Akamai's $11.6B CPU Bet With Anthropic Rewrites the AI Infrastructure Playbook

The Number That Isn’t the Point

Akamai’s stock surged more than 20 percent in after-hours trading on September 24, 2026. The trigger was a seven-year contract with Anthropic worth $11.6 billion, potentially expandable by another $9 billion. That headline number alone would be remarkable for a company best known as the internet’s plumbing — but the real story sits in a single word buried in the filing: CPUs.

Anthropic is committing $11.6 billion to general-purpose processors, not the graphics accelerators that dominate every AI infrastructure narrative. Akamai is front-loading $5.5 billion of its own capital to build out the capacity. Revenue from the deal doesn’t begin until late 2027. This is not a quick return. It is a structural bet, and it changes how you should think about who wins the next phase of the AI economy.

The CPU Question No One Is Asking

Every AI infrastructure story these days centers on Nvidia and its H100 and H200 GPUs. The assumption is that more compute power equals more GPUs, and the companies controlling GPU supply control the bottleneck. That assumption is wrong.

Akamai’s deal with Anthropic signals something different: a meaningful shift toward CPU-intensive workloads in the AI stack. CPUs handle inference, data preprocessing, orchestration, and a growing range of model operations that don’t require the massive parallelism GPUs excel at. As AI systems move from training to deployment — and deployment scales exponentially — the CPU layer becomes just as constrained as the GPU layer. Anthropic’s commitment tells us they see that coming.

This isn’t speculation. Akamai’s own press release explicitly stated the contract addresses “rapidly growing CPU processing demand” at Anthropic. The company also raised its 2026 supply chain and memory spending plan by $1.7 billion to secure components. Memory availability, incidentally, is another bottleneck distinct from chip supply — and one that benefits CPU architectures differently than GPU ones.

Who Actually Wins

Let’s trace the value.

Akamai wins first. The company is transitioning from a legacy CDN business into an AI infrastructure provider, and this contract is the proof point. The stock market already rewarded the pivot with a 20 percent jump. But the financial structure is aggressive: Akamai spends $5.5 billion upfront, sees zero revenue in 2026 from the deal, records $150 million to $300 million in 2027, and reaches approximately $1.7 billion in annual revenue only by late 2028 when full capacity comes online. That is a three-year cash burn before the payoff. The risk is real.

Akamai’s CFO Ed McGowan confirmed no revenue recognition in 2026. Akamai also granted Anthropic warrants to purchase up to roughly 5 percent of Akamai’s shares at $111.33 per share, with about 2 percent expected to vest immediately under the base contract and the remainder tied to any expansion. SiliconANGLE reported this is the first time Akamai has issued warrants in a similar agreement. That’s a concession — Akamai is aligning Anthropic’s incentives with its own stock price to secure long-term commitment. Smart, but costly if the deal stalls.

Anthropic wins second. They’ve locked in seven years of CPU capacity at a fixed price, insulating themselves from the kind of hardware scarcity that has plagued every major AI lab. While OpenAI and Google are negotiating GPU allocations through their own cloud partnerships, Anthropic is securing a different resource layer that competitors may underestimate. The fact that this deal is six times larger than the $1.8 billion contract Akamai disclosed in May 2026 — initially for an unnamed client, later identified as Anthropic — shows rapid escalation in their infrastructure strategy.

Nvidia loses relative to expectations. Not absolutely — demand for GPUs remains fierce. But if Anthropic and others begin allocating comparable spend toward CPU infrastructure, the total addressable market for AI chips fractures along different lines. The narrative that GPU supremacy guarantees winner-take-all economics weakens.

The Timing Signal

Revenue starting in late 2027 is deliberate. Akamai needs time to build, provision, and commission the infrastructure. Full capacity lands by late 2028. That timeline coincides with when the next generation of large language models will likely require dramatically more compute for inference at scale — the phase where CPU utilization spikes. Akamai is not just building for today’s Anthropic. It’s building for theAnthropic that will exist three years from now.

The $5.5 billion capital commitment is substantial but not unprecedented for Akamai. What’s notable is the composition: $1.7 billion of additional 2026 spending going toward supply chain and memory procurement. Memory is the quiet bottleneck. High-bandwidth memory availability constrains both GPU and CPU systems, but CPU-based inference workloads are particularly memory-sensitive. By front-loading memory procurement, Akamai is hedging against a constraint that could delay the entire project.

What Happens Next

Watch two things.

First, whether other AI labs pursue similar CPU-heavy contracts. Akamai’s disclosure of a prior unnamed client in May 2026 suggests this is not a one-off. If multiple models shift their compute mix toward CPUs for inference, the secondary market for server-grade processors will tighten alongside GPU markets. AMD and Intel — both players in this space — could find themselves in a suddenly strategic position.

Second, watch Akamai’s utilization rates once the infrastructure goes live. $1.7 billion in annual revenue at full capacity sounds healthy until you compare it to the $5.5 billion spent and the seven-year lockup. Margins matter. If Akamai can fill excess capacity with other clients — and the May 2026 contract proves there is demand — the unit economics improve dramatically. If not, the deal becomes a drag for years.

The warrants add another variable. If Akamai’s stock price rises significantly, Anthropic’s warrant holdings become valuable, creating a financial feedback loop between the two companies. If the stock stagnates, the warrants are underwater and the incentive alignment evaporates. Either way, Akamai has tied its fate to Anthropic’s trajectory more deeply than a standard cloud contract would.

The Bigger Picture

Japanese business media covered this deal with more structural curiosity than many English-language desks. Forbes Japan led with the CPU angle — the detail that matters most. The English-language AI infrastructure narrative remains fixated on GPU supply chains, Nvidia margins, and hyperscaler capex. The CPU shift is happening in parallel, quietly, in contracts like this one.

Akamai is no longer just the company that keeps your website from crashing. It is building the compute layer for the next wave of AI deployment. The $11.6 billion contract is the announcement. The real test is whether the infrastructure delivers returns before the money runs out.