Today in AI — 5 June 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
Uber burned through its entire 2026 AI coding budget in four months. GitHub Copilot users are reporting monthly bills that jumped from $29 to $750 overnight. The "use AI as much as possible" era didn't last long.
The bill comes due
Three stories tell the same tale from different angles. Uber imposed a $1,500 monthly cap per tool per engineer after exhausting its annual budget by April. Copilot's switch to token-based billing on 1 June caught power users off guard, with some heavy agentic workflows costing 10x to 50x more than the old flat rate. And Microsoft is pulling Claude Code licences across its Experiences + Devices division by 30 June, steering engineers back toward Copilot CLI. When companies encouraged unlimited AI usage, they assumed costs would scale linearly. They didn't.
- Uber caps employee AI spending after blowing through budget in four months — TechCrunch
- GitHub Copilot's new token-based billing spurs developer backlash over cost spikes — TechCrunch
- Microsoft cancels Claude Code licences and pushes developers back toward Copilot CLI — The Decoder
Platforms build their own stack
Microsoft launched MAI-Thinking-1 and MAI-Code-1-Flash, designed to reduce its reliance on OpenAI, alongside MXC, an OS-level sandbox for AI agents. Apple will debut a Gemini-powered Siri overhaul at WWDC. Each company is reducing its dependence on a single AI supplier.
- Microsoft unveils MAI-Thinking-1 and MAI-Code-1-Flash to reduce OpenAI reliance — CNBC
- Microsoft launches MXC, an OS-level sandbox for AI agents with OpenAI and NVIDIA on board — VentureBeat
- Apple readies Gemini-powered Siri overhaul and standalone AI app for WWDC 2026 — TechCrunch
Open models at a fraction of the price
NVIDIA released Nemotron 3 Ultra, a 550B hybrid Mamba-Transformer model. MiniMax shipped M3 at 5-10% of rival pricing. For anyone watching the spending stories above, the pricing pressure from open weights is about to make proprietary per-token billing even harder to defend.
- NVIDIA open-sources Nemotron 3 Ultra, a 550B hybrid Mamba-Transformer model — MarkTechPost
- MiniMax M3 debuts as open-weight frontier model at 5-10% of rival pricing — VentureBeat
Capital keeps flowing
Alphabet announced equity offerings totalling $80 billion for AI infrastructure, with backing from Berkshire Hathaway. Anthropic filed its S-1 confidentially after reaching a $965 billion valuation and expanded Project Glasswing to 150 new organisations across 15 countries. OpenAI's GPT-5.5 and Codex went generally available on Amazon Bedrock.
- Alphabet launches $80 billion capital raise for AI infrastructure with Berkshire Hathaway backing — CNBC
- Anthropic confidentially files S-1 for IPO after reaching $965 billion valuation — Fortune
- Anthropic expands Project Glasswing to 150 new organisations across 15 countries — CNBC
- OpenAI's GPT-5.5 and Codex go generally available on Amazon Bedrock — AWS News Blog
Video generation heats up
xAI's Grok Imagine Video 1.5 took the top spot on the video generation leaderboard.
- xAI launches Grok Imagine Video 1.5 and takes the top spot on the video generation leaderboard — xAI
For builders: the era of uncapped AI tool budgets is over. Now the real procurement decisions begin.