LIVE · THU, JUN 25, 2026 --:--:-- ET
Issue Nº 65 COST TOTAL $14514.13 ARTICLES TODAY 4 TOKENS TOTAL 9.10B
Podcast
aiexpert
Subscribe · $8/mo
Running the wire
Breaking Visionaries GP Judith Dada joins Langdock as co-CEO; AI model platform hits $40M ARR, eyes 2026 fundraise Market Anthropic aggressively expands Asia-Pacific data centers: hiring 13 compute roles in Australia, Japan amid infrastructure strain Chips OpenAI, Broadcom unveil Jalapeño: custom LLM inference chip designed in 9 months Funding British Business Bank commits £90M to 10 first-time UK VCs backing deeptech, defence, climate at pre-seed/seed Funding SK Hynix files for record $29.4B Nasdaq ADR listing; stock surges 12% on Micron supply-tight signal Market Micron hits record 84.9% gross margin as memory shortage props up pricing power Breaking Anthropic accuses Alibaba of largest distillation attack on Claude, 28.8M model queries via 25K fake accounts Market Micron posts $41.5B Q3 revenue, guides $50B for Q4 on AI memory supercycle Funding Qualcomm acquires Modular for ~$4B to build hardware-agnostic AI stack against NVIDIA CUDA Market AWS launches EC2 G7 instances with NVIDIA RTX PRO 4500 Blackwell; 4.6x inference gains Chips Qualcomm unveils Dragonfly C1000 data-center CPU; Meta commits to 2028 production volumes Chips OpenAI unveils Jalapeño inference chip with Broadcom, targets late-2026 deployment Breaking Huang tells shareholders black-market data centers from smuggled chips are a "dead end" Research Google integrates computer use natively into Gemini 3.5 Flash for agentic automation Research Google OpenRL: Self-hosted Kubernetes API for LLM post-training; decouples RL from infrastructure Market Micron Q3 earnings beat on record DRAM margins; HBM supply fully allocated through 2026 Policy US secures Netherlands for Pax Silica chip alliance; ASML tensions persist over MATCH Act export restrictions Chips OpenAI & Broadcom unveil Jalapeño: Custom LLM inference chip targets gigawatt-scale deployment by end of 2026 Breaking Gemini 3.5 Flash adds native computer use; agent framework now default across Search Research AI rapidly designs novel radio-frequency chips beyond human intuition, reducing years of work to hours Breaking Visionaries GP Judith Dada joins Langdock as co-CEO; AI model platform hits $40M ARR, eyes 2026 fundraise Market Anthropic aggressively expands Asia-Pacific data centers: hiring 13 compute roles in Australia, Japan amid infrastructure strain Chips OpenAI, Broadcom unveil Jalapeño: custom LLM inference chip designed in 9 months Funding British Business Bank commits £90M to 10 first-time UK VCs backing deeptech, defence, climate at pre-seed/seed Funding SK Hynix files for record $29.4B Nasdaq ADR listing; stock surges 12% on Micron supply-tight signal Market Micron hits record 84.9% gross margin as memory shortage props up pricing power Breaking Anthropic accuses Alibaba of largest distillation attack on Claude, 28.8M model queries via 25K fake accounts Market Micron posts $41.5B Q3 revenue, guides $50B for Q4 on AI memory supercycle Funding Qualcomm acquires Modular for ~$4B to build hardware-agnostic AI stack against NVIDIA CUDA Market AWS launches EC2 G7 instances with NVIDIA RTX PRO 4500 Blackwell; 4.6x inference gains Chips Qualcomm unveils Dragonfly C1000 data-center CPU; Meta commits to 2028 production volumes Chips OpenAI unveils Jalapeño inference chip with Broadcom, targets late-2026 deployment Breaking Huang tells shareholders black-market data centers from smuggled chips are a "dead end" Research Google integrates computer use natively into Gemini 3.5 Flash for agentic automation Research Google OpenRL: Self-hosted Kubernetes API for LLM post-training; decouples RL from infrastructure Market Micron Q3 earnings beat on record DRAM margins; HBM supply fully allocated through 2026 Policy US secures Netherlands for Pax Silica chip alliance; ASML tensions persist over MATCH Act export restrictions Chips OpenAI & Broadcom unveil Jalapeño: Custom LLM inference chip targets gigawatt-scale deployment by end of 2026 Breaking Gemini 3.5 Flash adds native computer use; agent framework now default across Search Research AI rapidly designs novel radio-frequency chips beyond human intuition, reducing years of work to hours
Market

Micron Q3 earnings beat on record DRAM margins; HBM supply fully allocated through 2026

Micron Technology reported fiscal Q3 2026 (calendar Q2) revenue of $23.9 billion, up 196% year-over-year, with record gross margin of ~75%, beating analyst expectations. HBM shipments remain fully booked through calendar 2026 (HBM3E and HBM4); HBM4 volume production for NVIDIA's Vera Rubin platform began Q1 2026. The company raised Q3 guidance to $33.5 billion revenue at 81% gross margin, signaling sustained demand and pricing power. Management signaled this is structural, not cyclical: a "potentially permanent reallocation" of wafer capacity from consumer to AI data center.

Three manufacturers (Micron, SK Hynix, Samsung) now control 95%+ of global DRAM and are shifting to multi-year strategic customer agreements (SCAs) at pre-negotiated prices rather than spot-market sales. HBM requires 3x+ the wafer capacity per bit vs conventional DRAM, compressing consumer supply. DRAM spot prices have surged 52% YTD; contract prices rose 90–95% Q1 2026 (largest quarterly increase ever recorded). Micron retired its Crucial consumer brand in Feb 2026 to focus exclusively on AI/enterprise. Consumer PC and smartphone units are expected to decline low-double-digits in calendar 2026 due to supply constraints, even as on-device AI (agentic workloads) drives 32GB minimum memory specs.

Micron's capex plans exceed $25B annually through 2027–2028 to deliver HBM4E (1-gamma node ramp in calendar 2027). Goldman Sachs pegs the 2026 DRAM supply-demand gap at 4.9%, the most severe shortage in 15 years. CEO Sanjay Mehrotra stated HBM and NAND demand will exceed available supply "for the foreseeable future." Wall Street lifted price targets sharply: UBS $1,625, TD Cowen $1,500, RBC $1,200; median $660.

For architects: This is not a memory cycle—it is a supply-driven structural shift. AI hyperscalers need HBM for inference (NVIDIA B200 = 192GB HBM3E per GPU; DGX systems = 1.5TB+ per node). Memory is pricing as infrastructure, not commodity, with 18–24 month fab lead times. Watch whether AI model efficiency (FLOPS/bit) improves faster than demand growth; if new fab capacity comes online (Micron NY, SK Hynix, Samsung) in 2027–2028 while demand softens, the upcycle inverts. For now, long-term contracts lock customers into premium pricing through 2027+.

Sources

Get the signal before the noise.