AI Market: Digest for July 29, 2026

The AI market is undergoing a structural shift: euphoria is being replaced by infrastructural pragmatism. This week's big money is in compute, cost control, and corporate security.

Compute Contracts as the New Venture

Safe Superintelligence (SSI), Ilya Sutskever's company, announced a long-term partnership with Nvidia to scale its research. Simultaneously, Recursive Superintelligence signed a $410 million deal with Amazon—an amount that consumes the lion's share of the company's raised funds.

In my view, we are seeing the emergence of a new monetization pattern. Startups aiming to create frontier models are no longer just competing for classic funding rounds in exchange for equity. They are purchasing compute capacity directly. This turns cloud providers and chipmakers into the primary beneficiaries of the AI race while minimizing their own risks.

AI Gateways: The Fight for Control and Savings

Satya Nadella stated that companies trusting a single AI model for everything may not survive. He emphasized the critical need for proprietary models or an AI gateway layer that decouples requests from any specific model. Following this, Snowflake introduced Cortex AI Gateway—a centralized layer for managing how AI agents access corporate data. Microsoft also launched its own specialized cybersecurity model and an agentic defense platform, betting on reducing the cost of corporate security.

This cluster of news reflects a clear trend: the enterprise market is tired of uncontrolled AI spending. Businesses need request routers that dynamically select the cheapest model capable of solving the task. Infrastructure for AI management is becoming a more marginal business than the base models themselves.

Agents Prove ROI in the Enterprise

Large businesses are reporting the first measurable results from AI agent deployment. Instacart's CTO stated that AI has allowed the company to stop worrying about technical debt by delegating routine engineering work to machines. At General Motors (GM), autopilot engineers now spend only 15% of their time writing code, and the use of agents has tripled the number of successfully merged pull requests. OpenAI also released a report confirming that coding agents are accelerating the build of scientific software.

Previously, AI agents were pilot projects for board reports. Now, they are tools for aggressive operational expense optimization. Businesses have stopped asking "what is AI capable of?" and started asking "how much money will this save?".

Google Changes the Rules of Traffic and Monetization

Google's AI Overviews now appear in 43% of US search queries according to Similarweb—up from 15% a year ago. Meanwhile, Pew Research Center data shows a collapse in clicks: if Google displays an AI answer, users click on traditional links in only 8% of cases (half as often as before).

This represents a fundamental break in the old model of content monetization on the web. Traffic as the foundation of the media business is rapidly depreciating. In my opinion, over the next year or two, only those publishers and platforms that can sell direct access to their databases for AI training—rather than relying on search engine referrals—will survive.

Bottom Line

Infrastructure, cost control, and agentic automation are the three pillars currently supporting AI monetization. In the coming quarters, expect market consolidation: corporations will continue to acquire orchestration tools (AI gateways and sandbox solutions), and cloud compute providers will solidify their status as the primary winners of the current technological cycle.