The faster you grow, the faster you go bankrupt
Sam strikes back
Sam barged in today wearing that look…
You know the one.
The “I’ve solved it” look.
The look that says she’s about to drop a tweet on my desk and demand - no COMMAND - that I revise everything I’ve ever written.
This is sadly becoming a pattern.
Last November she came in with NASDAQ multipliers proving we were “early in the bubble”:
A few weeks later it was a 5,000-word JPMorgan conspiracy theory about a quiet war on Bitcoin that turned out to be... not a war at all:
Now she’s back with this:
“Anthropic’s CEO admitted that at the beginning of the year, they predicted a tenfold increase in revenue. However, no one expected that in just four months, the revenue growth rate would surge 80 times. Those who hold Anthropic stock include Google, Amazon, Zoom, Microsoft, and Nvidia. Get ready to make a fortune!”
— Some random tweet
She slammed it down. “AI IS NOT IN A BUBBLE! THIS PROVES IT!”.
I sighed.
Then I started checking. Because yep… My thesis is built on data. Which is a pretty foreign concept to her…
Every time Sam discovers The Truth, I lose half a day digging for the things she didn’t read.
One good thing though: she’s a never-ending source of bubble-information!
Let’s start with what I found out: The 80x is real. Dario Amodei said it on stage at Anthropic’s developer conference on May 6 (link). The company planned for 10x growth, and got 80x instead, which pushed its annualized revenue from $9 billion at the end of 2025 to $30 billion by April. That’s faster than Salesforce, faster than ServiceNow, faster than basically any enterprise software story in history.
Genuinely impressive.
One problem… Sam (and probably a lot of AI-fanboys) didn’t read the next bit.
Anthropic’s gross margin is 40%.
Forty.
That means for every dollar of revenue, sixty cents goes straight back out the door to pay for inference compute. Before salaries. Before training the next model (which costs a few billion). Before R&D, sales, marketing, and the office snacks. Always keep the snacks flowing!
In January, Anthropic actually CUT its 2025 margin projection by 10 percentage points because inference costs ran 23% over budget (link). The opposite of operating leverage. They serve more customers, the cost of serving each one goes up.
Anthropic was burning roughly $2.8 billion in cash in 2025. Revenue tripled, but so did the burn.
So the strategy is: lose money on each customer, then make it up in volume?
“‘Tis but a flesh wound!”
The cherry on the cake comes from Amodei himself, on the record, to Fortune in March:
“If I’m just off by a year in that rate of growth, or if the growth rate is 5 times a year instead of 10 times a year, then you go bankrupt.”
— link (non-paywalled)
The CEO of the most successful AI company by revenue is saying that growing “only” 5x a year - which would be one of the fastest growth rates in software history - is a path to insolvency.
If that’s not a bubble walking around with a huge neon sign on its head, I don’t know what else it could be.
Now about Sam’s investor list. Google, Amazon, Microsoft, Nvidia.
She thinks this is the reassurance.
To me? It’s the smoking gun.
Those are Anthropic’s compute providers. Amazon and Google are where Anthropic rents its GPUs. Nvidia makes the chips. So Google invests cash into Anthropic, which then pays that cash back to Google for cloud capacity. Amazon does the same. Nvidia invested in Anthropic, and Anthropic uses that money to buy Nvidia chips. This is the exact circular financing pattern I covered here:
And updated here:
TL;DR: One big multi-trillion circle-jerk.
(Also worth noting: Zoom isn’t on the cap table. I have no idea where Sam got that one.)
(And you can’t even buy Anthropic stock. It’s private. The “make a fortune” pitch doesn’t apply unless you have a few hundred million sitting around, in which case you probably already own some.)
Right. So let me take a step back from all the negativity I’m spouting and grant Sam some positive news. Just anything positive in this AI field.
Don’t say I didn’t try!
There IS a profitable AI company. One that matters, anyway.
It’s Nvidia.
Last quarterly report: $68.1 billion in revenue. Full year revenue: $215.9 billion, up 65%. Net income $120 billion. GAAP gross margin above 70% on the data centre business. A 56% net profit margin. That is a real business, justifying some fraction of its market cap.
Notice though what makes Nvidia profitable.
It isn’t that AI works.
It’s that everyone else THINKS AI works.
Nvidia sells the picks and shovels to a gold rush where most of the gold seekers are losing money.
Anthropic burns cash on Nvidia chips. OpenAI burns cash on Nvidia chips. CoreWeave borrows money to buy Nvidia chips. Meta, Microsoft, Google, Oracle and Amazon between them spend hundreds of billions a year on Nvidia chips. Then Nvidia turns around and invests in those same companies.
So yes, Nvidia is profitable, but its profitability is the mirror image of everyone else’s losses.
And if the cash flow into the pile slows, Nvidia’s revenue will slow.
There is no second customer base. There is no diversified buyer pool. There are the hyperscalers, and there are the hyperscalers’ funded startups, and that’s basically it. (Plus Zoom, allegedly.)
Now look around at what else is happening because of all this.
