What the AI skeptics are missing

Hi, and happy Thursday.

The NVIDIA scandal is the latest blow to confidence in the AI industry.

Skepticism around AI has been growing in recent weeks, as we hurtle towards an IPO - of Anthropic, next week - that many have been expecting to “pop” the bubble.

It began a few weeks ago with a surge in interest rates demanded by lenders for CoreWeave’s debt financing. CoreWeave and similar startup “neoclouds” rely heavily on this capital to build out their data centers.

But lenders have started to grow increasingly concerned.

To justify the massive buildout in data center capacity that is currently underway, AI consumption by the likes of you and must scale to at least seven times current levels by 2030, according to an October 29, 2026 analysis by BCG.

This expansion also requires generating $1.8 trillion in revenue - equivalent to roughly one billion users spending $150 per month on AI services.

All this has to happen while, at the same time, more efficient chips and models are driving down the cost of AI.

The math might not quite add up.

But, it got worse.

In the US, opposition to AI data centers contributed to the Democrats’ sweeping victory in the November 3rd midterms - although it wasn’t the primary reason.

By the time of the election, an astonishing 76% of American voters opposed having a new AI data center built in their community. The dominant concerns were AI’s impact on jobs - particularly for early career professionals - as well as electricity costs, water consumption and environmental damage.

And now it's even worse!

On November 4th a congressional committee requested a closed-door briefing with Nvidia CEO Jensen Huang, scheduling it for November 18th. He may not be CEO by then.

Congress wants answers about his relationship with Alice Huang, a business executive suspected by US intelligence of spying for China, and whether sensitive information had been compromised.

What does this mean?

Next week may be one of the most historic weeks in AI.

On Wall Street, bankers are poised to celebrate “American technological supremacy” - Anthropic turning the application layer it has built on Nvidia’s chips into a trillion-dollar valuation.

On the other side of the country, Congress will be trying to determine if that “American technological supremacy” is protected by export controls or the marital fidelity of a billionaire in a leather jacket.

So what are the AI skeptics missing?

The first is that the AI technology investment boom is funded by hugely profitable companies. Nvidia, Microsoft, Google, Amazon and Meta have enormous, highly profitable existing businesses and decades of unspent earnings in the bank.

Secondly, when not building their own data centers, many of these same companies are the customers of CoreWeave and other neoclouds. CoreWeave reported this week that it expects approximately $9 billion in customer prepayments and deposits over the next three months, firming up support for the company.

Thirdly, there is unprecedented demand for AI. While you and I are stuck paying back the mortgage loan for our homes over a 30 year period, the payback period for a data center is 2.5 years.

A slowdown in the buildout of data centers, due to political opposition, will only reduce this payback period.

Lastly, the Anthropic IPO prospectus reveals the company - and the “state of AI” - to be much healthier than people had assumed.

Enterprise adoption is accelerating.

The number of customers spending at least $100,000 with Anthropic annually grew from 200 in December 2024 to 11,900 by September 2026 - a 60-fold growth in just 21 months.

More than 1,900 customers spend at least $1 million annually and approximately 250 customers spend at least $10 million annually.

At these growth rates, businesses - not consumers - look likely to meet that seven-fold data center capacity requirement.

Anthropic’s IPO’s disclosures also reveal remarkable capital efficiency. The company has raised $141 billion since inception but still holds approximately $132 billion in cash and short-term investments - equivalent to 93% of everything it has raised. (That doesn’t mean it has spent only 7%, because customer revenues are also funding its operations.)

More remarkably, Anthropic generated $6.6 billion in operating cash flow during the first nine months of 2026, demonstrating that its rapidly growing business is already generating significant cash.

How is this possible?

Anthropic can reportedly deliver the same AI capability within roughly one-tenth the cost the following year, while not necessarily passing on all the savings to customers.

So where is the ROI?

It is becoming increasingly clear - at least to me - that AI’s immediate economic value is coming less from creating entirely new industries and more from dramatically expanding what really capable individuals can accomplish.

Smartphones allowed people to work outside the office. AI goes further: it allows individuals to perform work that previously required specialized skills, substantial time or entire teams.

Someone with initiative, judgment and a willingness to learn can now research, write, code, analyze and execute across disciplines.

At Prescouter, this is increasingly what we are looking for from our teams and new hires.

The scarce resource is increasingly not the ability to perform individual tasks, but the ability to figure out what needs doing and take responsibility for getting it done.

And where we see this on display everywhere right now is with hackers, who often share the same traits we all need to be hiring for.

Approximately $6.34 billion was stolen from Binance. While the attack involved months of preparation, fake job interviews, compromised developer computers and stolen credentials, Binance’s own investigation confirms the attackers used AI tools as part of their infiltration.

See what I mean about individuals performing work that previously required specialized skills, substantial time or entire teams?

Meanwhile others have reverse engineered the entire suite of applications from companies such as Microsoft and Adobe and published them for free and open source. And fundamental problems in math and physics continue to be “unlocked” by new generations of AI models.

Best,

Dino