- Perplexity has struck a $750 million, multi-year cloud infrastructure deal with Microsoft to run advanced AI models via Azure.
- The pact demonstrates how AI developers are focusing on long‑term compute commitments and diversified cloud partnerships amid rising infrastructure pressures.
What happened
AI startup Perplexity has agreed to a $750 million, three‑year deal with Microsoft to use its Azure cloud services to operate a suite of advanced AI models, according to persons familiar with the matter. Under the arrangement—part of Microsoft’s Foundry program—Perplexity will deploy models from major developers including OpenAI, Anthropic, and xAI on Azure’s infrastructure.
Although Perplexity confirmed the partnership includes access to “frontier models from X, OpenAI, and Anthropic,” it also said this arrangement does not replace its existing spending with Amazon Web Services (AWS), which remains its primary cloud provider.
This cloud deal comes amid broader tensions in Perplexity’s infrastructure relationships: in 2025 Amazon sued the startup over its “agentic” shopping feature, alleging the AI tool accessed customer accounts and mimicked human browsing behavior—a controversy that underscores growing legal and operational challenges in agent‑based AI services.
Also Read: Microsoft CEO sounds a note of caution on AI
Also Read: Microsoft 365 services disrupted users
Why it’s important
The Perplexity–Microsoft agreement highlights a broader development in the AI landscape: competition increasingly centers on access to and long‑term commitments for compute and cloud resources, not solely on model performance or feature innovation. Infrastructure is costly, complex, and crucial to scaling deployments reliably, making strategic cloud partnerships a pivotal element of business planning for AI firms.
Locking in multi‑year cloud arrangements allows startups to negotiate capacity, pricing, and specialized tooling—but it also raises questions about dependency on specific providers and the risks that accompany such concentration. While diversifying compute sources can provide redundancy, it may also limit flexibility if strategic priorities or cost structures change over time.
Perplexity’s deal further reflects how the industry’s compute ecosystem is consolidating around a few major clouds and infrastructure programs. This has implications for competition: smaller players without access to deep cloud discounts or long‑term commitments might struggle to compete on speed, reliability, and cost efficiency.
There are also broader questions about balance in AI resource allocation: as development shifts from research experiments to large‑scale production, the economics of cloud spending could influence which types of models and applications are commercially viable. Firms that can secure favorable compute terms may gain an edge, potentially reshaping innovation incentives and market dynamics.
In this context, infrastructure strategy—whether through partnerships, diversified cloud usage, or on‑premises resources—is becoming as central to AI success as algorithm design itself.