Q1 2026 alone, the tech industry shed nearly 80,000 jobs. Roughly half of those - 37,638 to be specific - were attributed directly to AI. Through May we are already past 128,000 tech workers cut. Amazon: 14,000. Microsoft: 15,000. Salesforce: 4,000. Meta announced 8,000 in April with internal guidance hinting at up to 20% across the full year, with the freed compensation budget “redirected toward AI research and infrastructure”. Oracle dropped over 10,000. Cloudflare cut 20% of its workforce last week.
Half a million tech workers in two years.
Sam Altman himself admitted at the India AI Impact Summit that “there’s some AI washing where people are blaming AI for layoffs that they would otherwise do”. Translation: companies are firing people to free up cash, then putting AI on the press release because it sounds visionary. The juniors get fired so that capex can flow into Nvidia’s quarter. Salary line gets cannibalised to feed an infrastructure binge that 95% of enterprises report generates zero return.
This is the same dynamic I laid out here:
and here:
They ain’t buying productivity.
They are buying optionality.
The narrative requires that the spend be justified by AI productivity gains. The actual mechanism is that the spend is funded by removing the people who used to do the work, regardless of whether AI can actually do it.
The juniors get fired. The seniors stay. The hardware gets bought. The shareholders get nervous. The CFO restructures.
Then next quarter another round of layoffs. Another round of capex commitments. Another round of investor day slides about “AI-driven productivity gains”.
And Sam comes in again with another tweet about how this proves it’s all going to work.
Look.
Anthropic is one of the great companies in the AI world.
Genuinely.
I use it daily for my coding tasks.
Best technical product in the field by a wide margin.
Amodei isn’t a charlatan.
The 80x is real revenue that comes from real customers paying real money for software they actually use.
That doesn't make it not a bubble.
A bubble isn’t “the technology is fake”.
A bubble is “the capex required to chase this revenue dwarfs any plausible cash flow it can produce”.
That is my entire argument... Right there…
But I’m the stupid one not understanding how markets work. And Sam is gaining 30% on her stocks in a few weeks.
I truly hope she has a good sense of when to bail.
But I’m afraid she won’t.
You should show this to Sam:
https://cybercultural.com/p/internet-1995/
and this:
https://www.gwsmedia.com/articles/how-internet-system-requirements-have-changed
The point is, we're no longer in a growth market. The present market is more like a race to the bottom, where there is no first mover advantage anymore, in fact it can be a significant disadvantage as we see with AI chips.
I'm probably not the best example of market demand, but I'm not that far off. I haven't bought a new computer in over 10 years, my cell phone is a 3G Galaxy 5, which is practically an antique, but it does everything I need it to do. My provider keeps warning me that they're about to abandon 3G. Ha! They've been saying that for years. When it actually happens I'll get a new phone, but not until. Oh, and they were going to install optical for free last year. I told them forget it. I don't need the speed. The hookup is on the pole, I'll let the next owner of this house worry about it.
Not to belabour the point but my Korg MI which I bought in 1982 still works fine, although now I only use it for MIDI because the free open-source synthesizer I use is better than any current dedicated hardware (eat that Roland!). Same goes for my Bryston amp and Advent Legacy monitors which I bought around the same time. If it ain't broke, don't replace it! Yeah, maybe the tech has improved somewhat since then, but you'll never hear the difference in MP3-360. Even FLAC won't get you there unless you're willing to spend at least 10 grand on a sound system, and most people who'd do that are into vinyl anyway!
Most people don't think of 1929 as a tech bubble, but it surely was. Look at the hot stocks back then: automobiles, radio, phonographs, washing machines, refrigerators, electric lighting, toasters, curling irons. It was all new stuff that few people had and everybody wanted. The similarity to 1995 is striking.
Speaking of cars, here’s a compressed history of the automobile. Notice how many players fall by the wayside, and how everything rolls off towards the end.
https://americanbusinesshistory.org/a-century-of-automobile-industry-history-in-charts/
When I first got interested in cars in the late 60s it was all about speed and styling. They looked great and went fast, which was what everyone wanted. Chick magnets. Today there hasn’t been a significant improvement in automotive engineering for over 20 years. It’s all ‘form factor’ now - all bells and whistles in a desperate attempt to sell you something that’s no better than the last 3 or 4 cars you owned. Self-driving? Give me a break. That only appeals to fleet operators. The people who spend serious money on a car want something they can actually drive. EVs may have a future - that remains to be seen, but the point is, you’re still dealing with market saturation, as with everything else. All the seats are occupied.
I’m a country bumpkin… I do not understand how a population without wisdom and extra disposable income will be able to use or afford AI “knowledge” for anything meaning or productive and then in 20 years when there is total population collapse have this be meaningful or profitable for the people selling it. Won’t a huge part of the population need FOOD?!? Meaning: No Food on Shelves = Lowering the demand for creating AI cat videos and meaningless inquisitive searches on chatbots for hallowed (not) epistemological musings? Not to mention the ENERGY required on behalf of both the producer AND consumer to CREATE or USE AI? How does everything keep going UP?!??? From someone who is seeing everything go to $h!+ in real-time?!???