The State of AI - 2026-08-26

AI’s expansion is increasingly constrained and shaped by infrastructure, capital and labor—not just model capability—while prominent calls for stronger safeguards are becoming more concrete.

By OMIKINA Editorial · No human review recorded

Published

Executive summary

The day’s developments point to a maturing AI competition in which compute capacity, power access, monetization and deployment economics are becoming strategic differentiators. Bill Gates has escalated his public warnings about AI-enabled biosecurity, cyber, labor and control risks, but his proposed remedies remain high-level and their political feasibility is untested. Meanwhile, Chinese model developers and semiconductor suppliers are investing heavily in infrastructure and advanced packaging, while OpenAI faces another infrastructure-leadership departure amid a much larger build-out.

Gates shifts from general concern to a policy campaign on AI risk

Bill Gates says AI has already crossed meaningful thresholds in biological-design capability, cyber capability, psychosocial effects, labor disruption and potentially system control. He argues that highly capable models could substitute for a broad range of well-defined white-collar work at lower cost, and calls for governments to act before unemployment and social distrust rise. These are Gates’s assessments rather than independently established measurements of current AI impact, and he acknowledges that his essay is not a complete solution.

His specific proposals—monitoring models able to design novel molecules, taxes on AI tokens and robots used to replace labor, “human-reserved” jobs, and a new international coordinating body—put a high-profile business and philanthropic figure behind interventionist AI policy. The immediate executive implication is heightened scrutiny of frontier-model access controls, workforce plans, and evidence that AI deployment produces benefits beyond cost reduction. International coordination, especially with China, remains the central implementation uncertainty.

Sources: Bill Gates says we’ve passed AI’s danger thresholds. Now what? · Bill Gates is deeply worried about AI, and he’s no longer staying quiet

China’s AI race is becoming an infrastructure and monetization contest

MiniMax reported first-half revenue of $116.6 million, up 283% year on year, with enterprise services supplying 63.4% of revenue. Its reported loss remained substantial, however, and its first-half revenue equaled roughly 32% of the Bloomberg-compiled full-year analyst forecast cited by the report. The company says enterprise customers contributed 80% of August annual recurring revenue and is preparing new models alongside a larger domestically sourced-chip share from the fourth quarter.

The figures illustrate both demand and the cost burden of commercial AI: enterprise adoption is improving revenue mix, but model providers still need scale and infrastructure efficiency to close losses. Separately, DeepSeek is reportedly seeking a 50 billion yuan round at a roughly 500 billion yuan pre-money valuation and considering a 2027 Shanghai listing; those details are sourced to unnamed people and have not been confirmed by the company. If completed, the financing would reinforce the ability of leading Chinese labs to fund compute expansion despite price competition.

Sources: MiniMax revenue surges 283% but remains behind pace to meet forecast amid crowded AI race · DeepSeek nears pre-IPO funding round as 2027 market debut takes shape: sources

Compute supply broadens beyond GPU procurement

China’s three largest outsourced packaging and testing firms announced more than 15 billion yuan in first-half expansion investments, directed in part at advanced packaging for AI accelerators. JCET plans a 7.8 billion yuan Shanghai facility whose first phase is scheduled for the second half of 2028; Tongfu and Huatian also reported or forecast strong profit growth and new capacity investments. The report attributes the expansion to AI demand, constrained advanced-packaging capacity and Beijing’s technology self-sufficiency agenda.

Advanced packaging is a potential bottleneck and value-capture point for AI hardware, particularly where chiplet and stacked-memory designs matter. In parallel, Bitcoin-mining operators are repurposing power and data-center assets for AI or high-performance computing: the BBC reports long-duration Anthropic compute agreements involving Riot Platforms and Bitdeer. This suggests the competitive supply of AI-ready facilities may increasingly come from converted industrial infrastructure, though such contracts also lock in capacity for years.

Sources: AI boom, profit surge fuel China’s ambitious chip assembly buildout · AI build out: AI gold rush draws crypto firms away from Bitcoin

Infrastructure organizations and human-data pipelines are under pressure

OpenAI’s head of data centers, Chris Malone, has left after the company reorganized its infrastructure organization, according to CNBC and an OpenAI statement. The departure follows several other senior exits as the company pursues infrastructure plans reportedly targeting about $600 billion in compute spending by 2030. OpenAI says it retains an experienced data-center team and clear leadership, but the change adds execution risk at a time of growing public and political resistance to data-center development.

Amazon will close Mechanical Turk on September 30, ending a long-running marketplace for data-labeling and other human-review tasks. The closure does not mean human data work is disappearing: the report identifies competing specialist platforms and continuing customer dependence on the service. But it does underline a reorganization of the human-in-the-loop supply chain, with potential migration, pricing and quality consequences for model builders and enterprises that still need annotation or review.

Sources: OpenAI data center chief Chris Malone is out, the latest in a string of executive exits · Amazon service Bezos once called 'artificial artificial intelligence' is shutting down

Watch next

Sources

  1. Bill Gates is deeply worried about AI, and he’s no longer staying quiet — The Verge ·
  2. MiniMax revenue surges 283% but remains behind pace to meet forecast amid crowded AI race — South China Morning Post · China Tech ·
  3. AI boom, profit surge fuel China’s ambitious chip assembly buildout — South China Morning Post · China Tech ·
  4. Raised on AI — MIT Technology Review AI ·
  5. AI models flub these intelligence tests. Can you fare any better? — MIT Technology Review AI ·
  6. Bill Gates says we’ve passed AI’s danger thresholds. Now what? — MIT Technology Review AI ·
  7. DeepSeek nears pre-IPO funding round as 2027 market debut takes shape: sources — South China Morning Post · China Tech ·
  8. Amazon service Bezos once called 'artificial artificial intelligence' is shutting down — CNBC Technology ·
  9. AI build out: AI gold rush draws crypto firms away from Bitcoin — BBC Technology ·
  10. OpenAI data center chief Chris Malone is out, the latest in a string of executive exits — CNBC Technology ·
  11. Blykalla submits second SMR plant application — World Nuclear News ·
  12. Apple announces new Mac Mini and Mac Studio models with AI upgrades — CNBC Technology ·
  13. Myanmar SMR site work 'could start in 2027' — World Nuclear News ·

Editorial standards · Corrections