Anthropic just committed $11.6 billion over seven years to Akamai’s cloud infrastructure. That’s not a typo, and it’s not GPUs.

The deal centers on CPUs, a notable departure in an industry that’s been GPU-obsessed since the current AI wave began. The commitment could grow to roughly $20 billion depending on usage. In an arrangement that blurs the line between customer and investor, Akamai is giving Anthropic a potential stake of up to 5% of its stock that increases as Anthropic’s spending climbs.

This isn’t how cloud deals typically work. The equity component suggests Anthropic negotiated hard on pricing and Akamai needed the revenue certainty badly enough to offer ownership stakes. For context, $11.6 billion is significant even by AI standards. It’s more than Anthropic has raised in total funding to date.

The CPU angle is the real story. While competitors pile into Nvidia’s latest chips, Anthropic is making a bet that inference workloads, which is where the money actually gets made in AI, don’t need cutting-edge GPUs for every task. CPUs are cheaper and more available. If Anthropic can run enough of its workload on traditional processors, the economics shift dramatically.

Akamai gets revenue visibility it desperately needs. The company has been transitioning from its legacy content delivery business to cloud services, competing against AWS, Google Cloud, and Microsoft Azure. Landing a customer willing to commit $11.6 billion changes the trajectory of that business overnight.

The timing matters too. Anthropic is reportedly preparing for an IPO, and its seven co-founders are seeking 50.1% voting control in a structure they’ve asked shareholders to approve. Between the Akamai infrastructure deal and the governance ask, the company is making big moves to position itself before going public.

Meanwhile, the British neocloud play

While Anthropic is betting on CPUs, British AI cloud provider Nscale secured $3.36 billion in convertible financing ahead of its own planned US IPO. The round came from Third Point, Nvidia, and other investors.

Nscale is building AI-specific data centers, the kind purpose-built for training and inference workloads. The convertible structure is interesting because it delays the valuation question until the IPO, which suggests either the company or investors (or both) think the valuation will be substantially higher by then.

Nvidia’s participation is notable. The chip maker has been investing in cloud providers that buy its GPUs, essentially financing its own customers. It’s worked well so far. As long as demand for AI compute keeps growing, everyone in the chain benefits.

These two deals, Anthropic’s CPU commitment and Nscale’s GPU-focused raise, represent competing theories about AI infrastructure. Nscale is building for the current paradigm where more GPUs equals more capability. Anthropic is testing whether you can route around GPU scarcity and cost by redesigning workloads for different hardware.

Both are raising enormous sums. Both are racing toward public markets. The difference is in what they’re building toward. One assumes the current architecture is the future. The other is trying to prove there’s a cheaper path to the same destination.

We’ll find out soon enough which bet pays off. The infrastructure layer of AI is still being figured out in real time, and there’s enough capital flowing into both approaches that we’re going to get a real test of which model works at scale.