Month — By Accident">
<meta name="twitter:description" content="AI is the biggest breakthrough of our lives and a historic bubble. A half-billion-dollar accident shows why both are true.">
<meta name="twitter:image" content="https://jasoncolapietro.github.io/tokenmaxxing/assets/tokenmaxxing-card.png">
<meta name="twitter:image:alt" content="Total AI charges, one month: $500,000,000 — spent by accident. A Suede Labs AI field note on tokenmaxxing.">
<meta name="keywords" content="tokenmaxxing, AI bubble, is AI a bubble, OpenAI losses, Anthropic, AI spending, artificial intelligence, hyperscaler capex, AI revolution, circular money, AI investment, enterprise AI cost">
<meta name="theme-color" content="#0A0D12">
<link rel="sitemap" type="application/xml" href="https://jasoncolapietro.github.io/tokenmaxxing/sitemap.xml">
<link rel="alternate" type="application/rss+xml" title="Jason Colapietro on Substack" href="https://jasoncolapietro.substack.com/feed">
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Newsreader:ital,opsz,wght@0,6..72,400;0,6..72,500;0,6..72,600;1,6..72,400;1,6..72,500&family=IBM+Plex+Mono:wght@400;500;600&display=swap" rel="stylesheet">
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "Organization",
"@id": "https://suedeai.ai/#organization",
"name": "Suede Labs AI",
"alternateName": [
"Suede",
"Suede AI",
"Suede Labs"
],
"url": "https://suedeai.ai",
"logo": "https://suedeai.ai/suede-ai-logo-transparent.png",
"foundingDate": "2024",
"founder": {
"@id": "https://suedeai.ai/founder#person"
},
"sameAs": [
"https://suedeai.org/",
"https://x.com/AISUEDE",
"https://github.com/Suede-AI",
"https://www.youtube.com/@aisuede"
]
},
{
"@type": "Person",
"@id": "https://suedeai.ai/founder#person",
"name": "Jason Colapietro",
"alternateName": "Johnny Suede",
"jobTitle": "Founder and CEO of Suede Labs AI",
"url": "https://suedeai.ai/founder",
"knowsAbout": [
"programmable IP",
"music rights",
"provenance",
"ERC-8004",
"x402",
"royalty settlement",
"artificial intelligence"
],
"sameAs": [
"https://x.com/johnnysuede",
"https://github.com/JasonColapietro",
"https://www.linkedin.com/in/jasoncolapietro",
"https://www.youtube.com/@aisuede",
"https://suedeai.org/jason-colapietro/",
"https://jasoncolapietro.substack.com"
],
"worksFor": {
"@id": "https://suedeai.ai/#organization"
}
},
{
"@type": "BlogPosting",
"@id": "https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in",
"headline": "A Company Spent $500,000,000 on AI in One Month — By Accident. They Call It \"Tokenmaxxing.\"",
"description": "AI is the biggest breakthrough of our lives. It's also a historic bubble. A half-billion-dollar accident shows why both are true at once.",
"image": [
{
"@type": "ImageObject",
"url": "https://jasoncolapietro.github.io/tokenmaxxing/assets/tokenmaxxing-card.png",
"width": 1200,
"height": 630
}
],
"datePublished": "2026-05-31",
"dateModified": "2026-06-11",
"inLanguage": "en",
"url": "https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in",
"mainEntityOfPage": {
"@type": "WebPage",
"@id": "https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in"
},
"keywords": ["AI bubble", "is AI a bubble", "tokenmaxxing", "OpenAI losses", "Anthropic", "AI revolution", "artificial intelligence", "hyperscaler capex", "enterprise AI cost", "AI spending", "circular money"],
"wordCount": 2600,
"articleSection": "Technology",
"author": {
"@id": "https://suedeai.ai/founder#person"
},
"publisher": {
"@id": "https://suedeai.ai/#organization"
}
}
]
}
</script>
<style>
:root{
--ink:#0A0D12; --paper:#ECE9E0; --paper-dim:#9AA3AE; --muted:#6C7884;
--cyan:#2BE6D3; --red:#FF3A52; --line:rgba(255,255,255,.10); --line-soft:rgba(255,255,255,.06);
}
*{box-sizing:border-box}
html{-webkit-text-size-adjust:100%}
body{
margin:0; background:var(--ink); color:var(--paper);
font-family:"Newsreader",Georgia,serif; font-size:20px; line-height:1.75;
background-image: radial-gradient(130% 80% at 50% -12%, rgba(43,230,211,.07), transparent 60%);
background-repeat:no-repeat;
}
