LIVE · TUE, JUL 21, 2026 --:--:-- ET

Issue Nº 91 COST TOTAL $14871.87 ARTICLES TODAY 9 TOKENS TOTAL 9.57B

Running the wire

Breaking Moonshot Kimi K3 pauses subscriptions 48h post-launch as demand surges sixfold; 2.8T-param model strains GPU allocation Market Super Micro surges 15% on $60B order blitz; margin guidance raised to 15–17% on SpaceX XAI gigawatt build Market Semiconductor rebound: Micron +12%, Intel +8%, SMH ETF +4.5% as dip buyers return Market SK Hynix $26.5B Nasdaq debut, largest foreign IPO; stock up 13% day-one Chips TSMC pledges another $100B for US expansion; raises FY2026 revenue growth to 40%+ amid record Q2 $22B profit Funding Databricks raises strategic funding at $188bn valuation; Coatue-led round funds Unity AI Gateway and Lakebase expansion Funding Mistral closes €3bn Series D at €20bn valuation, backed by EU's Scaleup Fund Market GitHub reaches $100M open-source funding milestone; continued investment in maintainer support and community Market Goldman Sachs launches alternative investments platform; targets direct stakes in private AI unicorns pre-IPO Research Google launches Gemini 3.6 Flash with 17% token reduction, lower output pricing for agentic tasks Market OpenAI, Anthropic hit record lobbying: $3.17M combined in Q2 2026, up 23% QoQ Funding CuspAI raises $450M at $2.6B valuation for AI materials discovery; 45-company Foundry launches Breaking Iran claims fresh strike on AWS Bahrain data center with cruise missiles; ME-SOUTH-1 region offline since March, no Amazon updates Policy China weighs export controls on open-weight AI models, TSMC ban for Chinese chip designs; Alibaba, ByteDance, Zhipu consulted Chips TSMC commits additional $100B to Arizona, raising total US investment to $265B for 2nm and advanced packaging fabs Chips NVIDIA Vera Rubin NVL72 hits production with CoreWeave 10x throughput over GB200, draws Microsoft, Mistral, Tesla Chips NVIDIA Rubin GPU adds MoE descriptor management, 2x K-dimension throughput, 4x softmax for inference Breaking Google launches Gemini 3.6 Flash (17% fewer tokens), 3.5 Flash-Lite, and cyber-security model Chips NVIDIA Vera CPU ships with 88 Olympus cores, 1.5x agentic AI speedup over x86, starting with OpenAI Chips NVIDIA Spectrum-6 Ethernet hits 102.4 Tbps, deployed by CoreWeave, Microsoft, Nebius for gigascale AI Breaking Moonshot Kimi K3 pauses subscriptions 48h post-launch as demand surges sixfold; 2.8T-param model strains GPU allocation Market Super Micro surges 15% on $60B order blitz; margin guidance raised to 15–17% on SpaceX XAI gigawatt build Market Semiconductor rebound: Micron +12%, Intel +8%, SMH ETF +4.5% as dip buyers return Market SK Hynix $26.5B Nasdaq debut, largest foreign IPO; stock up 13% day-one Chips TSMC pledges another $100B for US expansion; raises FY2026 revenue growth to 40%+ amid record Q2 $22B profit Funding Databricks raises strategic funding at $188bn valuation; Coatue-led round funds Unity AI Gateway and Lakebase expansion Funding Mistral closes €3bn Series D at €20bn valuation, backed by EU's Scaleup Fund Market GitHub reaches $100M open-source funding milestone; continued investment in maintainer support and community Market Goldman Sachs launches alternative investments platform; targets direct stakes in private AI unicorns pre-IPO Research Google launches Gemini 3.6 Flash with 17% token reduction, lower output pricing for agentic tasks Market OpenAI, Anthropic hit record lobbying: $3.17M combined in Q2 2026, up 23% QoQ Funding CuspAI raises $450M at $2.6B valuation for AI materials discovery; 45-company Foundry launches Breaking Iran claims fresh strike on AWS Bahrain data center with cruise missiles; ME-SOUTH-1 region offline since March, no Amazon updates Policy China weighs export controls on open-weight AI models, TSMC ban for Chinese chip designs; Alibaba, ByteDance, Zhipu consulted Chips TSMC commits additional $100B to Arizona, raising total US investment to $265B for 2nm and advanced packaging fabs Chips NVIDIA Vera Rubin NVL72 hits production with CoreWeave 10x throughput over GB200, draws Microsoft, Mistral, Tesla Chips NVIDIA Rubin GPU adds MoE descriptor management, 2x K-dimension throughput, 4x softmax for inference Breaking Google launches Gemini 3.6 Flash (17% fewer tokens), 3.5 Flash-Lite, and cyber-security model Chips NVIDIA Vera CPU ships with 88 Olympus cores, 1.5x agentic AI speedup over x86, starting with OpenAI Chips NVIDIA Spectrum-6 Ethernet hits 102.4 Tbps, deployed by CoreWeave, Microsoft, Nebius for gigascale AI

Breaking Tuesday, July 21, 2026 at 03:32 PM

Google launches Gemini 3.6 Flash (17% fewer tokens), 3.5 Flash-Lite, and cyber-security model

Google released three new Gemini models on Tuesday as it prepares for earnings and faces intensifying competition from Chinese AI rivals. Gemini 3.6 Flash improves coding, multimodal, and knowledge-work performance while reducing output token usage by 17% compared to 3.5 Flash and lowering cost per output token to $1.50/M input, $7.50/M output. The model shows 49% accuracy on DeepSWE code editing (vs. 37% for 3.5 Flash) and 83.0% on OSWorld computer-use tasks (vs. 78.4% for prior generation).

Gemini 3.5 Flash-Lite targets high-volume, low-latency workloads at just $0.30/M input tokens and $2.50/M output tokens, running at 350 output tokens per second according to Artificial Analysis—the fastest in the 3.5 family. It significantly outperforms older Flash-Lite generations on agentic workflows and includes built-in computer-use capabilities. Gemini 3.5 Flash Cyber, launched in limited pilot for governments and trusted partners, is a specialized model paired with CodeMender, Google's code security agent, designed to detect and patch software vulnerabilities at lower cost per token than larger models.

Google's broader roadmap signals confidence: 3.5 Pro is in partner testing ahead of broader release, and the company has begun its largest pre-training run yet for Gemini 4. The launch responds to Chinese rivals' momentum—Moonshot AI's Kimi K3 hit capacity constraints from demand, and Alibaba is teasing Qwen 3.8 Max. For practitioners, 3.6 Flash's 17% token reduction and improved agentic performance at lower cost per task make it the clear choice for multi-turn agent systems and document analysis at scale.

Sources