Developing an open framework for AI-mediated discovery — how AI systems build trust, resolve decisions and form defaults.
London
Joined March 2012
- Tweets11K
- Following1.9K
- Followers1.4K
- Likes12.2K
Darren retweeted
The problem with using Dot com era strategy accounting practices to hide OpEx is that the disconnect from reality is so insanely obvious, it takes zero financial awareness to see the legal “fraud” hiding the truth.
Anthropics compute costs rise to $96 billion next year. 5x higher than 2026 when they got what appears to be a 75% off compute deal from $AMZN
They’re also facing a major drop in token prices.
Yes I’m a debt guru, but everyone s awe and amazement reveals the stupidity in the numbers and underlying OpEx manipulation that perfectly aligns to what was done in final months of DotCom crash.
Anthropic spent $7.2 billion for 1.4 GW of compute in 2025 (based on reports from investor presentation).
Anthropic told investors in presentation they were tracking 5 GW in 2026. They are tracking $20 billion spent.
Anthropic spent $5.14 billion / GW in 2025
Anthropic is tracking $4.0 billion / GW in 2026 spent (4x lower than industry standard of $16-$22 billion / GW)
Anthropic just signed a contract with N-Scale at $16 billion / GW in 2026, 4x higher than its reporting in its S-1.
The magic of $AMZN deal. 75% off compute.
Their true compute cost would have been closer to $80 billion in 2026 vs $20 billion.
Darren retweeted
leveraged neoclouds are bad.
except the one i’m running.
-@elonmusk
Pretty much.
As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.
SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.
That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
Darren retweeted
If you are wondering why semiconductors peaked in June this is the reason…the market sniffed out that incremental demand was coming from a debt fueled unprofitable speculative sector. Neo clouds.
Pretty much.
As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.
SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.
That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
Darren retweeted
Elon Musk just told the entire world that the AI bubble has popped... buckle up.
Pretty much.
As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.
SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.
That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
Darren retweeted
“That means the 200+ neoclouds are going to see guppy-level death and consolidation.”
If neocloud stocks tank on Monday, I may have accidentally accelerated the guppy-level death...
Darren retweeted
Jack confirms“there is no moat”
So what does that mean for the MASSIVE paper gains buried in the margins of $MSFT $AMZN $NVDA $GOOGL etc?
Oh yeah, and:
“Several asset managers have invested an estimated $6 billion in Anthropic, led by American Funds parent Capital Group, which had $10 billion invested. Fidelity ranks second with roughly $5 billion in the company, making the fund families Anthropic’s largest mutual fund shareholders.” Morningstar
Darren retweeted
There is no moat in the code anymore.
What's left:
Distribution.
Taste.
Community.
Anything physical.
If you don't have one of those, you're a feature someone will clone in a weekend.
Darren retweeted
On the bright side, without a free fall in token prices and a compute glut, the next generation of AI companies couldn't exist. It was only after the collapse of the Web 1.0 infrastructure layer that Web 2.0 and tech giants, like Netflix, Amazon Prime, smartphones, and Google Map could rise and benefit the masses.
Darren retweeted
the negative case against AI gets stronger every day; even Altman admits the problems are very serious with no obvious solution.
the positive case, meanwhile, looks weaker every day. a quote from Apollo’s Torsten Slok:
“Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech. Revenue per worker is… flat for the S&P 493…. Profit margins are… flat for the S&P 493…. No signs of AI boosting [measured] productivity growth…. Mainstream corporate America is pulling back and starting to view token spending… as a cost center…. Mainstream corporate America does not need the latest model and is unwilling to pay for it. As long as agentic AI and token spending are confined to a small part of the economy, the macroeconomic effects of AI will be very limited.”
apollo.com/wealth/insights-n…
Darren retweeted
Guess what, Anthropic S1 filing, a must have to pursue an IPO, not surprisingly hasn’t been made public by Friday. I am starting to think it might never publicly released because the moment the world sees what a horrible business Anthropic is, then AI bubble will implode
Darren retweeted
Not seeing a way forward for Anthropic and OpenAI revenue growth now that there are about 873 easy ways to bypass their models in favor of free or extremely cheap specialized alternatives. They went all-in on “best” when “sufficient” and “focused” solved 90% of use cases.
Darren retweeted
kimi k3.1 has mastered the art of distilling models until they outperform the models they learned from
hopefully anthropic will finally shut up about the whole "they just distill" nonsense
Darren retweeted
China did it again!
They cracked recursive self-improvement with a 27B model.
Researchers dropped an AI agent into a complex computer terminal environment.
They gave it a goal. And then they stepped back.
At first, the AI failed. It couldn't complete a single hard task. 0% success rate.
But then, the Recursive Self-Improvement loop kicked in.
Through pure trial and error, the AI discovered a tiny success.
Instead of waiting for a human to log it, the AI rewrote its own memory of the success.
It analyzed what worked, refined its own core policy, and permanently upgraded its own intelligence.
Then it went back to the terminal. But this time, it was smarter.
It repeated this loop. Over and over.
Generate. Verify. Rewrite. Refine.
The results are terrifyingly impressive.
On the hardest coding and system administration benchmarks, the model went from literally 0% capability to nearly 20% completion.
On another complex task suite, it skyrocketed from 31% to 79%.
Not by training on the internet. By training on its own internal discoveries.
Darren retweeted
This is probably one of the most ridiculous markets I have ever seen. I get that stocks can’t always go up, I get that they need to go down, but nothing about this price action makes sense.
The market is essentially pricing in an AI slow down that isn’t being signaled anywhere.
