AI Market: Digest for July 24, 2026

The AI market is entering a phase of aggressive commercialization. While leaders like OpenAI and Anthropic are showing astronomical revenue growth and attracting tens of billions of dollars, a fierce battle is emerging at the corporate implementation level to reduce inference costs and manage new security risks.

Capitalization and Revenue Records

OpenAI has announced an unprecedented infrastructure spending program, planning to spend $750 billion by 2030. Investments comparable to the GDP of a mid-sized European country underscore a bet on total monopoly over compute power.

Meanwhile, the primary competitor in the foundation model market—Anthropic—reached an estimated annual revenue run rate of $47 billion by May 2026 (up from $9 billion a year prior). Matt Murphy, an investor at Menlo Ventures, notes that he hasn't seen this pace of growth in his 25-year career. In my view, Anthropic's success is driven less by the quality of the model itself and more by a smart focus on the enterprise segment and security, allowing corporations to pay a premium.

The Infrastructure Race and GPU Alternatives

The infrastructure market continues to expand. NVIDIA is actively scaling production of Vera Rubin systems, covering more than 350 factories across 30 countries. Simultaneously, the company is strengthening its position in the public sector (launching a DGX supercomputer at the US Naval Postgraduate School) and developing the concept of physical AI for robotics.

New players are also emerging: Etched, a startup founded by three Harvard dropouts, hit a $10.3 billion valuation, promising to radically accelerate LLM inference without using standard GPUs. AMD is also entering the fray, introducing the Helios rack-scale system to compete directly with NVIDIA.

Optimizing Token Costs

Euphoria is being replaced by pragmatism. Compute costs are becoming the primary barrier to scaling AI agents. Google responded by releasing Gemini 3.6 Flash and Flash-Lite, specifically targeting lower latency and token costs for enterprise developers.

Microsoft is demonstrating a similar approach: the company's new internal models (MAI-Image and MAI-Voice) reduce operational costs by up to 89% compared to OpenAI solutions. The Writer platform went even further, introducing an architectural solution that cuts token consumption by nearly 40% without sacrificing accuracy. Businesses no longer want to overpay for redundant capacity.

AI Agent Security as a New Market

The mass adoption of autonomous agents has led to a systemic security crisis. A Cisco study showed that multi-turn attacks break flagship models in 88.3% of cases, while standard tests fail to detect this.

The problem is exacerbated by the human factor and architectural errors. The Hugging Face breach, for which OpenAI took responsibility (due to a sandbox configuration error), highlighted the vulnerability of supply chains. According to VentureBeat, 54% of enterprises have already experienced AI agent security incidents, but only a third use strict credential isolation. This is creating a new market: the startup Glow emerged from stealth with a $1.2 billion valuation, offering endpoint protection solutions for the AI era.

Geopolitics and Business Restructuring

Washington is increasing pressure on Chinese open-weight models. Following allegations that Moonshot AI (creator of Kimi K3) performed unauthorized distillation of Anthropic's Fable model, the US Treasury threatened sanctions. This is forcing corporations to reconsider the risks of using cheap Chinese models in critical infrastructure.

Parallelly, business models are shifting. Monday.com laid off 20% of its staff (about 630 people) to pivot fully toward an AI platform. Jack Dorsey is launching Buzz—a collaboration chat for humans and AI agents—positioning it as a challenge to Slack. Runway introduced Media Router, a tool for automatically selecting the optimal media generation model based on a balance of price, speed, and quality, signaling the start of consolidation in the generative tools market.

Bottom Line

The era of mindless consumption of cheap AI tokens is ending. In the coming quarters, the winners in the market will not be the companies training the largest models, but those who can offer infrastructure with the lowest inference cost, ensure reliable AI agent isolation, and protect intellectual property from geopolitical risks.