2026-05-19
By Vadym · Generated with AI, curated by me
The conversation about who pays for AI — and who pays the price — dominated the past 24 hours. Valuations climbed into the hundreds of billions, billing models sharpened into per-token precision, and workforce cuts that were telegraphed for weeks finally hit payroll systems.
Anthropic is negotiating at least $30 billion in new financing at a pre-money valuation exceeding $900 billion, Bloomberg reported — more than double the $380 billion it commanded in its Series G just three months ago. Dragoneer, Greenoaks, Sequoia Capital, and Altimeter Capital are expected to co-lead. No term sheet has been signed yet, but the round is expected to close as soon as the end of May. The valuation acceleration reflects Claude Code’s rapid enterprise adoption and Q1 2026 ARR reported above $44 billion. At $900B pre-money, Anthropic would trail only a handful of public tech giants and top every other private AI company by a wide margin.
Cursor released Composer 2.5 on May 18, its first proprietary AI coding model built using reinforcement learning trained on long, complex coding sessions. The model matches Opus 4.7 and GPT-5.5 on SWE-bench and Aider polyglot benchmarks while pricing at $0.50/M input and $2.50/M output — roughly one-tenth the cost of frontier models. Cloud agent dev environments, Microsoft Teams integration, and Build in Parallel shipped alongside. The strategic implication is significant: Cursor is no longer simply layering on third-party APIs. It’s training its own foundation model to compete on both capability and cost, which puts it in a different category than tools that just route to whoever wins the benchmark war this month.
Meta started notifying 8,000 employees — 10% of its workforce — on May 20, while simultaneously cancelling 6,000 open roles. Teams are being restructured into AI-focused pods under new Chief AI Officer Alexandr Wang’s Superintelligence Labs as Meta commits $115–135 billion to AI infrastructure in 2026. The cuts are structural rather than performance-based; affected US workers receive 16 weeks of base pay plus two additional weeks per year of tenure. Meta’s move is the largest single-employer cut in a quarter that already saw 80,000 tech sector jobs eliminated industry-wide, with nearly half attributed to AI displacement. The math is blunt: Meta’s annual AI infrastructure spend now exceeds its total annual payroll.
Starting June 1, all GitHub Copilot plans transition from flat monthly access to an AI Credits model: 1 credit = $0.01 USD, consumed based on token usage at listed API rates. Copilot Pro gets $10/month in credits included, Business $19/user, Enterprise $39/user. Code completions and next-edit suggestions remain unlimited for all paid plans; the change primarily hits heavier model usage and agentic code review sessions. GitHub is rolling out a preview billing dashboard so users can see projected costs before the cutover. The move makes Copilot billing structurally identical to API pricing, ending the flat-rate buffer that made heavy agentic use effectively free for Pro and Business subscribers.
Starting June 15, programmatic Claude usage — Agent SDK calls, claude -p, Claude Code GitHub Actions, and third-party agents — moves out of subscription limits into a separate monthly credit pool billed at full API list prices. Interactive Claude Code in the terminal is unaffected. Credit allotments: Pro gets $20/month, Max 5x gets $100, Max 20x gets $200; credits expire monthly with no rollover. The change targets developers running automated workflows at scale — CI pipelines, scheduled agents, and scripts that call Claude programmatically without a human in the loop. Between this and GitHub Copilot’s June 1 transition, the era of flat-rate AI tooling for heavy users is definitively over.
Five stories, one through-line: AI’s adolescence is ending. A $900 billion valuation for a private lab. Token-based billing replacing flat monthly subscriptions at two major developer platforms simultaneously. An 8,000-person layoff explicitly framed as a reallocation to AI infrastructure. Cursor training its own model rather than reselling someone else’s. None of these are signals of a bubble deflating — they’re signs of an industry that has moved past proving itself and is now optimizing for unit economics. The companies that survive this phase will be the ones who figured out that the cost structure, not the benchmark, determines the winner.
— Boba
Curated by Vadym