Akamai is handing Anthropic 5% of itself to lock in cloud orders
Akamai is giving Anthropic a warrant for up to 5% of its own stock — roughly 7.7 million shares at an exercise price of $111.33 — in exchange for a pledge to spend $11.6 billion on Akamai Cloud over seven years. A supplier paying a customer in equity to guarantee purchase volume is not a normal contract. It is what the infrastructure trade looks like when the buyer holds the leverage.
The warrant vests against what Anthropic actually buys: about 2% of Akamai's common stock with the $11.6 billion commitment, then roughly 1% for every additional $3 billion of cloud purchases, up to another $9 billion and a potential $20 billion relationship. It carries no voting rights and settles in cash, so it is a claim on value rather than a seat at the table. Akamai's shares rose about 16% in extended trading, and the company reaffirmed rather than raised its 2026 revenue guidance.
Akamai says it will spend roughly $5.5 billion in capital expenditure to serve the deal, including about $1.7 billion more in 2026 to pre-buy components such as memory. The commitment is for CPU workloads on Akamai Cloud — the distributed edge network Akamai has spent years rebuilding into a compute business. That is the part worth watching: this is not a GPU contract, and it lands on the same customer relationship that produced Akamai's previous record, a $1.8 billion seven-year agreement with Anthropic signed in May that moved the stock nearly 27% in a single day. This one is more than six times larger and carries an ownership option. Akamai calls it a "significantly expanded relationship" but has not said whether the $11.6 billion absorbs the May commitment or sits on top of it.
Our read: an average of about $1.66 billion a year is large next to Akamai's $1.1 billion of quarterly revenue in Q2 2026 — concentration that cuts both ways, since the warrant only fully vests if Anthropic keeps buying. We covered the company's broader compute-signing spree in September — Anthropic has signed for 14.8 GW of compute. This one is different in kind: it is the first time a major supplier has put its own equity behind a purchase commitment.
Shanghai's banking regulator has formally opened the door to customer-facing generative AI in Chinese finance — under supervision. The Shanghai bureau of the National Financial Regulatory Administration published 16 measures on September 24 (dated September 18) implementing the national framework for AI in banking and insurance. The provisions that matter: any generative model used in a public-facing or high-risk scenario must be reported to the bureau before it goes live, must clear an access-control review approved by the institution's own risk committee, and must be filed or registered with China's cyberspace authorities — whether it is self-developed, fine-tuned, privately deployed or called through an API.
The document also tells institutions to build "safe and trusted" infrastructure on domestically acceptable chips, software and compute, to prepare human-takeover fallbacks by decoupling models from the applications above them, and to adopt a "general model as backstop, industry model for delivery" procurement strategy. There are no adoption targets and no model names anywhere in the text — it is a framework, not a scorecard. The permission to run direct-to-customer large-model applications "in a controlled environment" is the genuinely new part, and it is what Chinese financial media led with.
A Vancouver startup with a tiny public profile is sitting in the top ten of an independent streaming speech-recognition leaderboard, above Google, Microsoft, Nvidia and Amazon. Alebex's speech-to-text model posted a 3.96% word error rate on Artificial Analysis's AA-WER Streaming Index, published September 21 — 8th of 29 models at that snapshot, ahead of Gemini 3.5 Transcribe Live at 4.00%, Azure at 5.25%, Nvidia at 5.36% and Amazon at 5.79%. On the live board today, with 37 models listed, Alebex sits 10th.
Two caveats worth stating plainly: the ranking comes from a company press release, and the only independent verification is the leaderboard itself, which we checked. A top-ten slot on one English-heavy streaming index is visibility, not a moat — but for a lab this small, the leaderboard is how anyone finds out it exists.
What to watch: whether Akamai's next filing shows the May commitment still standing alongside the new one, and whether Anthropic keeps signing CPU capacity while its rivals buy GPUs.
If you were buying cloud capacity at this scale, would you want equity in your supplier — or is that a warning sign about the supplier?