.grain{position:fixed;inset:0;z-index:0;pointer-events:none;opacity:.04;mix-blend-mode:overlay;
background-image:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='160'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.85' numOctaves='2' stitchTiles='stitch'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E");}
.wrap{position:relative;z-index:1;max-width:720px;margin:0 auto;padding:0 24px 100px}
.topbar{display:flex;justify-content:space-between;align-items:center;font-family:"IBM Plex Mono",monospace;
font-size:12px;letter-spacing:.22em;text-transform:uppercase;color:var(--muted);padding:20px 0;
border-bottom:1px solid var(--line-soft);margin-bottom:52px}
.topbar .cy{color:var(--cyan)}
.eyebrow{font-family:"IBM Plex Mono",monospace;font-size:13px;letter-spacing:.28em;text-transform:uppercase;
color:var(--cyan);display:flex;align-items:center;gap:10px;margin-bottom:22px}
.eyebrow .dot{width:7px;height:7px;border-radius:50%;background:var(--cyan);box-shadow:0 0 12px 1px var(--cyan)}
h1.title{font-size:clamp(34px,5vw,52px);line-height:1.07;font-weight:600;letter-spacing:-.015em;margin:0 0 22px}
.subtitle{font-size:clamp(20px,2.4vw,24px);line-height:1.4;color:var(--paper-dim);font-style:italic;margin:0 0 30px}
.byline{font-family:"IBM Plex Mono",monospace;font-size:13px;letter-spacing:.04em;color:var(--muted);
text-transform:uppercase;margin-bottom:40px}
.byline b{color:var(--paper);font-weight:600} .byline a{color:var(--cyan);text-decoration:none}
.hero{width:100%;height:auto;border-radius:12px;border:1px solid var(--line);display:block;margin:0 0 48px;
box-shadow:0 34px 90px -34px rgba(0,0,0,.85)}
.prose{font-size:20px}
.prose p{margin:0 0 1.35em}
.prose h2{font-size:clamp(26px,3.1vw,33px);line-height:1.2;font-weight:600;margin:2.5em 0 .65em;
color:var(--paper);letter-spacing:-.012em}
.prose a{color:var(--cyan);text-decoration:underline;text-underline-offset:3px;text-decoration-thickness:1px}
.prose strong{font-weight:600;color:#fff}
.prose em{font-style:italic}
.prose ul{padding-left:0;list-style:none;margin:1.5em 0}
.prose ul li{position:relative;padding-left:1.6em;margin:.75em 0}
.prose ul li::before{content:"";position:absolute;left:0;top:.72em;width:7px;height:7px;background:var(--cyan);
border-radius:1px;box-shadow:0 0 8px rgba(43,230,211,.55)}
.prose blockquote{margin:1.7em 0;padding:.1em 0 .1em 1.2em;border-left:3px solid var(--cyan);
font-style:italic;color:#fff;font-size:1.14em}
.prose img{width:100%;height:auto;border-radius:12px;border:1px solid var(--line);margin:2.2em 0;
box-shadow:0 26px 64px -30px rgba(0,0,0,.85)}
.prose hr{border:none;border-top:1px solid var(--line);margin:2.8em 0}
.endbar{margin-top:60px;padding-top:30px;border-top:1px solid var(--line)}
.cta{display:inline-flex;align-items:center;gap:10px;font-family:"IBM Plex Mono",monospace;font-size:14px;
letter-spacing:.07em;text-transform:uppercase;color:var(--ink);background:var(--cyan);padding:15px 24px;
border-radius:9px;text-decoration:none;font-weight:600;box-shadow:0 0 34px rgba(43,230,211,.28)}
.meta-links{font-family:"IBM Plex Mono",monospace;font-size:12.5px;letter-spacing:.05em;color:var(--muted);margin-top:24px}
.meta-links a{color:var(--cyan);text-decoration:none}
@media(max-width:600px){ body{font-size:18px} .prose{font-size:18px} .wrap{padding:0 20px 72px} }
</style>
</head>
<body>
<div class="grain"></div>
<div class="wrap">
<div class="topbar"><span>Suede Labs AI · Field Note No. 05</span><span class="cy">suedeai.ai</span></div>
<div class="eyebrow"><span class="dot"></span>Tokenmaxxing</div>
<h1 class="title">A Company Spent $500,000,000 on AI in One Month — By Accident. They Call It “Tokenmaxxing.”</h1>
<p class="subtitle">AI is the biggest breakthrough of our lives. It’s also a historic bubble. A half-billion-dollar accident shows why both are true at once.</p>
<div class="byline">By <b>Jason Colapietro</b> · <a href="https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in">Originally on Substack</a></div>
