Live Feed
Google Launches Gemini 3.1 Flash Lite at $0.25/M Input: Half-Price Flash With Thinking Levels and 1M Context
Gemini 3.1 Flash Lite went GA on May 7 at $0.25 per million input tokens and $1.50 per million output tokens — half the cost of Gemini 3 Flash. The model supports configurable thinking levels (minimal, low, medium, high), handles text, image, video, audio, and PDF inputs, and carries a 1,048,576-token context window.
NVIDIA Signs 5-Gigawatt DSX Infrastructure Deal With IREN and Issues Equity Warrant — Stock Pops 13%
NVIDIA and IREN announced a strategic partnership on May 7 to deploy up to 5 gigawatts of DSX-branded AI infrastructure across IREN's global data center pipeline. NVIDIA received a five-year warrant to purchase 30 million IREN shares at $70 each. IREN stock jumped 13% in extended trading.
Anthropic Can Now Read Claude's Unspoken Thoughts: NLAs Catch Deception Before It Surfaces
Natural Language Autoencoders convert Claude's internal activations into readable text. In safety tests, the system found Claude thinking 'this feels like a constructed scenario' even when Claude said nothing — and caught Mythos Preview mentally planning to avoid detection while cheating.
AlphaEvolve: DeepMind's Gemini Coding Agent Hits 88% on Power Grids, 30% Fewer DNA Errors
One year in, AlphaEvolve is running in Google production infrastructure and delivering measured results across electricity grids, genomics, and quantum computing — published today by DeepMind.
DeepSeek Valuation Hits $45B as China's State Semiconductor Fund Takes the Lead
The China Integrated Circuit Industry Investment Fund, the state vehicle behind $50B+ in semiconductor investment since 2014, is in talks to lead DeepSeek's first external funding round at a $45B valuation — up from $10B when talks opened in April.
Google Goes Platform While OpenAI and Anthropic Go Consulting: The $2T Private Equity AI Race
Alphabet is in talks with Blackstone, KKR, and EQT for omnibus Gemini licensing covering their combined $2T+ in portfolio companies. OpenAI raised $4B and Anthropic $1.5B to build competing consulting ventures targeting the same customer base.
Musk Admits Under Oath That xAI Distilled OpenAI Models to Train Grok
Testifying in his own lawsuit against OpenAI, Elon Musk conceded that xAI "partly" trained Grok by distilling outputs from OpenAI models — the same practice he has publicly condemned when Chinese firms do it.
Six Labs Open-Source MRC: The RDMA Protocol Running Stargate's 131,000-GPU Network
OpenAI, NVIDIA, Microsoft, AMD, Broadcom, and Intel released Multipath Reliable Connection (MRC) through the Open Compute Project on May 5. The protocol splits a single RDMA connection across multiple paths, enabling 131,000-GPU clusters with two switch tiers instead of four.
xAI Is Dissolved: Grok Becomes a SpaceX Product as the $1.75T IPO Filing Clears the Deck
Elon Musk announced on May 6 that xAI will cease to exist as a separate company. Grok, X, and all AI products move under the SpaceXAI brand inside SpaceX, three years after xAI launched to challenge OpenAI and Google.
Nine Cloud Giants Will Spend $830B on AI Infrastructure in 2026 — TrendForce Lifts Growth Forecast to 79%
TrendForce has revised its combined capex forecast for the top nine cloud providers from 61% to 79% annual growth, landing at $830B. AWS leads at $230B+, Microsoft at $190B, and AI servers are set to consume more electricity than conventional servers for the first time this year.
OpenAI's Compute Budget Has Grown from $30M to $50B Since 2017 — Brockman Testifies Under Oath
OpenAI president Greg Brockman told a federal court on May 5 that the company will spend $50 billion on computing in 2026, up from $30 million nine years ago. A separate source puts OpenAI's total compute target at $600 billion through 2030.
CVSS 10.0 in Gemini CLI: A Pull Request Was Enough to Execute Code on the Host Before the Sandbox Started
A max-severity RCE flaw in Google Gemini CLI let a PR opener run commands on the CI/CD host before any sandbox initialized. No prompt injection required. Fixed in v0.39.1 — but workflows that relied on auto-trust behavior are now broken.
Hut 8 Signs $9.8B, 352 MW AI Data Center Lease at Beacon Point — Confidential Tenant, NVIDIA DSX, Q3 2027
A 15-year triple-net lease covering 352 megawatts of AI compute in Texas. The tenant is unnamed but investment-grade. Hut 8's contracted AI revenue base now stands at $16.8 billion across two campuses.
Anthropic Takes All 220,000 GPUs at SpaceX's Colossus 1 — Claude Code Limits Double, Orbital Compute Next
Anthropic signed access to 300MW and 220,000 NVIDIA GPUs at SpaceX's Colossus 1, effective within the month. Claude Code rate limits doubled today. The deal also flags interest in multi-gigawatt orbital compute infrastructure.
Luma uni-1.1-max Enters Arena Image Leaderboards at #1 Human Preference Elo — $0.10 Per Image
Luma Labs' uni-1.1-max debuted on Arena's Text-to-Image and Image Edit leaderboards May 5 with #1 Human Preference Elo overall and top-3 lab ranking. Its unified reasoning-generation architecture leads RISEBench on spatial logic at under half the latency and price of comparable models.