Every story in this edition, newest first.
Chrome is downloading Gemini Nano — a roughly 4 GB on-device model called weights.bin — onto qualifying laptops without an explicit prompt. Google says it powers scam detection. Users say they were never asked. Both are true.
The new Kumamoto venture ties Sony's sensor lead to TSMC's foundry scale as automakers and robotics groups demand better machine vision.
Akamai says a leading U.S.-based frontier-model provider has committed $1.8 billion over seven years for cloud infrastructure services, and Reuters, citing Bloomberg, reports the customer is Anthropic — a sign that AI compute demand is broadening beyond the biggest cloud platforms.
Nintendo is lifting Switch 2 prices by as much as 20% in Japan and by $50 in both the United States and Canada, a rare early-cycle increase that arrives alongside strong sales and a ¥370 billion operating-profit forecast.
Microsoft's latest diffusion report says generative AI reached 17.8% of the world's working-age population in the first quarter of 2026, but the sharper signal is how quickly adoption is separating richer and poorer economies.
OpenAI has released two new transcription models and a steerable text-to-speech system, pairing broader language coverage with low per-minute pricing as the market for enterprise voice agents accelerates.
AMD's latest numbers suggest the global AI infrastructure boom is starting to reward more than one chip supplier, as data-center revenue jumps and hyperscalers widen deployments.
Alphabet, Amazon, Meta, and Microsoft are collectively pouring up to $700 billion into AI infrastructure in 2026 — more than the GDP of most nations — as Q1 results show cloud revenue surging but investors split on whether the returns will justify the spend.
Malicious versions of the widely-used deep learning framework stole cloud credentials, GitHub tokens, and CI/CD secrets across AWS, Azure, and GCP — and planted persistence hooks in Claude Code and VS Code for the first documented time in a production attack.
Amazon crushed Wall Street estimates with $2.78 EPS against $1.64 expected, powered by the fastest AWS growth in 15 quarters and a custom silicon business that CEO Andy Jassy says would be worth $50 billion as a standalone.
Meta delivered Q1 revenue of $43.4 billion and diluted EPS of $5.95, beating Wall Street, but raised full-year capex guidance to $130-135 billion as Llama 5 infrastructure and the Andromeda ad-ranking stack absorb spend faster than ad pricing can offset.
Apple reports fiscal Q2 results Thursday after the closing bell, with consensus at $109.7B revenue and $1.95 EPS as Tim Cook prepares his September 1 handoff to John Ternus and Siri's Gemini-backed overhaul anchors the AI reset.
Operating profit jumped 56 percent year-on-year as Samsung joined SK Hynix as a qualified HBM4 supplier to Nvidia, while foundry losses narrowed and the Galaxy S26 launch lifted mobile margins.
Redmond beat consensus on revenue and EPS but warned AI infrastructure spending will accelerate through fiscal 2027, drawing fresh scrutiny over the cloud capex cycle.
The round — anchored by an Amazon top-up and a new sovereign-led syndicate — values the Claude maker at roughly twice its mid-2025 mark and underscores how much of frontier AI's cost has migrated from talent to data center capacity.
The HKMA's go-slow strategy is the real story: Hong Kong picked quality over volume, and Beijing has a yuan-adjacent bridge it can quietly tolerate.
Stockholm-based streamer reports record paid additions and ad-tier growth, while a 30-language AI DJ rollout positions personalization as the next moat — even as Apple, Amazon, and YouTube intensify bundling pressure.
USDT issuance climbs past $215B while USDC clears $72B; ESMA's tighter euro-coin rules push regulated issuance toward Paris and Frankfurt as Tether's preliminary Q1 attestation guides to $4.8B in net profit.
New Delhi roughly doubles its empanelled compute pool inside 14 months; allocations weighted toward Nvidia H200 and B200, with a 15% carve-out for AMD and indigenous C-DAC silicon.
Both Korean memory giants confirmed this week that 12-high HBM4 stacks have cleared Nvidia qualification and are now flowing into TSMC's CoWoS-L lines for the Rubin GPU launch, formally opening a memory generation that will set AI-infrastructure economics through the rest of the decade.
The IIM-1 plant in Chitose ran its first EUV-patterned 2nm test wafers this month, putting Japan's national champion roughly 18 months behind TSMC but on track for its 2027 commercial-production deadline.
Mark Zuckerberg reports Wednesday with consensus expecting roughly $48B in revenue, but the real test is whether Family of Apps ad uplift from generative AI can absorb a Reality Labs loss inching toward $5B per quarter.
Brett Adcock's startup is the first humanoid-robot maker to break the pilot threshold in a U.S. auto plant, with a fleet large enough to actually shift labor economics and a Series E in motion at $40 billion.
Revenue rises to $20.1B and free cash flow burn narrows, but MAX 7 and MAX 10 certification remain Ortberg's unfinished business.
