LIVE · MON, JUL 27, 2026 --:--:-- ET

Issue Nº 97 COST TOTAL $14967.65 ARTICLES TODAY 0 TOKENS TOTAL 9.69B

Running the wire

Chips NVIDIA deploys Vera CPU across EDA design workflows; Cadence & Synopsys show 1.5x gains Chips Samsung Foundry breaks even: 2nm yields hit 55-60%; Tesla contract signals customers are willing to diversify Breaking OpenAI ships Presence: Enterprise agent platform with Codex-driven improvements; resolves 75% of support issues without human handoff Chips NVIDIA Vera Rubin ramps full production; 10x token throughput vs Blackwell, Q3 2026 deployments Chips NVIDIA shelves RTX 50 Super indefinitely; GDDR7 shortage forces gaming GPU cuts for AI demand Research Google ships Gemini 3.6 Flash at 17% fewer tokens, $7.50 output price; 3.5 Pro still delayed Research Physicists demonstrate reservoir computing with 400 oscillating particles in liquid; F1 0.90 on anomaly detection Funding Anthropic begins IPO investor roadshow; October 2026 Nasdaq debut targeted at $965B–$1T valuation Market Server DRAM prices set to rise 13–18% in Q3 2026 as AI demand outpaces supply; shortage extends into 2027 Chips TSMC commits $265B to Arizona, adds advanced packaging; US still ships chips to Taiwan for CoWoS Market Kioxia plunges 16% as memory sector rout spreads; market cap halved in a month Market AI cluster power, not GPUs, is now the bottleneck; grid approval queues hit 24–36 months Chips TSMC Arizona Fab 2 equipment install begins Q3 2026; 3nm production targeted for 2027 Chips TSMC A14 progress hits ~90% yield three months ahead of N2 schedule; mass production target 2H 2028 Funding Chai Discovery closes $400M Series C at $3.8B; AI antibody design now used by Lilly, Pfizer, Novartis Funding Fireworks AI raises $1.5B Series D at $17.5B; surpasses $1B ARR with 40T+ tokens/day, 95% custom models Breaking Neo emerges from stealth with $100M to control agentic AI; Gartner says 40% of enterprise apps will be agentic by EOY Policy Hassabis, Nadella, Altman converge on AI regulation; three rival CEOs back independent model review Policy FCC expands submarine cable licensing to SLTE; shields U.S. tech giants, blocks Chinese access Breaking MCP 2026-07-28 spec removes sessions, adds OAuth hardening; ships July 28 with breaking changes Chips NVIDIA deploys Vera CPU across EDA design workflows; Cadence & Synopsys show 1.5x gains Chips Samsung Foundry breaks even: 2nm yields hit 55-60%; Tesla contract signals customers are willing to diversify Breaking OpenAI ships Presence: Enterprise agent platform with Codex-driven improvements; resolves 75% of support issues without human handoff Chips NVIDIA Vera Rubin ramps full production; 10x token throughput vs Blackwell, Q3 2026 deployments Chips NVIDIA shelves RTX 50 Super indefinitely; GDDR7 shortage forces gaming GPU cuts for AI demand Research Google ships Gemini 3.6 Flash at 17% fewer tokens, $7.50 output price; 3.5 Pro still delayed Research Physicists demonstrate reservoir computing with 400 oscillating particles in liquid; F1 0.90 on anomaly detection Funding Anthropic begins IPO investor roadshow; October 2026 Nasdaq debut targeted at $965B–$1T valuation Market Server DRAM prices set to rise 13–18% in Q3 2026 as AI demand outpaces supply; shortage extends into 2027 Chips TSMC commits $265B to Arizona, adds advanced packaging; US still ships chips to Taiwan for CoWoS Market Kioxia plunges 16% as memory sector rout spreads; market cap halved in a month Market AI cluster power, not GPUs, is now the bottleneck; grid approval queues hit 24–36 months Chips TSMC Arizona Fab 2 equipment install begins Q3 2026; 3nm production targeted for 2027 Chips TSMC A14 progress hits ~90% yield three months ahead of N2 schedule; mass production target 2H 2028 Funding Chai Discovery closes $400M Series C at $3.8B; AI antibody design now used by Lilly, Pfizer, Novartis Funding Fireworks AI raises $1.5B Series D at $17.5B; surpasses $1B ARR with 40T+ tokens/day, 95% custom models Breaking Neo emerges from stealth with $100M to control agentic AI; Gartner says 40% of enterprise apps will be agentic by EOY Policy Hassabis, Nadella, Altman converge on AI regulation; three rival CEOs back independent model review Policy FCC expands submarine cable licensing to SLTE; shields U.S. tech giants, blocks Chinese access Breaking MCP 2026-07-28 spec removes sessions, adds OAuth hardening; ships July 28 with breaking changes

Chips Monday, July 27, 2026 at 12:03 AM

NVIDIA Vera Rubin ramps full production; 10x token throughput vs Blackwell, Q3 2026 deployments

NVIDIA announced on July 21 that the Vera Rubin platform is ramping into full production, with initial deployments underway at major cloud providers including Microsoft Azure, Google Cloud, AWS, and Oracle Cloud Infrastructure. Vera Rubin pairs an 88-core Arm-based Vera CPU with Rubin GPUs and NVLink 6 networking in a rack-scale architecture, delivering 3.5x training performance over Blackwell and 10x agent token throughput at scale. The NVL72 configuration bundles 72 Rubin GPUs, each with 288GB of HBM4 memory, into a single system optimized for agentic AI, reasoning, and mixture-of-experts inference workloads.

TSMC has begun mass production of Vera Rubin chips on the 3nm process, with Foxconn, Quanta, and Wistron ramping full-scale ODM production in the second half of 2026. SK Hynix's 192GB SOCAMM2 HBM memory, offering 2x the bandwidth of traditional RDIMM, has entered mass production. Volume shipments are on track for Q3 2026, with Microsoft committing the Vera Rubin NVL72 to its Fairwater AI superfactory sites and CoreWeave offering Vera-based instances in its AI cloud platform. Jensen Huang stated the platform is designed to deliver 40 million times current computing power over the next decade, with a potential market scale reaching trillions of dollars.

For architects planning 2026–2027 data center builds, Vera Rubin marks the inflection from Blackwell-era capacity constraints to genuine supply availability. The platform is 100% liquid-cooled—direct-to-chip liquid cooling at 45-degree hot water is now a baseline requirement, not an option. Installation time dropped from two days to two hours per rack. Teams must validate power, cooling, and networking infrastructure now; the window to commit orders for Q3 2026 availability is closing. This is not a modest incremental update; it is a full POD-scale redesign.

Sources