<img class="hero" src="assets/tokenmaxxing-card.png" alt="Total AI charges, one month: $500,000,000 — spent by accident. A Suede Labs AI field note on tokenmaxxing.">
<div class="prose">
<p>A company accidentally spent <strong>$500,000,000</strong> on AI in
one month — because they forgot to set a usage limit.</p>
<p>It’s not a glitch. It’s the whole AI economy in a single invoice.</p>
<p>And it is <em>not</em> an argument against AI. It’s the best evidence
we have for the strangest fact about this moment: AI is the biggest
breakthrough of our lives <strong>and</strong> a historic bubble — at
the same time. Start with the receipts.</p>
<p><strong>The receipts — four things that happened this
spring:</strong></p>
<ul>
<li>A company <strong>reportedly spent ~$500M on Claude in a single
month — by accident</strong> — after rolling it out org-wide with no
usage caps (a consultant described it to <em>Axios</em>; the company is
unnamed). Agentic tools can burn ~<strong>1,000× the tokens</strong> of
a simple query.</li>
<li><strong>Amazon shut down its own internal AI-usage
leaderboard</strong> after employees gamed it with low-value prompts —
so the “leaderboard burns” pattern isn’t just Uber.</li>
<li><strong>Uber:</strong> 95% of engineers now use AI tools monthly;
per-engineer costs ran <strong>$500–$2,000/month</strong>; the year’s
budget was gone in 4 months.</li>
<li><strong>Microsoft pulled internal Claude Code licenses</strong>
after hitting that same $500–$2k/engineer range.</li>
</ul>
<p>Let me start with that first one.</p>
<h2 id="the-500-million-accident">The $500 million accident</h2>
<p>$500 million. In a single month. By accident.</p>
<p>Not on a data center. Not on a warehouse of GPUs. On <em>tokens</em>
— the metered little units you burn every time a model thinks for you.
According to a consultant who described the episode to <em>Axios</em>,
the company rolled Anthropic’s Claude out across the entire organization
and simply forgot to set a spending cap. No limits, no guardrails — just
thousands of employees pointing autonomous AI at their work. And
“agentic” tools, the kind that grind through multi-step tasks on their
own, can devour roughly a <em>thousand times</em> the tokens of a simple
chatbot question. Run that across a whole company with the meter
spinning and no ceiling, and you get an invoice the size of a Series A
round.</p>
<p>Hold that image, because it is the entire AI economy compressed onto
a single invoice. The technology was so capable, so eager, so
<em>useful-feeling</em> that an entire workforce reached for it
constantly — and the bill, when someone finally looked, was an
absurdity. Enormous consumption. Unknown value. A number nobody meant to
spend.</p>
<p>That is not a one-off horror story. That is the business model.</p>
<h2 id="the-leaderboard-that-ate-the-budget">The leaderboard that ate
the budget</h2>
<p>Take Uber. Last year it wanted its engineers using AI, so it did the
most Silicon Valley thing imaginable: it gamified the burn. The company
built an internal leaderboard ranking engineering teams by how many AI
tokens they consumed. Climb the board, win the glory.</p>
<p>It worked terrifyingly well. Ninety-five percent of Uber’s engineers
now reach for AI tools every month, at a cost that reportedly ran
anywhere from $500 to $2,000 <em>per engineer.</em> The contest was such
a runaway success that Uber torched its <em>entire</em> 2026 budget for
AI coding tools in four months. By April, a full year’s money was
gone.</p>
<p>Then Uber’s own president and COO, Andrew Macdonald, went on a
podcast and admitted he couldn’t tell you it was worth a dime.</p>
<blockquote>
<p>“If you’re not actually able to draw a direct line to how [many]
useful features and functionality you’re shipping to your users, that
trade becomes harder to justify.”</p>
</blockquote>
<p>The link between all that frantic token-burning and anything an
actual Uber rider would ever notice? “That link,” he said, “is not there
yet.” He even has a word for the dynamic — engineers racing one another
to consume tokens with nothing to show for it.
