The AI industry placed four bets at unprecedented scale this year, each on ground that shifted. U.S. export controls targeted China’s chip supply, and the four largest tech companies committed more than $650 billion to infrastructure. Patent filings on generative AI technology doubled as application-layer startups compressed five-year revenue milestones into months. The collisions between these ambitions and the realities beneath them reshaped the industry in ways that none of their architects planned.
Real-world restrictions
Jensen Huang acknowledged in April that NVIDIA held zero market share in China, down from roughly 95% a few years earlier. The company took $4.5 billion in charges tied to export restrictions in a single quarter. Those restrictions, designed to slow Chinese AI development by restricting chip access, catalyzed the domestic chip ecosystem. Huawei’s Ascend 950PR entered mass production in March, delivering two petaflops on locally manufactured memory. The company’s AI chip revenue will reach an estimated $12 billion this year, a 60% increase over 2025. China’s total AI accelerator market grew to $30–35 billion from near zero in roughly three years. Morgan Stanley forecasts that 86% of China’s AI chip supply will be domestic by 2030, as independent software stacks replace CUDA and establish an ecosystem that functions without Western infrastructure at any layer.
Alphabet, Amazon, Meta, and Microsoft committed more than $650 billion to AI infrastructure this year. Seventy-five projects representing $130 billion in investment stalled in Q1 alone, nearly matching all of 2025. Electrical transformers, whose delivery times have extended to five years, account for less than 10% of project costs and prevent the other 90% from coming online. Data centers’ share of national electricity consumption will reach an estimated 12% by 2028, triple the 2023 level. OpenAI and Oracle abandoned a 600 MW expansion at Stargate in Abilene, Texas. Projects entering utility queues this year cannot realistically expect power before 2030.
Paper explosions
GenAI patent families nearly doubled between 2024 and 2025, to more than 37,800. China alone filed more in those two years than in its entire prior decade. SoftBank filed roughly 3,000 families, the largest single corporate portfolio globally. Only about 9% of these filings cross international borders. The U.S. Federal Circuit, the UK Supreme Court, and Japan’s Supreme Court have each ruled that AI cannot hold a patent as an inventor, requiring that a natural person demonstrate control of the inventive process. Claims on AI technology accumulate faster than the legal system that governs them can determine who holds them. The Patent Eligibility Restoration Act remains under congressional consideration, and Section 101 rejection reversal rates roughly tripled in late 2025.