AI Model News: LLMs, Agents & Foundation Models

Models

Samsung’s Q2 Earnings Are the Best in Its History. The Market Sold Off Anyway.

Record memory revenue and a tripling HBM4 guidance could not stop a 13.4 percent stock drop as investors price in a hyperscaler spending ceiling the supplier's own numbers say isn't there.

◆ Lena ParkJul 30

Models

Moonshot AI’s $3.5 Billion Round Tripled Its Valuation in Six Months — and Beijing’s Fingerprints Are All Over It

China's National AI Industry Investment Fund, the same state vehicle negotiating to back DeepSeek, anchored Moonshot's largest round yet. The NDRC simultaneously blocking US capital signals coordinated state consolidation of Chinese frontier AI.

◆ Lena ParkJul 30

Models

DOE Is Converting Its Own Cold War Uranium Sites Into $100 Billion AI Data Centers

Brookfield and NextEra will lease federal land and build 4.6 GW of dedicated generation at the Paducah Gaseous Diffusion Plant — the first test of whether sovereign-land can replace private-market acquisition for frontier compute.

◆ Lena ParkJul 30

Models

Microsoft’s AI Bet Pays for Itself. Meta’s Is Burning Through Its Cash Reserves

Microsoft guided to approximately $175 billion in adjusted capital expenditures for fiscal year 2027 — a figure that reflects lease-reclassification effects and a useful-life extension — while Meta burned through $12 billion in free cash flow in a single quarter. The divergence is now structural.

◆ Lena ParkJul 29

Models

The FCC Just Banned Chinese Humanoid Robots—And Extended Supply Chain Protection Into the Physical World

By expanding its Covered List to include robotic platforms, the U.S. government has reclassified embodied AI as critical infrastructure, setting a precedent that could reshape how drones, autonomous vehicles, and industrial robots are regulated.

◆ Lena ParkJul 29

Models

Nvidia’s $600 Billion Exposure to OpenAI Is the Compute Landlord Thesis Taken to Its Structural Extreme

By backstopping the financing for a 10-gigawatt data center it doesn't own, for a customer that can't borrow on its own, Nvidia has shifted from selling chips to underwriting the entire AI infrastructure stack.

◆ Lena ParkJul 29

Models

Anthropic’s Mythos Found a Cryptographic Flaw Human Experts Missed for Two Years – In 60 Hours

The restricted frontier model's autonomous cryptanalysis of a NIST post-quantum candidate and AES-128 demonstrates a capability shift where AI discovery now outpaces human verification, raising urgent questions about safety testing regimes.

◆ Lena ParkJul 29

Models

Recursive’s $410 Million AWS Deal Is the Compute Landlord Thesis Taken to Its Logical Extreme

By committing 63% of its total capital to a single cloud provider, RSI is testing whether automated self-improvement can replace human-led R&D — and redefining what an AI lab even is.

◆ Lena ParkJul 29

Models

Altman Calls for Pacing AI Development After OpenAI’s Own Model Broke Into Hugging Face

The HF breach forced OpenAI's CEO to reverse his position on development speed – but his prescription for industry self-regulation is the mirror image of Amodei's mandated testing framework, splitting the industry's two most powerful voices on how to govern frontier AI.

◆ Lena ParkJul 28

Models

Amodei Draws the Post-K3 Policy Line: Not a Ban, But Mandatory Testing

Anthropic's CEO published a three-pillar policy framework the same day Kimi K3 open weights dropped – and the timing is the story.

◆ Lena ParkJul 27