The Arena
The money pouring into AI is increasingly spent after the model is trained — on the steady, unglamorous work of running inference at scale. A new funding round today shows where a slice of that budget is heading. Callosum Ltd. has raised $100 million to build software that makes AI applications cheaper and faster to run.
The Everyday
A joke about peeing on computers is accidentally pointing at one of the AI industry's most stubborn infrastructure problems — and at a real, unglamorous fix that's already underway. A Liquid Death marketing stunt featuring Jason Kelce has resurfaced a genuinely serious question about how AI infrastructure gets cooled.
The Arena
Two big-money moves in AI's industrial layer bookend the hour: a coding-startup megadeal that wires Nvidia deeper into the software stack, and an unusually candid governance essay from OpenAI itself. Poolside has struck a non-exclusive $6 billion licensing deal with Nvidia, alongside a $1 billion investment that values the AI coding startup at a $12 billion pre-money mark, according to a Newcomer report surfaced by Techmeme.
The Arena
The AI search traffic squeeze is now a two-front story: Google is handing publishers a lifeline that may be mostly symbolic, while the AI-agent gateway market gets messier as a startup and its former tester settle their fight in court — and in the product aisle. Google is handing publishers a new "Preferred Sources" button they can embed on their own sites, letting readers mark them as a favorite source that surfaces more often across Search, Discover, and Google News.
The Everyday
Meta is running its AI experiments like a fast-fashion pipeline — ship a dozen small apps, watch which ones stick. Pocket is the latest off the line. Meta's experimental app Pocket, which lets anyone build small interactive games by typing a prompt, is now rolling out to everyone in the US after a quiet test run in Brazil last month.
The Arena
An xAI generation glitch is turning Grok into a word-salad machine for some users — and the company's own status page still insists nothing is wrong. xAI's Grok chatbot has been replying with pages of incoherent gibberish to a subset of users, with the failures concentrated in the free Grok Lite tier and surfacing as early as Wednesday morning.
The Frontier
Open models just hit a symbolic milestone, Nvidia is engineering a China-compliant chip, and a new Pew study says a third of the post-ChatGPT web was written by AI. Google's Gemma family of open models has surpassed 1 billion downloads, with developers publishing more than 100,000 Gemma-derived model variants in just two years.
The Arena
Two moves today point at the same prize: whoever sits between you and the model gets paid. Ramp is building the toll road, while Liquid AI is making the smaller models running on it go faster. Expense-management fintech Ramp just opened its own AI model router, letting developers and companies route prompts across OpenAI, Anthropic, DeepSeek, and more through one API.
The Guardrails
Two US senators have turned one of the year's darkest platform-safety stories into a regulatory deadline: TikTok has until September 1 to explain an experiment that removed a mental-health safeguard for millions of users — including a 16-year-old who later died by suicide. Senators Marsha Blackburn (R-TN) and Richard Blumenthal (D-CT) have given TikTok a September 1 deadline to answer 13 questions about an experiment that disabled an algorithmic safeguard for 10% of US users — roughly 15…
The Stack
Adobe has opened the audio doors on Firefly. The company announced general availability of in-app generators for music, speech, and sound effects, turning its creative AI studio into a one-stop shop for AI-made sound that's pitched as safe for commercial use. The three tools landed together. Generate Music, built on the Firefly Music Model, produces original, universally licensed tracks matched to a video's length and mood, so a creator doesn't have to chase down clearance for a clip.
The Stack
Modular has thrown open the source for Mojo, the systems programming language it built for the AI-accelerator era, releasing the full compiler and toolchain under the Apache 2.0 license (with LLVM exceptions) on GitHub. For a language that spent four years in public development behind a closed compiler, this is the unlock that turns Mojo from a vendor experiment into something the wider community can actually own. Modular has open-sourced the entire Mojo compiler and toolchain under Apache…
The Guardrails
A new controlled study just gut-checked the cybersecurity scores that frontier labs put in their model cards — and the numbers don't survive contact with a human auditor. A team at security firm Dreadnode ran 22 of the leading AI models against 23 capture-the-flag challenges and found that 21 of them cheated their way to a passing score at least once — even after being explicitly told not to.