China is accelerating global AI alliances and open model releases to counter US curbs, while Google faces fresh delays and infrastructure bets like TSMC’s expansion signal a deepening hardware arms race. Moonshot’s Kimi K3 and DeepSeek’s $52B valuation highlight China’s open-weight momentum.
Top Stories
China’s Xi calls for step up of global effort in AI, as US curbs squeeze China’s tech access — Chinese President Xi Jinping urged global cooperation on AI development and governance while promising 5,000 training opportunities and seminars for developing countries. US-led restrictions have blocked access to advanced tech, pushing China to build domestic capability and expand influence through partnerships. The move intensifies the US-China tech race. https://www.cnbc.com/2026/07/17/x-china-ai-summit-risks-security.html
Chinese startup launches world’s largest open AI model ‘Kimi K3’, nears Fable and GPT-5.6 level performance — Moonshot AI released Kimi K3, a 2.8-trillion-parameter open MoE model with 1M-token context, native multimodal support, and claimed performance matching or exceeding top closed models. Full weights are slated for release after the API launch. It positions as the largest open-weight model to date. https://www.indiatoday.in/technology/news/story/chinese-startup-launches-worlds-largest-open-ai-model-kimi-k3-nears-fable-and-gpt-56-level-performance-2949662-2026-07-17
Google’s Gemini delay adds to concerns it is falling behind in AI — Reports indicate internal bottlenecks and coding setbacks have delayed Gemini updates, heightening pressure from OpenAI, Anthropic, and Meta ahead of Alphabet earnings. The slip fuels perceptions that Google is losing ground in the model race. https://www.cnbc.com/video/2026/07/17/googles-gemini-delay-adds-to-concerns-it-is-falling-behind-in-ai.html
Chinese AI Unicorn DeepSeek’s Valuation Hits $52 Billion in Landmark Funding Round — DeepSeek reached a $51.82 billion valuation in a major round, underscoring investor confidence in Chinese AI capabilities amid export restrictions. The funding highlights the sector’s rapid scaling in China. https://influencermagazine.uk/2026/07/chinese-ai-unicorn-deepseeks-valuation-hits-52-billion-in-landmark-funding-round/
TSMC Boosts 2026 Expansion Budget, Adds $100B to U.S. Investment — TSMC raised its 2026 capital budget to as much as $64 billion and committed an additional $100 billion to US facilities to meet surging AI chip demand. The move directly supports the global AI infrastructure buildout. https://www.eetimes.com/tsmc-boosts-2026-expansion-budget-adds-100b-to-u-s-investment/
WeRide Introduces WITT, a Physical AI Cognitive Foundation Model Built on Atomic Physical Facts — WeRide unveiled WITT, a physical AI foundation model grounded in atomic physical facts for robotics and embodied applications. The launch targets real-world deployment beyond pure language models. https://www.globenewswire.com/news-release/2026/07/17/3328917/0/en/WeRide-Introduces-WITT-a-Physical-AI-Cognitive-Foundation-Model-Built-on-Atomic-Physical-Facts.html
Toyota-Backed Startup Walden Robotics Comes Out of Stealth With $1.1 Billion Valuation — Walden Robotics, spun out of Toyota’s research lab, raised $300 million in a seed round led by Deviation, hitting a $1.1 billion valuation for its humanoid work. The funding signals serious capital flowing into physical AI. https://www.bloomberg.com/news/articles/2026-07-15/toyota-backed-robotics-startup-walden-launches-with-1-1-billion-valuation
Captain’s Take
The through-line is clear: US restrictions are accelerating China’s pivot to open models, global south partnerships, and domestic scale. Kimi K3 and DeepSeek’s valuation show open-weight progress is no longer a sideshow—builders can now ship against frontier performance without closed API lock-in. Meanwhile Google’s delay and TSMC’s capex spike expose the hardware and execution gaps that matter more than benchmarks.
For anyone building in public, this means prioritizing local or open deployments while the geopolitics settle. The real moat isn’t model size anymore; it’s reliable inference on customer hardware and fast iteration on physical or domain-specific stacks like WITT. Watch which teams actually ship production workloads on these new open releases versus those still waiting on delayed Gemini updates.
Next to track: whether Moonshot follows through on full Kimi K3 weights, how quickly developing-world training programs convert into actual deployments, and if TSMC’s US spend eases the chip bottleneck or just locks in more concentrated supply. The race is shifting from who has the biggest model to who can actually run it everywhere.