IA · 27 September 2026 · 5 min read
Anthropic Bets $11.6 Billion on Akamai: AI Infrastructure Pivots Toward CPUs
In brief: Anthropic has committed $11.6 billion over seven years to Akamai's distributed cloud infrastructure. The landmark partnership highlights a major shift in artificial intelligence infrastructure: alongside massive GPU clusters dedicated to model training, demand is surging for general-purpose CPUs and edge compute capable of running sandboxes, code, and live tools for autonomous AI agents.
by Team Mocchi's
Beyond GPUs: A Record-Breaking Cloud Deal for the Edge
In the global race for artificial intelligence infrastructure, tech industry attention has long been captured by next-generation GPUs and massive energy-hungry data centers. However, as frontier models evolve into autonomous agents capable of performing multi-step tasks, computing priorities are undergoing a structural shift. Anthropic confirmed this transition by striking a massive cloud computing commitment with Akamai, agreeing to spend $11.6 billion over seven years on the company's distributed cloud platform.
According to reporting by TechCrunch, the commitment is more than six times larger than an initial $1.8 billion estimate cited in market reports earlier this year, making it the largest contract in Akamai’s corporate history. Securities filings detail that the contract is not unconditional: spending remains subject to Akamai meeting performance milestones and infrastructure availability targets, with termination clauses available to both parties under specific operating conditions.
The transaction also carries a strategic equity component. Akamai issued Anthropic warrants for nonvoting preferred stock convertible into 7.7 million common shares at an exercise price of $111.33 per share. The arrangement allows the AI lab to acquire up to roughly 5% of Akamai's outstanding stock, with approximately 2% vesting as soon as Anthropic issues its first contract payment.
The General-Purpose Shift: Why AI Agents Need Millions of CPUs
The architectural driver behind the multi-billion-dollar deal lies in the changing profile of AI workloads. While initial foundation training and deep inference remain tightly coupled to dedicated GPU and TPU clusters hosted by traditional hyperscalers, agentic workflows demand a fundamentally different type of machine.
When models such as Claude Code orchestrate automated tasks—such as executing automated unit tests, interacting with Git repositories, scraping dynamic web interfaces, and running background code within sandboxed Linux virtual machines—the primary system bottleneck moves to the CPU. General-purpose processors are indispensable for managing operating system virtualization, secure sandboxing, network I/O, and file system orchestration.
Akamai's global edge footprint, strengthened through its cloud compute expansion, provides a geographically distributed network optimized for low latency. This setup allows Anthropic to instantiate isolated execution environments closer to users and enterprise systems, avoiding the network lag and architectural strain of routing every small operational command back to centralized GPU facilities.
Capital Spending Plans and Revenue Timelines
To accommodate Anthropic’s massive scale, Akamai is significantly expanding its hardware base. During a call with financial analysts, executive leadership revealed plans to allocate approximately $5.5 billion in capital expenditures to build out the required capacity. In addition, the firm is deploying $1.7 billion in immediate 2026 capital investments to pre-purchase critical hardware inventory, including memory modules and network gear.
The commercial payoff will scale over time. Akamai does not anticipate immediate revenue from the arrangement throughout 2026. Financial projections indicate top-line contributions between $150 million and $300 million starting in the second half of 2027, before accelerating to an annualized run rate of approximately $1.7 billion by the end of 2028.
The arrangement also underscores a broader shift across the cloud landscape: AI labs are actively diversifying away from exclusive single-hyperscaler ecosystems. Rather than relying solely on monolithic partners, frontier AI providers are building hybrid stacks, pairing specialized accelerators for tensor calculations with edge-native compute platforms tailored for real-time agentic execution.
Mocchi's take
As custom software architects integrating autonomous models into business systems, we see this deal as clear proof that AI is evolving from a conversational interface into an operational runtime environment. Enterprises deploying AI agents can no longer evaluate projects solely on model token pricing; they must design secure, low-latency execution sandboxes where generated code can run safely. Understanding this structural separation between the semantic reasoning layer (GPUs) and the operational execution layer (distributed CPUs) is becoming a prerequisite for building reliable, scalable software architectures.