#AnthropicSigns35BCloudDeal

The $35 billion figure is huge, but what's really interesting is the financing structure surrounding it.

Anthropic signed an agreement with Nvidia-backed Lambda for approximately 350 MW of capacity in a Texas data center; Nvidia itself is involved in leasing and chip supply. Anthropic also recently invested another $45 billion in Nscale for AI cloud capacity.

My view: This is becoming the financial engine of AI infrastructure.

1. Anthropic isn't buying computing power because it thinks GPUs are cheap. It's buying because computing power scarcity has become a strategic constraint.

Claude usage—especially Claude Code—is rapidly increasing, and Anthropic seems to be securing capacity years in advance rather than risking exposure to GPU constraints. Signing $35 billion, $45 billion, and other massive computing deals simultaneously suggests the company expects demand to remain extremely high.

2. Nvidia is shifting from a chip supplier to an infrastructure financier.

This part is the most fascinating to me.

Nvidia sells GPUs → helps finance/guarantee the infrastructure → the infrastructure is built → Anthropic leases computing resources → Lambda buys/distributes Nvidia GPUs → Nvidia generates enormous demand for additional hardware.

It's a powerful mechanism.

And this isn't an isolated case: Nvidia is reportedly committing $36 billion as part of its broader AI computing financing initiative, while other news raises questions about the potential financial risk Nvidia might ultimately bear.

3. However, the "AI bubble" argument also becomes legitimate here.

There's a crucial distinction between:

"AI is being exaggerated."

and

"The financing structure supporting AI infrastructure has become too aggressive."

I think the second concern is far more credible.

Claude, ChatGPT, Gemini, etc. If projects generate enough revenue to justify the enormous amount of repetitive processing power spent, infrastructure investment can work perfectly well.

However, if AI model prices fall rapidly while inference efficiency increases, today's $35 billion/$45 billion capacity commitments could become expensive legacy contracts.

This is the risk of locking in processing power without knowing what the cost of a unit of AI intelligence will be in 3-5 years.

The biggest winner could actually be Nvidia.

Ironically, Anthropic's competitive success could benefit Nvidia even if Anthropic eventually develops more proprietary silicon.

Nvidia's strategy increasingly seems to be:

“I don't need to own the AI ​​company. The AI ​​ecosystem needs to consume enormous amounts of processing power.”

This is a very different business model than simply selling GPUs.

And Nvidia is also investing in companies and technologies that could ultimately challenge parts of its own hardware monopoly.

Here's what I'll be watching next:

For me, there are three numbers more important than the headline figure of $35 billion:

1. Human-driven revenue growth — can revenue ultimately support computing spending at this scale?

2. Computing usage — are these data centers actually operating near capacity?

3. Cost per token — how quickly is the cost of producing useful AI output falling?

If the cost per token falls while revenue and usage continue to explode, it's not necessarily a bubble. It's potentially the early infrastructure phase of a massive new computing industry.

If capacity continues to be financed faster than AI companies can monetize it, then we're facing something far more bubble-like.

In conclusion: I'm optimistic about the demand for core AI infrastructure, but increasingly cautious about the financing architecture. The technology race is turning into a capital race, and Nvidia is positioning itself not just as an arms dealer, but increasingly as a banker.

$NVDA ‌