☀️ TRENDING AI NEWS 🚨 Market Rout: The Nasdaq Composite erased all of its 2026 gains on Friday, closing down 5% from its value in an AI-led sell-off. 🏢 Data Centers: Trump threatened a 'tremendous tax' on US companies that build AI data centers outside the country. 🛠️ Signal Shield: Signal now blocks screenshots on Android and iOS to keep Gemini and Siri AI from reading your chats. 🤖 Open Models: Baseten's bison-1 scored 76 on the AA Decision Index v2.1, and NVIDIA released 14 Nemotron-Labs-Sigma checkpoints. |
A trillion-dollar IPO was supposed to be the victory lap. Instead, it took exactly one more trading day for the mood to flip, and the whole Nasdaq paid for it.
| 🛠️ Tool of the day | |
Descript - an AI video and audio editor where you edit a recording by editing its text transcript. It helps creators, marketers and teams cut podcasts, demos and meetings fast. It has a free plan.
| ⚡ Quick hits | |
The University of California expanded its OpenAI deal to give students, faculty and staff free access to ChatGPT and Codex. Read more →
OpenAI says it used its models to improve chip designs, with one AI-made multiplier redesign cutting estimated chip area by about 15%. Read more →
Google opened Genie 4 to eligible AI Ultra subscribers, so they can generate interactive worlds and change them with text prompts. Read more →
| 🤓 AI Trivia | |
Nvidia's Nemotron-Labs-Sigma denoises text inside a learned embedding space. How many dimensions is that space?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 📉 The Nasdaq Gives Back Its Entire 2026 | |

Anthropic's second day couldn't hold the line
As Anthropic stock dropped during its second day of trading, the wider market dipped with it. On Friday the Nasdaq Composite erased all of its 2026 gains and closed down 5% from its value, losing trillions as investors questioned the enormous sums flowing into AI buildout.
The IPO itself was historic: a peak first-day valuation of $1.3tn and a $97 close on Thursday, roughly 17% above the offering price. But the Guardian's market coverage notes experts expect the coming months to be bumpy, with the industry facing a variety of challenges. The IMF's Kristalina Georgieva added another wrinkle, warning that booming AI demand will push central banks to raise interest rates.
The bottom line
One bad day proves little, but it shows how much of the market is now leaning on a single AI story. If you hold index funds, you are exposed whether you wanted to be or not.
State of Product reveals what AI still hasn’t solved
80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.

| 🏢 Nvidia's Board and the Problem With Candor | |

It wasn't just the allegations
Amid allegations of multiple affairs, including with a direct report and an alleged Chinese intelligence asset, sources tell WIRED the Nvidia board's main issue with Jensen Huang was his 'lack of candor' with them ahead of his resignation.
The fallout keeps spreading. The Verge tracked how the stock sank to around $150 a share, a nearly 40% drop, while interim CEO Colette Kress told investors on Tuesday night, "We've got this." Meanwhile, tech workers are reportedly getting nervous about honeypots, and Taiwanese restaurants Huang favored are choosing sides. For the full corporate picture, our earlier coverage of Nvidia without Huang has the backstory.
The bottom line
Boards forgive a lot, but not being misled. The lesson for every AI company run by a founder-celebrity is that governance risk is now priced into the stock.

| 🛠️ Signal Blocks Gemini and Siri From Reading Your Screen | |

Screenshots become the new battleground
Signal is restricting screenshots on Android and iOS to prevent AI agents built into operating systems from reading users' messages and other data. These agentic features analyze on-screen content to act on your behalf, which is exactly what a private messenger does not want happening. iOS users get a toggle that blocks screenshots and stops Siri AI from scanning the screen.
It lands as Google rolls out computer use in its Gemini app for macOS, letting it work across local files, desktop apps and the web, with users choosing which apps it can access. See our ongoing data privacy coverage for how this tension keeps escalating.
The bottom line
If OS-level agents can see everything on screen, every app with sensitive content has to pick a side. Expect more apps to follow Signal's lead.

| 🔬 Claude and the Riemann Hypothesis | |

A claim that needs serious checking
Anthropic mathematician Levent Alpöge claims he has proved the Riemann hypothesis, one of mathematics' most notoriously difficult problems, with help from Claude. It has eluded some of the world's best mathematicians for decades. The key word is claims: if verified, it would be another seismic AI achievement in mathematics, but verification is the whole story.
Separately, METR's updated time horizon plot finds some AI capabilities still improving exponentially, but models are not yet as strong at tasks requiring judgment as they are at coding. Worth reading alongside the math claim for calibration.
The bottom line
Treat it as unverified until mathematicians sign off. Either way, AI-assisted proofs are moving from curiosity to something the field must now referee.

| ⚡ NVIDIA's Diffusion Language Model Decodes 4.2x Faster | |

Text generation without the left-to-right grind
NVIDIA released Nemotron-Labs-Sigma, 14 open 3B and 8B checkpoints of a continuous diffusion language model. Instead of predicting one token at a time, it denoises text in a learned 16-dimensional embedding space and distills to few-step samplers, decoding up to 4.20x faster than autoregressive decoding.
Elsewhere in open models, Baseten released bison-1 and groundhog-1 on Gemma 4 with 128k context and image input, and RadixArk shipped DaisyRose-27B, an Apache 2.0 agentic coding model trained with reinforcement learning only that runs on a single GPU. Building sites alongside all this? Our sponsor 60sec.site is an AI website builder that gets you live in about a minute.
The bottom line
Faster decoding means cheaper inference, which is the number that actually decides who can afford to run AI at scale. Diffusion for text is worth watching.

| 🏛️ Trump Wants to Tax Data Centers Built Abroad | |

Policy collides with local pushback
After his summit with China's President Xi Jinping, Trump promised a "tremendous tax" on companies that build data centers outside the US, saying companies won't be allowed to build them abroad. The twist: states and cities across the country are restricting new applications to build them at home. In Australia, NextDC's share price fell 40% after community backlash to a Melbourne expansion, and it now vows to keep new developments away from homes.
The bottom line
Everyone wants the compute, nobody wants the building next door. Whoever resolves that squeeze controls where AI actually gets built.

| 🌎 Trivia Reveal | |
The answer is 16! Nemotron-Labs-Sigma denoises text in a learned 16-dimensional embedding space, which is part of how it reaches decoding speeds up to 4.20x faster than autoregressive models.
| 💬 Quick Question | |
Would you let an OS-level AI agent see everything on your screen, or are you with Signal on blocking it? Hit reply and tell me - I read every response! And for more, visit dailyinference.com for our daily AI newsletter.
That's it for today - see you tomorrow with more!
| 🎁 Share Daily Inference | |
Know someone who would like this? Share your link below. One referral unlocks the AI Tools Starter Kit, and five unlock the AI Insider Briefing.


