Anthropic has agreed to spend $11.6 billion over seven years on Akamai's cloud infrastructure, Akamai announced Thursday — the largest contract in the company's history and more than six times the size of a $1.8 billion deal Bloomberg reported between the two companies in May. Investors loved it: Akamai shares surged more than 20% on the news. But the headline number isn't the most interesting thing about this deal. The hardware is. Anthropic isn't buying GPUs, the chips everyone associates with artificial intelligence. It's buying CPUs — the general-purpose processors that run code, serve web traffic, and power the unglamorous middle of the internet.

That detail says a lot about where the frontier labs think compute demand is heading. Training a model takes GPUs. Serving an army of AI agents that browse, click, execute code, and retry — that takes an awful lot of CPUs.

The deal, in numbers#

The basics, per Akamai's announcement and its securities filing:

  • $11.6 billion committed over seven years, announced on September 24.
  • Zero revenue this year. Akamai expects $150–300 million in 2027, starting in the second half, ramping to an annual pace of about $1.7 billion by the end of 2028.
  • $5.5 billion in capex for Akamai to build the capacity, including an extra $1.7 billion this year to buy components such as memory in advance.
  • A stock sweetener. Anthropic gets warrants for nonvoting preferred stock convertible into 7.7 million common shares — roughly 5% of Akamai — at $111.33 a share. About 2% vests once Anthropic makes its first payment; each additional $3 billion it commits unlocks roughly another percentage point.
  • Room to grow. The two companies have an option to expand the agreement by another $9 billion, which would take the potential total past $20 billion. JPMorgan analysts called the expansion "a path of low resistance."

The deal also builds on $2.8 billion in other multi-year cloud commitments Akamai has already secured from customers this year, disclosed in its second-quarter earnings.

Illustration of a rising growth chart over cloud computing nodes and data stacks
AI-generated illustration — no free public image of this subject was available.

The fine print: a commitment, not cash#

Before treating $11.6 billion as money in the bank, read the conditions. According to Akamai's filing, Anthropic's payments depend on Akamai meeting delivery and service-availability requirements, and either company can walk away under certain conditions. This is a commitment with gates, not an unconditional purchase order.

The build-out itself is a bet on Akamai's execution. Spending $5.5 billion of capex ahead of an eventual $1.7 billion annual revenue pace is aggressive, and the company may need fresh capital: CFO Edward McGowan told analysts Akamai holds $4.6 billion in cash plus a $1 billion credit facility — and pointedly didn't deny the possibility of a raise next year.

Why CPUs are the interesting part#

Every AI infrastructure headline this year has been about GPUs — who has them, who can't get them, what they cost. This deal is pointedly about the other chip. Akamai didn't specify what Anthropic will run on the CPUs, but the timing lines up with a broader shift: as AI agents take on more of the work, demand for general-purpose compute has been growing alongside the GPU frenzy. Agents don't just generate text; they run code, browse the web, call tools, and retry — workloads that look a lot more like traditional cloud computing than model training.

There's also a geography angle. Akamai Cloud runs across thousands of distributed locations worldwide. Agents and inference workloads benefit from being close to users, while training clusters can sit wherever power is cheapest. A distributed CPU footprint may be better matched to serving agents than a single giant GPU campus.

Akamai co-founder and CEO Tom Leighton said Anthropic chose the company after assessing its ability to build and operate large-scale AI infrastructure — a sign that the old CDN business has quietly become a credible cloud contender.

Illustration of a data center corridor lined with server racks
AI-generated illustration — no free public image of this subject was available.

One more stop on Anthropic's compute shopping spree#

The Akamai deal is the latest in a string of enormous infrastructure commitments from Anthropic this year. The lab agreed to spend $45 billion leasing AI computing capacity from Nscale at a data center campus in West Virginia, signed a 20-year lease with TeraWulf for roughly 401 megawatts at the Hosville data center in Kentucky, and inked a memory and storage supply contract with Micron in June.

The pattern is unmistakable: compute is the binding constraint on frontier AI, and the labs are buying power, silicon, and memory years in advance. Infrastructure commitments that used to be measured in quarters are now measured in decades.

What to watch#

Can Akamai build it? The delivery gates in the filing are real, and a $5.5 billion capex program with a possible capital raise is a lot of execution risk for a company pivoting from CDN to cloud.

Does the expansion trigger? If Anthropic exercises the $9 billion option, the deal becomes a $20 billion relationship — and Akamai's cloud becomes a structurally important part of frontier AI infrastructure.

What actually runs on the CPUs? If this capacity ends up serving agentic inference at scale, it's a leading indicator: the AI infrastructure boom is moving beyond training GPUs toward the everyday compute that agents consume. Watch who else starts buying CPUs at this scale.

Sources#