<strong>Tokenmaxxing.</strong></p>
<p>And Uber is not alone in discovering that its leaderboard had become
the point. Amazon reportedly shut down its <em>own</em> internal
AI-usage leaderboard after employees started gaming it with low-value,
throwaway prompts — racking up the meter to climb a ranking that
measured nothing but the meter. Microsoft, an AI superpower and an
Anthropic investor, quietly canceled most of its internal Claude Code
licenses once per-engineer costs hit that same $500-to-$2,000 range, and
moved its people onto cheaper in-house tooling.</p>
<p>So: a $500 million accident, a gamified budget set on fire, two of
the most sophisticated technology companies on earth pulling back on the
very tools they’re betting the future on. If you only read those
headlines, you’d conclude AI is a scam.</p>
<p>You’d be wrong. That’s the trap. Because here is the thing I’d
actually stake money on: <strong>all of this is happening, and AI is
still the most important technology of our lifetimes.</strong> Both
things are true at once. AI is, in all likelihood, the most
transformative technology any of us will ever witness — <em>and</em> it
is, right now, a bubble of historic proportions. Those aren’t competing
claims. They’re the same sentence. Tokenmaxxing is just the receipt.</p>
<h2 id="the-case-for-the-revolution-take-it-seriously">The case for the
revolution (take it seriously)</h2>
<p>It’s easy to roll your eyes at “most transformative technology of our
lifetime.” We’ve heard it about crypto, about the metaverse, about 3D
TVs. Pattern-matching says: hype.</p>
<p>But pattern-matching is exactly the trap. The honest version of the
bull case doesn’t rest on a demo or a keynote. It rests on the fact that
the thing is already load-bearing.</p>
<p>Uber’s own CEO, Dara Khosrowshahi, says roughly <strong>10% of the
code his company commits is now written by autonomous agents.</strong>
That’s not a pilot — that’s a double-digit slice of a flagship
engineering org’s output, shipping today. The same $500-a-seat tools
driving those terrifying invoices are also the reason 95% of his
engineers won’t work without them. People don’t voluntarily build their
daily workflow around a toy.</p>
<p>The adoption curve is the tell. The technologies we now call
transformative — electricity, the automobile, the internet — took
<em>decades</em> to reach the households and workflows that AI saturated
in about two years. The capability is real, it compounds, and it shows
no sign of plateauing. When skeptics sniff that “it’s just
autocomplete,” they’re describing a product that shipped 18 months
ago.</p>
<p>So: revolution. Genuinely. The people calling this a fad are not
going to age well.</p>
<p>And yet.</p>
<h2 id="the-case-for-the-bubble-follow-the-money">The case for the
bubble (follow the money)</h2>
<p>A revolution and a bubble are measured on different instruments. The
first you measure in capability. The second you measure in cash flows.
And the cash flows are <em>deranged.</em></p>
<p>Start with the company at the center of the story. OpenAI burned
roughly <strong>$9 billion in 2025</strong>, and is on track to burn
something like <strong>$17 billion in 2026.</strong> It pulls in around
$13 billion in revenue and spends north of $22 billion to do it. The
most valuable AI company on earth loses money on a staggering scale —
and the losses <em>grow with success.</em></p>
<p>That last clause is the one that should keep you up at night.
