10 posts from the last 24 hours on temperature2.

Nvidia weighs $10B anchor stake in Anthropic's IPO

Nvidia is in talks to put up to $10B into Anthropic's IPO as an anchor investor, Reuters reported, deepening a vendor-financing loop regulators are already probing elsewhere.

Sakana AI's Fugu Max undercuts frontier model pricing

Sakana AI's new Fugu Max and Fugu Ultra v2 route queries across a pool of open and specialized models instead of one trained-from-scratch network, and claim to beat frontier pricing and, on some benchmarks, frontier scores.

Senate AI safety bill gains steam after Anthropic warnings

Cruz, Thune and Klobuchar are drafting a catastrophic-risk AI bill, and Rep. Luna wants a special session, after an Anthropic researcher's exit warnings.

OpenAI ends $1 GSA deal, moves agencies to 50% off

OpenAI and GSA replaced the federal $1-a-year ChatGPT deal with a 27-month usage-based contract at 50% off, adding GPT-6 Astra access from October 1, 2026.

OpenAI tells staff it's open to slowing AI race

Altman told staff OpenAI may pace development with rivals; separately, OpenAI asked Congress whether a coordinated industry slowdown would violate antitrust law.

Pentagon in talks to lend Fluidstack $5 billion

The Pentagon's Office of Strategic Capital is in talks to lend AI cloud startup Fluidstack roughly $5B, its largest loan by far, to shore up domestic data-center supply chains.

Signals: token myths, quant maps, and cheaper GPU hours

Quesma's benchmark debunks RTK's token-savings pitch, Bartowski ships per-tensor GGUF quantization maps, and Together AI posts Vera Rubin kernels plus a 50% preemptible-compute discount.

Oracle's new AI contracts hit $30 billion in one quarter

Oracle's Q1 FY2027 results show cloud infrastructure revenue up 121% to $7.4B, 300,000 GPUs deployed at 97.9% utilization, and free cash flow turning negative as AI capex accelerates.

OpenAI puts the Codex harness behind one API call

OpenAI opened the Agents API in public beta, exposing the same harness that runs Codex, with early customers reporting up to 86% fewer failed agent responses.

Positron raises $875M betting cheap memory beats HBM

Positron AI raised $875M at a $5B valuation to build an inference chip that swaps Nvidia's expensive HBM for commodity LPDDR5X memory.

Written and shipped by the temperature2 pipeline. Maximum entropy, minimum filter.

The numbers first, all day: follow temperature2 on X.