LIVE · SAT, JUL 25, 2026 --:--:-- ET
Issue Nº 95 COST TOTAL $14942.87 ARTICLES TODAY 1 TOKENS TOTAL 9.65B
Podcast
aiexpert
Subscribe · $8/mo
Running the wire
Policy Jensen Huang debuts on X with open-weight AI letter, backing models against Washington restriction Chips Intel Foundry Q2: $5.8B revenue, external customer mix still 5% of total Chips Samsung locks $200B with Broadcom: HBM4 + 2nm foundry through 2030 for AI chips Breaking Moonshot AI halts Kimi K3 subscriptions as demand hits infrastructure limits amid U.S. chip export curbs Funding Paper raises $34M Series A from Accel, ICONIQ for AI-native design platform Research Anthropic releases Claude Opus 5: matches Fable on benchmarks, 50% cheaper, praised for coding Funding Cathedral: DOGE alumni raise $160M for stealth AI military cybersecurity startup at $1.4B valuation Market NVIDIA locks down SK Hynix HBM supply with $500B AI infrastructure deal Research Poolside releases Laguna S 2.1, 118B open-weight model beating DeepSeek V4 Flash on agentic tasks Funding Etched raises $300M Series C at $10.3B valuation, values inference chips above general training Market SpaceX Starship Flight 13 deploys 20 Starlink V3 satellites post-IPO, stock down 43% from peak Funding Glow emerges from stealth at $1.2B; AI-native endpoint security for enterprise devices Funding Sila raises $300M for silicon anode expansion; pivots beyond EV to AI, defense, robotics Market Samsung launches RX robotics division; Korea pledges 312T won to Yeongnam AI hub Chips Fortinet becomes Intel Foundry's first named external customer on Intel 4 Market Tesla Q2 free cash flow turns negative $1.1B as capex explodes 142% funding Robotaxi, Optimus, chip fab Market BlackRock markets $12.3B bonds for Meta El Paso data center as AI debt fuels infrastructure buildout Market Waymo launches standalone app in Atlanta, Austin as Uber exclusivity ends January 2028 Funding Etched raises $300M at $10.3B valuation; aims for frontier-scale AI inference at lower power, higher density Funding Atoms raises $1.7B from a16z for physical AI in mining, food, transport Policy Jensen Huang debuts on X with open-weight AI letter, backing models against Washington restriction Chips Intel Foundry Q2: $5.8B revenue, external customer mix still 5% of total Chips Samsung locks $200B with Broadcom: HBM4 + 2nm foundry through 2030 for AI chips Breaking Moonshot AI halts Kimi K3 subscriptions as demand hits infrastructure limits amid U.S. chip export curbs Funding Paper raises $34M Series A from Accel, ICONIQ for AI-native design platform Research Anthropic releases Claude Opus 5: matches Fable on benchmarks, 50% cheaper, praised for coding Funding Cathedral: DOGE alumni raise $160M for stealth AI military cybersecurity startup at $1.4B valuation Market NVIDIA locks down SK Hynix HBM supply with $500B AI infrastructure deal Research Poolside releases Laguna S 2.1, 118B open-weight model beating DeepSeek V4 Flash on agentic tasks Funding Etched raises $300M Series C at $10.3B valuation, values inference chips above general training Market SpaceX Starship Flight 13 deploys 20 Starlink V3 satellites post-IPO, stock down 43% from peak Funding Glow emerges from stealth at $1.2B; AI-native endpoint security for enterprise devices Funding Sila raises $300M for silicon anode expansion; pivots beyond EV to AI, defense, robotics Market Samsung launches RX robotics division; Korea pledges 312T won to Yeongnam AI hub Chips Fortinet becomes Intel Foundry's first named external customer on Intel 4 Market Tesla Q2 free cash flow turns negative $1.1B as capex explodes 142% funding Robotaxi, Optimus, chip fab Market BlackRock markets $12.3B bonds for Meta El Paso data center as AI debt fuels infrastructure buildout Market Waymo launches standalone app in Atlanta, Austin as Uber exclusivity ends January 2028 Funding Etched raises $300M at $10.3B valuation; aims for frontier-scale AI inference at lower power, higher density Funding Atoms raises $1.7B from a16z for physical AI in mining, food, transport
Chips

Samsung locks $200B with Broadcom: HBM4 + 2nm foundry through 2030 for AI chips

Samsung Electronics and Broadcom signed a memorandum of understanding on July 24 for a strategic partnership estimated at over $200 billion through 2030, covering high-bandwidth memory supply and advanced foundry services. Samsung will supply HBM4 and next-gen HBM4E memory to power Broadcom's next-generation AI accelerators, while its foundry arm manufactures Broadcom's products on 2-nanometer and below process nodes. The deal also covers advanced 2.3D and 2.5D packaging for tighter memory-to-logic integration.

The agreement reflects tight AI memory supply: global memory shortages have driven DRAM and NAND prices up 50–95% quarter-over-quarter in early 2026, with supply expected to remain constrained into 2027. Samsung, which fell behind on HBM yields initially, now codevelops HBM4E (targeting 64GB, 16-layer stacking, 16Gbps pins) to compete with SK Hynix's supply lock with Nvidia. Broadcom gains supply resilience and access to Samsung's vertically integrated memory and packaging capabilities for custom AI chips.

Part of a larger $950 billion Korea–US semiconductor cooperation package announced the same day—with SK Hynix committing $750 billion to Nvidia alone. For architects, this stabilizes HBM availability for non-Nvidia ecosystems (AMD, Broadcom custom accelerators) and signals Samsung's return as a credible premium memory supplier after years of losing share to competitors.

Sources

Get the signal before the noise.