Anthropic buys computing capacity from Grok's owner

Anthropic competes with Grok for customers while buying computing capacity from the company behind it. That relationship is a useful antidote to the idea that AI adoption must produce one winner. A company can lose a subscription sale and still supply the machines that help a rival serve its customer.

The week 40 podcast covers Hammer's provider research from September 28 to October 4. Its central argument is that AI competition and cooperation can grow together. If that holds, businesses gain more ways to buy useful AI, and providers have more ways to earn revenue than winning the chat window.

The September disclosure adds scale to a May agreement

On May 6, SpaceXAI announced an agreement giving Anthropic access to Colossus 1. Anthropic planned to use that additional computing capacity for Claude Pro and Max subscribers. Computing capacity here means the machines and processing time needed to train or run AI.

Source: SpaceXAI's May 6 Anthropic partnership announcement.

The new development in last week's research is the reported scale. Reuters wrote on September 29 that Anthropic's agreements with xAI could lead to up to $84.5 billion in spending through 2029, largely cancelable with 90 days' notice. Reuters based its account on a confidential prospectus it had seen. Hammer has verified the published reporting; we have not inspected that prospectus.

The qualification matters commercially. That figure describes potential spending under agreements, with substantial cancellation flexibility. Revenue depends on what happens under those agreements. It also describes a supplier relationship between two competitors, which survives the caveats about the final amount.

Source: Reuters reporting on Anthropic's infrastructure commitments, via KSL.

Grok reaches customers through Amazon's catalog

A separate September 28 development concerns distribution. AWS added Grok 4.7 to Amazon Bedrock, its service for accessing AI models. The model itself had launched on September 21; last week's change was an additional route to customers.

For a business already using AWS, this makes Grok a candidate within an existing technology environment. It gives xAI a route to buyers who might prefer that environment to a direct vendor integration. Model development and customer distribution can belong to different companies.

The route has its own terms. AWS describes cross-region inference profiles, so the location of an endpoint alone does not establish where every request runs. The commercial opportunity is another purchasing option; each account still needs applicable access and suitable terms.

Source: AWS: Grok 4.7 is now available on Amazon Bedrock.

Cooperation gives rivals another way to grow

Consider the two transactions together. Anthropic buys computing capacity from the company behind Grok. Grok gains distribution through Amazon. Our interpretation is that a provider can benefit from activity outside its own branded assistant. Selling infrastructure or reaching customers through a partner can support a business while competition between the products continues.

For an organization choosing AI, that makes the search for a permanent industry winner less useful. Start with a recurring piece of work and the commercial routes available to you. A good product can fit alongside another provider's service. Buying it does not require a prediction that every competitor will disappear.

There is a strong counterargument: better models can still pull customers toward one product, and a few large infrastructure suppliers can gain bargaining power. Partnerships can widen access while increasing dependence on those suppliers. The evidence supports commercial interdependence; it does not establish cheaper prices, shared customer data or common outage exposure.

The concrete next step is a purchasing conversation: ask your current technology supplier which relevant models it can offer through your existing agreement. Then compare one real task with your current approach. That is a business question worth bringing to Hammer's Mindset Forge work.

A week of product changes alongside the deals

The English audio draws on research covering all five active providers. OpenAI's September 29 DevDay announcements cover delegated work and collaborative workspaces. Anthropic's research includes Sonnet 5.5, Claude Code Mods and changes to Cowork. Google announced Gemini 4 Argon with restricted initial access; wider availability remains a separate question. Grok's research also covers Team Bots in public beta and its managed Intune app.

Mistral supplies a different rhythm. Its September 28 report recovers earlier September changes to Vibe, spreadsheets and mini apps. Those are dated background in this research batch, rather than fresh October launches. The episode brings these different developments together without treating every provider as if it made the same move.

This is an AI-generated masterclass built from Hammer's deep daily research into AI-provider updates and features, processed with NotebookLM. The research window is September 28–October 4, 2026. This companion article uses the source research and a NotebookLM synthesis; it is not a transcript.