AI Industry Pulse

Nvidia six-die superchip platform full production; Vera Rubin architecture with 3x faster attention cores. Google IO 2026 confirms agentic shift with TPU split, Gemini 3.5 Flash, Vertex AI orchestration. Google-Blackstone AI cloud company with $5B initial investment targeting 500 MW by 2027. Blackstone invests in Neysa Networks ($1.2B, 20K GPUs in India). Memory now 63% of AI chip cost; HBM tightness; Micron sold out through 2027. Data center AI chip market $12.45B (2025) to $50.31B (2034). TSMC CoWoS doubling; Nvidia Vera/Rubin ramp; NVIDIA Q1 FY2027 $81.6B rev (+85% YoY), stock dip on priced-in beat; non-hyperscaler business growing faster. Google-Blackstone $25B TPU; Amazon-Anthropic $25B; Cerebras IPO; Broadcom AI networking. $700B capex amid power/edge ramps. Applied Digital pivot from crypto to AI data centers highlights power/location bottlenecks. Exa's $250M raise signals AI-native search infrastructure as critical layer for agentic AI. Nvidia Q1 earnings show deceleration but market expansion; TPU v7 vs AWS Trainium 3 competition intensifies. NVIDIA $2B Marvell investment (silicon photonics, custom XPUs); AMD plans $10B+ Taiwan AI market investment (2nm Venice CPUs). Microsoft launches Azure Linux 4.0 for enterprise AI workloads; GPU capacity as industrial capacity narrative strengthens with US-Germany gap. Qualcomm jumps 11% on AI chip rotation; TS Lombard calls semiconductor hypercycle with 80% YoY sales growth; Google Cloud Next 2026 announces Gemini Enterprise Agent Platform and Agentic Data Cloud. Custom AI chips (ASICs) growing 44.6% vs GPU 16.1%, Midjourney 65% cost reduction with TPUs; Dell on-premise AI bet paying off (5,000 customers, $43B backlog, 320x token consumption); Google Virgo network scaling 134K TPU chips with 47 petabits/sec; ePlus launches private AI infrastructure managed service; HDD object storage validated for AI at scale (420 GB/s reads). OpenRouter more than doubles valuation to $1.3B (100T tokens/month, 5x growth) confirming multi-model gateway thesis; Qumulo Cloud AI Accelerator addresses GPU utilization crisis (5% utilization stat) by fixing data staging bottlenecks. New: SK Hynix hits $1T market cap, underscoring HBM dominance and memory cost share. Cerebras CEO post-IPO interview covers memory shortages, enterprise adoption blockers, and $20B OpenAI deal. Marvell earnings signal AI infrastructure cycle winners/losers. Baikal announces Russian AI chip challenging Nvidia, facing manufacturing hurdles. Qualcomm designing custom ASICs for ByteDance to sidestep US export controls. Micron hits $1T market cap. Broadcom launches first 50G PON edge AI gateway SoC. Supply chain constraints: semiconductor sales $791.7B, PCB laminate lead times doubled, prepreg pricing up 90%. Latest: Broadcom AI revenue up 106% YoY, targeting $100B annual chip revenue by 2027; expanded Google TPU deal and OpenAI first-gen XPU in 2027. FuriosaAI partners with Broadcom on 2nm HBM4 inference chip (2028). Baseten, Fireworks AI, Modal Labs raises signal inference infrastructure gold rush. China adds nine homegrown AI chips to government procurement list. New: Nvidia CEO pledges $150B annual investment in Taiwan; KLA gaining share as AI chip complexity drives yield costs. Nvidia plans 4,000 new jobs in Taiwan. Snowflake signs $6B AWS deal for Graviton CPUs, signaling AI inference driving CPU demand. OpenRouter's $1.3B valuation reinforces multi-model inference infrastructure. New: Micron starts building $100B NY memory fab (first production 2030); Broadcom faces VMware backlash despite 106% AI revenue growth; Elon Musk warns on TSMC dependency, highlights $16.5B Samsung deal and power bottlenecks.