According to financial data that leaked early this year, OpenAI’s
<strong>inference costs — the cost of actually answering your prompts —
exceeded its subscription revenue.</strong> Read that again. Every
additional ChatGPT user, every “wow, it did my homework” moment, made
the unit economics <em>worse.</em> The more people loved it, the faster
the money went.</p>
<p>This is the part the public never sees, because the price you pay has
almost nothing to do with the cost you incur. You are not paying for the
compute behind your chatbot. You’re paying for a sliver of it, and a
wall of venture and hyperscaler capital is paying the rest — betting
that someday, 2028, 2030, <em>someday,</em> the curve bends and the
subsidy ends. One analyst’s phrase for the moment we’re entering:
<em>“the era of subsidized AI usage is over.”</em> That subsidy is the
only reason the magic feels free. It isn’t free. Someone is eating the
difference, and they are eating billions of it.</p>
<p>The $500 million accident is simply what it looks like when the meter
becomes briefly, accidentally visible. Most of the time the true cost is
hidden behind a flat monthly price and a mountain of someone else’s
money. For one month, one company saw the real number. The rest of us
are still looking at the subsidized one.</p>
<h2 id="the-circular-money-this-is-the-1999-part">The circular money
(this is the 1999 part)</h2>
<figure>
<img src="assets/tokenmaxxing-quote.png"
alt="The clouds fund the labs. The labs rent the clouds. Every lap books as “demand.”" />
<figcaption aria-hidden="true">The clouds fund the labs. The labs rent
the clouds. Every lap books as “demand.”</figcaption>
</figure>
<p>Now layer on the financial plumbing, because this is where it stops
looking like a frontier and starts looking exactly like the dot-com
peak.</p>
<p>The same handful of balance sheets are funding every side of the
table. Amazon committed up to <strong>$25 billion</strong> more to
Anthropic this spring — and then turned around and reportedly poured
<strong>$50 billion</strong> into OpenAI, Anthropic’s chief rival,
becoming its exclusive third-party cloud distributor. Microsoft has tens
of billions in OpenAI <em>and</em> billions in Anthropic. Read that
twice: the cloud giants are bankrolling <em>both</em> of the warring
labs, while the labs spend that very money renting the giants’
clouds.</p>
<p>That’s the loop. The chipmakers invest in the labs; the labs spend
the money on chips. The cloud providers invest in the labs; the labs
spend the money on cloud. Capital flows out one door and back in
another, and every lap around the track books as “revenue,”
“investment,” and “explosive demand” — often all three at once. When a
single dollar can be counted three times, you are not looking at a
market. You are looking at a hall of mirrors.</p>
<p>And the people inside the mirrors know it. OpenAI’s revenue chief
reportedly accused Anthropic, in an internal memo, of <strong>inflated
accounting</strong> — of juicing its reported numbers. Anthropic, for
its part, claims its run-rate revenue tripled past <strong>$30
billion</strong> in a matter of months. They may both be right about
each other. When the biggest players start publicly questioning whether
anyone’s revenue is even real, the growth story has entered its decadent
phase.</p>
<p>Then there’s the quietest signal of all — the one I keep coming back
to. <strong>The true believers are trimming their own usage.</strong>
Microsoft canceled its internal Claude licenses. Amazon killed its gamed
leaderboard. Uber’s COO is doing arithmetic out loud on a podcast. These
are not the doubters. These are the most committed, most sophisticated
buyers on the planet — and even <em>they</em> can’t draw the line from
the spend to the value.</p>
<h2 id="this-is-what-a-revolution-looks-like-from-the-inside">This is
what a revolution looks like from the inside</h2>
<p>Here’s the reframe, and it’s the whole point: <strong>the bubble is
not evidence against the revolution. The bubble is what a revolution
looks like while it’s happening.</strong></p>
<p>Go back to the railways of 1840s Britain. Real technology — it
rewired civilization. It also produced “Railway Mania,” a speculative
frenzy that wiped out a generation of investors when it burst. The
tracks stayed. The railway <em>companies</em> mostly didn’t.</p>
<p>Go to 1999. The internet was every bit as transformative as the
prophets promised — more, probably. And the dot-com bubble was every bit
as fake as the cynics said. Pets.com died. So did hundreds of others.
The fiber they over-built in their delusion became the backbone that
Google and Netflix and Amazon were later built on. The economist Carlota
Perez has a name for this rhythm: the “installation phase” of every
technological revolution, when financial capital floods in, builds far
more capacity than the present can use, and detonates — <em>and
then</em> the technology quietly delivers on everything the bubble
overpromised, just on a slower clock and for different owners.</p>
<p>The infrastructure outlives the investors. The capability outlives
the valuations. <strong>Amazon is the lesson, not the exception</strong>
— it lost more than 90% of its value when the bubble popped, then went
on to become one of the most important companies in history. Both things
were true about Amazon in 2000: it was a real revolution <em>and</em> a
stock in a bubble. Most people could hold only one of those ideas. The
ones who held both got rich.</p>
<h2 id="what-tokenmaxxing-actually-teaches">What tokenmaxxing actually
teaches</h2>
<p>So what do we do with that $500 million invoice, and with Macdonald’s
confession? Not “AI is fake.” Something far more useful.</p>
<p>“Tokenmaxxing” is the perfect word because it names the exact failure
mode of a bubble: <strong>mistaking activity for value.</strong> A
leaderboard measured token consumption, so token consumption is what it
got. A rollout had no spending cap, so spending is what it got. Usage
went vertical. Whether anything <em>good</em> came out the other end — a
feature, a fix, a happier customer — was left, conveniently, unmeasured.
