AI competition is increasingly about two linked questions: who can secure enough computing capacity, and who can turn that capacity into useful products? Anthropic’s $11.6 billion agreement with Akamai Technologies addresses the first. Microsoft’s decision to bring its consumer and workplace Copilot tools together addresses the second. Neither move makes model quality irrelevant, but both suggest that a strong model alone is not enough.
Anthropic commits to capacity years ahead
Anthropic has agreed to buy computing power from Akamai Technologies under a seven-year contract valued at $11.6 billion. Akamai plans $5.5 billion in capital expenditures tied to the agreement, with the computing arrangement targeted to begin operating in the second half of 2027. The contract is the largest in Akamai’s history and marks a significant expansion of its work in specialized AI computing infrastructure.
The timing matters. Anthropic has reported that demand for computing resources for Claude grew eightyfold year over year. As more people and organizations use an AI model, its developer needs capacity for inference—the computing work required to produce answers to requests. Securing that capacity well before it is needed may help Anthropic serve demand that its existing resources cannot cover.
▲ Hardware behind AI computing demand
The agreement also gives Anthropic stock warrants, which provide an option to acquire an equity stake in Akamai. That feature links the customer and supplier beyond a conventional purchase of computing services. OpenAI has used a warrant-linked structure in a computing agreement with AMD. Such arrangements raise a question worth watching: when an AI developer can also benefit from a supplier’s equity, how clearly do contract values reflect independent demand? The structure itself does not answer that question.
Other developments point to the same pressure to build capacity. Elon Musk said xAI’s Colossus 2 facility in Memphis has deployed 550,000 NVIDIA GB200 and GB300 graphics processing units, or GPUs, the specialized chips used for AI workloads. He said xAI aims to double that number by year-end. Separately, an AI data center provider raised $3.36 billion ahead of a planned initial public offering, including a $1 billion commitment from NVIDIA. Those figures describe commitments and plans, not a guarantee that every facility will deliver capacity on schedule or earn a return.
Microsoft narrows Copilot’s job
Microsoft is taking a different kind of strategic step. It is moving away from a race to build a universal personal digital companion and combining its consumer and workplace versions of Copilot into a unified software environment. Its renewed focus is on practical tasks in productivity applications such as Excel and PowerPoint.
▲ Copilot’s focus on workplace tasks
The shift centers on what Microsoft appears to want Copilot to do: tie AI assistance to work that people already perform in business software. Competition from Meta’s consumer-facing Muse assistant adds context to the decision. Meta’s shares gained roughly 30% during September as enthusiasm for Muse rose, though its longer-term revenue potential remains to be demonstrated.
Microsoft’s choice may make the distinction between consumer and workplace AI less important inside Copilot. For a user, the test will be whether that combined approach makes a task easier to complete, not whether the assistant can sustain a general conversation. For a business, it will also matter whether the software handles access to company data appropriately. That is an interpretation of where the product strategy could lead, not a promised result.
The test moves from models to delivery
The two strategies operate at different points in the same chain. Anthropic needs computing resources to keep Claude available as demand grows. Microsoft needs Copilot to make those resources valuable in everyday tasks. Spending on infrastructure does not, by itself, establish that an AI product will be useful; a useful product still needs reliable capacity behind it.
For organizations planning AI deployments, the practical questions follow from that distinction:
- Check capacity against timing. A large contract is not the same as computing power available today. Anthropic’s Akamai arrangement has an operational target in the second half of 2027.
- Judge tools by complete tasks. In workplace software, look at whether an assistant helps finish a workflow, rather than only producing a plausible answer.
- Examine control of data and actions. Permissions, data security and clear limits on what an AI system may do matter when it works inside business processes.
What to watch next
Anthropic’s deal shows how far ahead an AI developer may commit to secure computing power. Microsoft’s Copilot shift shows the other side of the challenge: turning AI into dependable work within familiar applications. Readers assessing these moves should track when promised capacity becomes available and whether unified AI tools improve real tasks while respecting organizational controls. Those outcomes will say more than contract size or product positioning alone.