China's AI sector saw 1,203 equity financing events totaling 307.68 billion yuan in the first half of 2026, already exceeding the full-year 2025 total. DeepSeek's 51 billion yuan Series A round in June was the single largest deal and drove a monthly surge to over 100 billion yuan. The top four cities—Beijing, Hangzhou, Shanghai, and Shenzhen—accounted for 74% of deals and 86% of total funding, with Beijing leading at 95.5 billion yuan. By sub-sector, large model companies captured half of all capital at 159.85 billion yuan, followed by AI-embodied intelligence (90.64 billion yuan) and AIGC applications (59.61 billion yuan). Early-stage investments heavily favored world-model startups, while no pure large model ventures appeared among the top 10 early-stage deals, signaling that the large model startup window is closing.
On June 30, 2026, Anthropic launched Claude Science, a life-sciences AI platform integrating biology-specialized Claude Opus 4.6 and 4.8 models, a physical wet lab for experimental data feedback, and the acquisition of startup CoefficientBio. The initiative aims to slash overall life-science R&D cycles to one-tenth. Training on biological data is challenging because there is often no single ground-truth answer; the wet lab enables closed-loop model improvement. The platform has attracted capital inflows, with Amazon evaluating pricing, while Anthropic reports annualized revenue of $30 billion and is developing safety classifiers to mitigate dual-use risks.
OpenAI has reportedly developed a new system optimization that can cut model inference costs by more than half, reducing GPU requirements from tens of thousands to just hundreds. The optimization is believed to focus on KV cache efficiency, a strategy DeepSeek previously pioneered with its Multi-head Latent Attention (MLA) and caching discounts. This software push is complemented by hardware efforts: the Jalapeño inference chip co-designed with Broadcom, and a $10 billion-plus deal with Cerebras for wafer-scale inference. OpenAI's 2025 revenue reached $13.07 billion but operating losses hit $20.9 billion, with cloud bills alone exceeding $17.2 billion, making deep cost cuts essential for its delayed IPO. API gross margin improved to 39% in Q1 2026, with a target of 52% by year-end.
In early 2026, major AI companies began imposing mutual restrictions. Google limited Meta's Gemini access due to compute capacity shortages, delaying internal Meta projects. Google banned most employees from using Claude Code and Codex, while Microsoft revoked internal Claude Code licenses and later restricted Claude Fable5 over data retention concerns. Meta also restricted the use of Claude and Codex in model building, fearing that model outputs could enter its training pipeline and trigger distillation or legal risks. These moves mark the end of AI tools' free-trial era, as resource scarcity, data security, and asset protection became the three key defensive barriers.
The AI boom is causing a massive memory chip shortage, dubbed 'memory doomsday,' with prices forecast to rise 40–50% in Q3 2026 and continue through 2028. Upstream makers SK Hynix, Samsung, and Micron are shifting capacity toward high-margin HBM but face risks from over-reliance on Nvidia orders and internal competition; Micron has locked customers into aggressive 3–5 year 'strategic customer agreements' with take-or-pay clauses. Midstream module makers such as Jiangbolon, Biwin, and Demingli report record profits from high inventory and rising prices, while distributors endure razor-thin margins. Downstream, terminal companies are raising prices—Apple, Microsoft, and Nintendo have all announced increases—and GoPro warned of bankruptcy due to reduced memory supply. Consumers are coping by recycling old phones, building DIY PCs, and sharing cost-saving guides on GitHub.
Wired revealed that Meta ran a covert project codenamed ‘Cannes’, where hundreds of contractors from outsourcing firm Covalen created fake minor accounts and systematically sent tens of thousands of harmful prompts to ChatGPT, Gemini, and Character.AI chatbots. One internal document alone listed 3,748 malicious prompts, including at least 239 sexually explicit prompts involving minors, and over 45,000 high-risk prompts were submitted in a single August 2025 test round. The prompts aggressively probed safety filters around self-harm, suicide, eating disorders, and child endangerment, without the target companies’ knowledge or consent. Meta defended the activity as legitimate ‘comprehensive AI safety benchmarking’, while Character.AI, OpenAI, and Google stated they had not authorized such testing. The exposure sparked criticism that Meta weaponized AI safety as an anti-competitive tool under the guise of responsible testing.