The metric became the mission.</p>
<p>That gap — between <em>use</em> and <em>value</em> — is the entire
bubble in miniature. The technology is real. The capability is real. But
a great deal of today’s spending is the corporate equivalent of climbing
a leaderboard: adopting because everyone’s adopting, because the board
asked about the “AI strategy,” because the subsidy makes it feel cheaper
than it is. When that subsidy ends and someone finally has to draw
Macdonald’s line from cost to value, an enormous amount of this spending
will not survive the audit.</p>
<p>And that’s <em>fine.</em> That’s the system working. The unit
economics will get fixed, or the companies with broken ones will die.
The valuations will reset. Some labs that look invincible today will be
footnotes. The frenzy will burn off like fog. And underneath it, the
actual technology — the agents writing 10% of the code, then 30%, then
more — will still be there, compounding, indifferent to whose stock
cratered.</p>
<h2 id="how-to-hold-both-ideas">How to hold both ideas</h2>
<p>The dumbest position in 2026 is certainty in either direction.</p>
<p>The bull who can’t see the bubble is going to get vaporized when the
subsidy ends and the circular money stops circling. The bear who thinks
the bubble <em>is</em> the whole story is making the precise mistake
people made shorting Amazon in 2001 — confusing the death of the
financing with the death of the technology.</p>
<p>I’m not writing this from the cheap seats. I build an AI company,
<strong>Suede Labs AI</strong>, on a bet about what’s left when the fog
clears — not the valuations or the leaderboards, but
<strong>ownership</strong>: proof of who created what the machines make,
and the right to keep it. When the $500 million accidents get audited
and half the spend evaporates, the durable question won’t be “how many
tokens did you burn.” It’ll be “what did you actually create — and can
you prove it’s yours.”</p>
<p>So hold both at once. Yes, you are living through the most
consequential technology shift of your life. Also yes, a staggering
amount of money is about to be lit on fire by people who confused tokens
consumed with value created. The revolution is real <em>and</em> the
price is wrong. Both things are true.</p>
<p>Uber’s COO couldn’t draw the line from the spend to the product, and
he was honest enough to say so out loud — which is worth more than a
thousand triumphant keynotes. But “I can’t yet draw the line” is not the
same sentence as “there is no line.” It’s the sentence every
transformative technology whispers right before it stops being a bubble
and starts simply being the world.</p>
<p>The tracks stay. The fog burns off. Don’t mistake one for the
other.</p>
<hr />
<p><em>Jason Colapietro writes and builds as <strong>Johnny
Suede</strong>, founder of <strong>Suede Labs AI</strong>. Suede builds
AI tools that secure ownership — proof of creation, IP, and attribution
— so that when the hype burns off, creators still hold what they made.
More at <a href="https://suedeai.ai">suedeai.ai</a>.</em></p>
<p><strong>Sources:</strong> <a
href="https://www.tomshardware.com/tech-industry/artificial-intelligence/mystery-company-accidentally-blew-usd500-million-on-claude-in-a-single-month-failed-to-put-usage-limit-on-licenses-for-employees">Tom’s
Hardware — the $500M Claude bill</a> · <a
href="https://fortune.com/2026/05/26/uber-coo-ai-spending-tokens-claude-code/">Fortune
— Uber’s 2026 AI budget</a> · <a
href="https://www.axios.com/2026/04/13/openai-microsoft-anthropic-amazon">Axios
— OpenAI, Microsoft & Anthropic</a> · <a
href="https://www.geekwire.com/2026/amazon-doubles-down-on-anthropic-with-25b-investment-mirroring-its-openai-cloud-deal/">GeekWire
— Amazon’s $25B Anthropic deal</a> · <a
href="https://www.uncoveralpha.com/p/the-era-of-subsidized-ai-model-usage">UncoverAlpha
— the end of subsidized AI</a> · <a
href="https://fortune.com/2025/11/12/openai-cash-burn-rate-annual-losses-2028-profitable-2030-financial-documents/">Fortune
— OpenAI’s losses through 2028</a></p>
</div>
<div class="endbar">
<a class="cta" href="https://jasoncolapietro.substack.com/p/a-company-spent-500000000-on-ai-in">Read & subscribe on Substack →</a>
<div class="meta-links">Source & graphics on <a href="https://github.com/JasonColapietro/tokenmaxxing">GitHub</a> · © 2026 Jason Colapietro (Johnny Suede)</div>
</div>
</div>
</body>
</html>