The $1.5B bet against GPU dominance.

The Frontier

Science & tech breakthroughs in AI models, hardware, and computing that define what's possible

Tenstorrent raises $1.5 billion to challenge GPU dominance. AI infrastructure company Tenstorrent closed a $1.5 billion Series E at a $6.2 billion post-money valuation. It plans to use the capital for volume manufacturing of Galaxy servers, next-generation hardware and intellectual property, advancing its open-source-oriented inference infrastructure against GPU incumbents.

Tenstorrent’s $6.2B valuation, up 2.4× since 2024, is a bet that open-source-oriented inference can justify server-scale manufacturing. Its $1.5B round funds Galaxy volume production and next-gen hardware/IP; the test is whether that spend creates a credible GPU alternative.

Betakit

Error-correction record brings useful quantum computing closer. Researchers have set a new reliability record for logical qubits, the error-corrected computational units needed for useful large-scale quantum computers. The result advances quantum error correction, still a central hurdle in turning unstable physical qubits into fault-tolerant systems.

Newscientist’s logical-qubit reliability record is more important than larger physical-qubit totals because it targets fault tolerance’s central constraint. Watch for whether subsequent logical-qubit results preserve that reliability as systems scale.

Newscientist

Anthropic gives an outside AI watchdog employee-like access. Anthropic has given nonprofit evaluator METR employee-like access, including an embedded team with access to internal hardware and systems, to assess risks from Claude models. The arrangement expands third-party model evaluation beyond conventional external testing and follows Anthropic's public commitments to stronger AI safety safeguards.

Anthropic is making METR an embedded counterweight, not a conventional outside tester, with access to internal hardware and systems. The test over the next year is whether that access produces risk findings Anthropic would not surface itself.

Ars Technica

Capital & Control

Tracking the flow of capital and influence that shapes the tech landscape

Anthropic plans to commit 5% of shares to a public-good fund. Anthropic said it will commit 5% of its shares to a public-good fund, subject to shareholder approval, and add another 2% for every 50% increase in value above an undisclosed baseline. Using Anthropic's May valuation of $965 billion, the shares were estimated at about $48 billion as the company prepares for a potential stock-market listing.

Anthropic’s $48B public-good carve-out is IPO governance design: it could trade a 5% stake for regulatory legitimacy while preserving founder control. The test before any stock-market listing is whether shareholders approve it.

The Guardian

Infrastructure & Power

The physical backbone of AI: compute, data centers, supply chains, and the geopolitics of energy

European carriers explore a direct-to-device satellite service without Starlink. Deutsche Telekom, Orange, Telefónica and Vodafone signed a letter of intent to consider a joint spectrum bid for a European direct-to-device satellite service. The carriers would provide ground infrastructure while an unnamed partner would operate the satellite constellation, positioning the effort against SpaceX's Starlink service.

Deutsche Telekom, Orange, Telefónica and Vodafone still face the real squeeze: securing a contiguous 2 GHz band. They must also power ~3 000 gateway sites at 10 kW each, a 30 MW draw equal to a midsize wind farm. If auctions stall, Starlink holds Europe past 2028.

The Verge

Digital Defense

Threats and defenses shaping the global security landscape

AI agents reportedly breached John Deere, SharkNinja and Cummins. A hacker reportedly used an array of AI agents to conduct much of the intrusion activity against John Deere, SharkNinja and Cummins with limited human involvement. The reported incidents offer a concrete example of AI agents automating cyberattack operations against major industrial and consumer-technology companies.

John Deere, SharkNinja and Cummins were reportedly targeted by a hacker using AI agents for much of the intrusion activity. The test is whether future reporting shows those agents maintained access after initial entry.

The Verge

The Performance Edge

Peak performance science for sustained execution in high-stakes environments

3,000 daily steps may help slow Alzheimer's-related decline. A new study linked daily physical activity with slower Alzheimer’s-related tau accumulation and slower cognitive and functional decline among people showing early disease signs. Benefits appeared from roughly 3,000 daily steps, with the strongest association at 5,000 to 7,500 steps rather than the commonly cited 10,000-step threshold.

3,000 daily steps slowed Alzheimer’s cognitive-functional decline by almost 50% across 9–11 years in a cohort of ~300 adults already showing early pathology. Less tau accumulated, with peak benefit at 5,000 to 7,500 steps. Clinics can recommend step-count goals now; watch payers trial activity-based coverage by 2028.

Science Daily

Market Pulse

Daily market data: indices, sectors, top movers, and volatility

Markets climb as technology leads. The S&P 500 rose 0.7%, the Nasdaq 100 gained 1.5%, the Dow Jones added 0.1% and the Russell 2000 advanced 0.4%. Technology led sectors with a 1.7% gain while Industrials fell 0.4%; SMCI climbed 4.0%, AMZN rose 3.6% and PLTR gained 3.3%.

SMCI’s 4% pop shows money crowding back into AI server supply chains while Industrials fade. Today’s sector gap (+1.7% Tech vs -0.4% Industrials) pulled AMZN +3.6% and PLTR +3.3%, underscoring AI’s capital magnet. Expect funds to keep rotating into GPU-adjacent names ahead of Q4 datacenter capex guides.

Yahoo Finance