LIVE · MON, JUN 22, 2026 --:--:-- ET
Issue Nº 62 COST TOTAL $14474.15 ARTICLES TODAY 5 TOKENS TOTAL 9.06B
aiexpert
Running the wire
Breaking Samsung deploys ChatGPT Enterprise & Codex to all 300K+ employees globally Chips NVIDIA Rubin 100% liquid-cooled, eliminates need for data center chillers Market NVIDIA Projects $1T Hyperscaler Capex in 2027; Vera Rubin Supply Constrained Through Lifecycle Chips TSMC 2nm Achieves 70% Yield, Targets 140K Wafers/Month by Late 2026 Amid CoWoS Bottleneck Policy China Adds US Rare Earth Producers to Export Control List; Supply Chain Escalation Funding Snowflake, Anthropic expand $200M partnership to deploy Claude AI agents at scale across 12,600 enterprises Breaking Samsung reverses 3-year ChatGPT ban, deploys enterprise OpenAI, Gemini, Claude to 300K+ employees Breaking Anthropic Claude handles 95% of internal analytics—demonstrating enterprise scale Breaking Samsung reverses 2023 ChatGPT ban, deploys enterprise OpenAI/Google/Anthropic AI companywide Breaking Apple Core AI ships in-device LLM framework supporting up to 70B-parameter models Market Meta boosts Texas AI datacenter investment to $10B; targets 1 GW by 2028 Chips Amazon custom silicon hits $20B ARR; Trainium3 nearly sold out, Trainium4 largely reserved Funding Anthropic signs multi-GW Google + Broadcom compute deal; $30B+ ARR, 1,000+ $1M+ customers Funding OpenAI files confidential S-1 for IPO targeting $1T+ valuation, Goldman Sachs and Morgan Stanley lead Funding Anthropic closes $65B Series H at $965B, overtaking OpenAI as most valuable private AI company Funding Generalist AI raises $400M at $2B valuation for embodied robot foundation models Chips Apple Launches Core AI Framework for On-Device LLM Inference on Apple Silicon Breaking Samsung Reverses 2023 AI Ban: Rolls Out ChatGPT Enterprise Globally Breaking Noam Shazeer, Transformer co-author, leaves Google for OpenAI; less than 2 years after $2.7B acqui-hire Breaking OpenAI acquires Astral; adds uv, Ruff to Codex as coding-assistant arms race intensifies Breaking Samsung deploys ChatGPT Enterprise & Codex to all 300K+ employees globally Chips NVIDIA Rubin 100% liquid-cooled, eliminates need for data center chillers Market NVIDIA Projects $1T Hyperscaler Capex in 2027; Vera Rubin Supply Constrained Through Lifecycle Chips TSMC 2nm Achieves 70% Yield, Targets 140K Wafers/Month by Late 2026 Amid CoWoS Bottleneck Policy China Adds US Rare Earth Producers to Export Control List; Supply Chain Escalation Funding Snowflake, Anthropic expand $200M partnership to deploy Claude AI agents at scale across 12,600 enterprises Breaking Samsung reverses 3-year ChatGPT ban, deploys enterprise OpenAI, Gemini, Claude to 300K+ employees Breaking Anthropic Claude handles 95% of internal analytics—demonstrating enterprise scale Breaking Samsung reverses 2023 ChatGPT ban, deploys enterprise OpenAI/Google/Anthropic AI companywide Breaking Apple Core AI ships in-device LLM framework supporting up to 70B-parameter models Market Meta boosts Texas AI datacenter investment to $10B; targets 1 GW by 2028 Chips Amazon custom silicon hits $20B ARR; Trainium3 nearly sold out, Trainium4 largely reserved Funding Anthropic signs multi-GW Google + Broadcom compute deal; $30B+ ARR, 1,000+ $1M+ customers Funding OpenAI files confidential S-1 for IPO targeting $1T+ valuation, Goldman Sachs and Morgan Stanley lead Funding Anthropic closes $65B Series H at $965B, overtaking OpenAI as most valuable private AI company Funding Generalist AI raises $400M at $2B valuation for embodied robot foundation models Chips Apple Launches Core AI Framework for On-Device LLM Inference on Apple Silicon Breaking Samsung Reverses 2023 AI Ban: Rolls Out ChatGPT Enterprise Globally Breaking Noam Shazeer, Transformer co-author, leaves Google for OpenAI; less than 2 years after $2.7B acqui-hire Breaking OpenAI acquires Astral; adds uv, Ruff to Codex as coding-assistant arms race intensifies
Chips

TSMC 2nm Achieves 70% Yield, Targets 140K Wafers/Month by Late 2026 Amid CoWoS Bottleneck

TSMC's next-generation 2nm process, which entered volume production in Q4 2025 using Gate-All-Around (GAA) nanosheet transistors, has already achieved 70% initial yields—a strong benchmark for a cutting-edge node. The company targets an 80% yield rate for the enhanced N2P variant launching in H2 2026. This rapid maturation is critical: the company expects to scale 2nm production to over 100,000 wafers monthly by end-2026, with some estimates reaching 140,000 wafers/month, nearly matching current 3nm capacity. Fab 22 in Kaohsiung is the primary production hub, with three additional 2nm fabs planned in Taiwan to support long-term demand.

The N2 node delivers a full-node performance-per-watt improvement over 3nm: 10–15% speed gains at the same power, or 25–30% power reduction at the same speed, plus 15%+ transistor density increase. These PPA gains are particularly valuable for AI inference chips where memory bandwidth and thermal efficiency matter. Apple has locked in over half of TSMC's initial 2nm output for its A20 and M6 chips, but the real story is data center: NVIDIA, Google, Amazon, and other hyperscalers are competing for capacity to deploy 2nm chips into their next-generation AI accelerators throughout 2026–2027.

However, the bottleneck is not logic nodes—it is packaging. TSMC's CoWoS (Chip-on-Wafer-on-Substrate) 2.5D packaging, essential for bonding compute dies to HBM stacks, has become the critical constraint. Demand is doubling from ~35,000 wafers/month in 2024 to 70,000+ by end-2025, with further growth into 2026. TSMC is dedicating 10–20% of 2025–2026 capex to advanced packaging and test facilities, plus launching two dedicated packaging plants in Arizona. For architects designing next-generation AI systems, the lesson is clear: 2nm logic availability is no longer the gating factor. Access to CoWoS capacity is now the real supply bottleneck limiting when your H200- or Blackwell-class accelerator chiplets can ship in volume.

Sources