Anthropic clinches massive infrastructure pacts — $200B Google TPU commitment and $1.8B Akamai CDN deal
- Anthropic locked in two huge infrastructure deals in early May — a reported $200 billion Google Cloud pact and a $1.8 billion Akamai contract. - The Google agreement reportedly runs five years and leans on TPU capacity, while Akamai’s seven-year deal sent its stock up roughly 25%. - This matters because frontier AI demand is spreading past Nvidia-heavy clouds into TPUs and edge networks built for inference.
AI labs are starting to look less like software companies and more like industrial buyers of power, chips, and network capacity. That is the real story behind Anthropic’s latest infrastructure spree. In the span of a few days, reports tied the company to a roughly $200 billion, five-year commitment with Google Cloud and a separate $1.8 billion, seven-year compute deal with Akamai. Put simply — Anthropic is locking down places to run Claude wherever demand shows up. ### Why are there two different deals? They solve different bottlenecks. The Google arrangement is about core cloud capacity and custom AI chips — especially TPUs — at an enormous scale. The Akamai deal is smaller, but it points at a different problem: getting inference closer to users, with more distributed infrastructure instead of relying only on the biggest centralized clouds. (usnews.com) ### What is the Google piece, exactly? The reported commitment is about $200 billion over five years, tied to cloud services and chips from Google. That is an eye-watering number even by AI standards. Reuters’ pickup of The Information said the deal could amount to more than 40% of Google Cloud’s disclosed revenue backlog, which tells you this is not a side arrangement — it is big enough to matter to Google’s financial plumbing. (usnews.com) ### Why do TPUs matter here? Because this is one of the clearest signs yet that frontier-model demand is no longer just a GPU story. Anthropic already said in April that it trains and runs Claude across AWS Trainium, Google TPUs, and Nvidia GPUs, and that Google plus Broadcom were helping expand TPU capacity by multiple gigawatts. Basically, Anthropic wants optionality — different chips for different workloads, better pricing leverage, and less dependence on any single supply chain. (usnews.com) ### So what is Akamai doing in this picture? Akamai is the surprising part because most people still associate it with CDN services, not frontier AI. But the company has been building out cloud infrastructure, and it disclosed a seven-year deal with a “leading frontier model provider.” Bloomberg identified that customer as Anthropic. Forbes called it the largest deal in Akamai’s history, which helps explain why investors treated it like a category shift rather than just a nice contract win. (anthropic.com) ### Why would Anthropic want Akamai at all? Inference is the likely answer. Training likes giant clusters. Serving models to real users is messier — lots of requests, lots of geographies, lots of latency sensitivity. A distributed network operator can help with that. Think of it like this: Google is helping secure the power plant, while Akamai may help extend the grid to more neighborhoods. That is an inference about the fit, but it matches how Anthropic has described using multiple hardware and cloud platforms for resilience and workload matching. (bloomberg.com) ### Did the market care? Very much. Akamai shares jumped about 25% to 27% after the company disclosed the long-term AI cloud agreement. Analysts called it a landmark win because seven-year AI infrastructure contracts are unusual, and because it suggests Akamai can sell more than legacy delivery services into the AI buildout. (anthropic.com) ### What changed versus a year ago? A year ago, the clean mental model was simple: frontier AI ran on hyperscalers and mostly on Nvidia. That model is breaking. Anthropic’s recent moves suggest the next phase is multi-chip, multi-cloud, and increasingly edge-aware. The bottleneck is no longer just training the biggest model — it is serving enough tokens, cheaply enough, in enough places, without running out of infrastructure. (seekingalpha.com) ### Bottom line? These deals look like capacity reservations, but they are really a map of where AI is heading. Anthropic is not just buying more compute. It is spreading Claude across TPUs, clouds, and edge-style infrastructure so demand spikes do not trap the business in one vendor’s queue. (usnews.com) (anthropic.com)
