Sold Out Through 2027

Capability & Integration

• Microsoft announced a family of seven in-house models at Build 2026 (June 2), headlined by MAI-Thinking-1, a 35B-active-parameter sparse MoE reasoning model with a 256k-token context window. Microsoft says it was trained entirely on commercially licensed data without distilling from any third-party model — a notable framing as large tech firms move to reduce dependency on external labs. Microsoft claims MAI-Thinking-1 scores 97.0% on AIME 2025 and is preferred over Claude Sonnet 4.6 in blind side-by-side evaluations; those comparisons are self-reported.
• MAI-Code-1-Flash, a 5B-active-parameter companion model, began rolling out to all GitHub Copilot tiers on June 2 and is positioned as a low-latency coding assistant deeply integrated into the Microsoft stack.
• Anthropic expanded Project Glasswing on June 3 to approximately 150 organizations across 15+ countries, giving partners early access to Claude Mythos Preview to hunt and patch critical software vulnerabilities. Partners have collectively surfaced more than 10,000 high- or critical-severity security flaws across major operating systems and browsers. As of June 9, Anthropic shipped Claude Fable 5 and Claude Mythos 5, making a safeguarded Mythos-class model generally available for the first time.
• NVIDIA unveiled Ising, a family of open-source models built to accelerate quantum computing error-correction decoding, claiming 2.5x speed and 3x accuracy improvements over traditional approaches.

Robotics

• Figure AI reports its BotQ factory is producing Figure 03 at one robot per hour, with units running in BMW’s Spartanburg manufacturing plant. This is a limited initial factory deployment, not a broad commercial rollout.
• Boston Dynamics’ electric Atlas is shipping initial units to committed early customers Hyundai and Google DeepMind, following its CES 2026 production announcement. The 56-degree-of-freedom, 50kg-lift-capacity platform is in controlled deployment, not general availability.
• Tesla Optimus continues internal deployment within Tesla’s own factories; Fremont volume production is reportedly planned for late July or August 2026. No third-party commercial deployments have been confirmed.
• Automate 2026 (Chicago, June 22–25) will feature Universal Robots’ UR AI Trainer — an imitation-learning platform co-developed with Scale AI that lets factory workers physically guide robot arms through tasks to train Vision-Language-Action models for direct factory deployment. Described by Teradyne as “production-ready,” though independent assessment of that claim awaits post-show reporting.

Hardware & Supply Chain

• At TSMC’s June 4 annual shareholders meeting in Hsinchu, CEO C.C. Wei warned that global chip supply will “lag behind” AI-driven demand “for years,” with advanced-node capacity sold out through at least 2027 and demand running roughly 25–30% above available capacity. TSMC reaffirmed 30%+ revenue growth for 2026 in USD terms.
• NVIDIA’s fiscal year 2026 annual report shows total revenue of $215.9 billion, up 65% year over year, with data center revenue reaching $194 billion. The company disclosed it is “effectively foreclosed” from China’s data center market due to export controls.
• The Bureau of Industry and Security shifted the export license review policy for NVIDIA H200 chips from “presumption of denial” to case-by-case review for China in January 2026, under conditions including a 25% tariff, a 50% volume cap, and third-party testing requirements. The more advanced B200 and Rubin architectures remain prohibited.
• Google has reportedly placed an order with Intel to manufacture more than 3 million in-house TPUs in 2028, as major customers diversify production routes beyond TSMC’s constrained capacity. AMD announced more than $10 billion in Taiwan investment at COMPUTEX 2026 (June 2–5), focused on advanced packaging.

Environmental & Cultural Impact

• A UN University report released in early June projects that AI data centers could consume 945 terawatt-hours of electricity annually by 2030 — nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, countries home to more than 650 million people.
• The same report estimates AI-related water consumption could equal the basic annual domestic water needs of 1.3 billion people by the end of the decade, driven by cooling systems and the water footprint embedded in electricity generation.
• AI infrastructure is projected to generate up to 2.5 million tonnes of e-waste annually by 2030; the report notes the disposal burden is likely to fall disproportionately on lower-income countries with limited safe processing capacity.
• In 2026, lawmakers in more than 30 U.S. states have introduced over 300 bills addressing data center development, including moratoriums, tax incentives, and grid-impact requirements.

AI in the Wild

A fan-made AI video depicting Los Angeles mayoral candidate Spencer Pratt as Batman fighting crime in a dystopian version of the city — with incumbent Karen Bass cast as the Joker — accumulated more than 5 million views on X ahead of the June 3 primary. Pratt actively reshared the content; its AI origin was confirmed and widely reported by Time, NBC News, and ABC7. Pratt did not advance to the runoff (Bass and City Councilmember Nithya Raman did). The episode is a clean data point: AI-generated political content can achieve significant reach without translating to votes, and in this case, the candidate’s actual grievance — his house burned in the Palisades fire — was more legible than the synthetic imagery amplifying it.

Takeaway