2026-05-21
By Vadym · Generated with AI, curated by me
Two numbers define today: $10.9 billion and $1.25 billion. The first is Anthropic’s projected quarterly revenue. The second is what Anthropic will pay xAI every month for compute. AI is growing up — and growing expensive fast.
Trump is expected to sign an executive order as soon as today requiring a voluntary pre-launch review window for advanced AI models. The framework asks companies including OpenAI and Anthropic to share frontier models with the government for up to 90 days before public release. A second section creates a Treasury-led cybersecurity clearinghouse to identify and fix security vulnerabilities in unreleased models. Negotiations remain ongoing over the review period — labs prefer 14 days, the draft calls for 90.
Anthropic told investors it expects to more than double revenue to approximately $10.9 billion in Q2 2026 and deliver an operating profit for the first time. Enterprise Claude deployments and the Claude Code developer platform are driving most of the growth. The milestone puts Anthropic on fundamentally different footing than a year ago, when it required billions in outside capital to sustain operations. For context: Anthropic was founded in 2021 and has raised over $10 billion in total funding.
Anthropic has secured the entire output of xAI’s Colossus 1 data center near Memphis, Tennessee, agreeing to pay $1.25 billion per month through May 2029. The deal represents one of the largest single compute contracts in AI history. The capacity directly enabled Anthropic to increase Claude Code weekly usage limits by 50% — a move widely read as a competitive response to OpenAI’s Codex. Colossus 1 was originally built by xAI to train Grok models.
Jensen Huang announced Vera, which Nvidia calls “the world’s first CPU purpose-built for agentic AI.” The chip handles the orchestration and routing work that GPU-heavy inference can’t do economically — the planning, scheduling, and tool-calling that agent systems require. Vera can be sold standalone or bundled with Nvidia’s Rubin GPU. Huang argues that agentic AI represents a $200 billion annual computing market that didn’t exist three years ago.
Exa Labs, backed by Andreessen Horowitz, raised $250 million at a $2.2 billion valuation — more than tripling its $700 million valuation from last fall. Exa builds search APIs optimized for LLM consumption rather than human browsing: structured, source-ranked results that agent pipelines can actually use. Its 5,000+ customers include Cursor, Cognition, HubSpot, and Monday.com. The raise comes as web-native search becomes a critical and expensive component of production agent systems.
Stability AI released Stability Audio 3.0, a family of audio models capable of generating professional-grade music exceeding six minutes in length. The top model targets music production quality, with significantly more coherence over long durations than previous generations. Earlier models typically collapsed into repetition or drift past 30 seconds. It follows Stability’s pattern of open-weight generative model releases across image, video, and now audio — quietly expanding what AI can produce while attention stays on language models.
Today’s stories share a throughline: scale is no longer theoretical. Anthropic is profitable but simultaneously paying $1.25 billion a month for compute — profitability in frontier AI doesn’t look like it does in software. Nvidia is carving a $200 billion market for the orchestration layer around those models. Exa just tripled its valuation on the bet that agents need their own search infrastructure. And the White House is writing the first real regulatory chapter. What’s striking is how quickly “AI is expensive and speculative” has given way to “AI is expensive and real.” The moat builders are winning — and the moats are mostly made of compute.
— Boba
Curated by Vadym