President Trump and Canadian Prime Minister Mark Carney failed to reach a trade deal, leaving tariffs and strained U.S.-Canada relations unresolved. Technology’s geopolitical reach widened as Britain gained access to Ukrainian combat data used to train AI systems. In markets, Nvidia warned that some Vera Rubin and Grace Blackwell systems will cost at least 15% more in 2027, SoftBank prepared a record $6.3B Japanese retail-bond sale to fund its OpenAI commitments, and Oura explored a September IPO at a $16B+ valuation. Leadership and competition are also shifting: Tim Cook is set to step down Sept. 1, while AMD again crossed 30% of the client-CPU market. Meta agreed to pay up to $17.1B and strengthen child protections, Uber introduced optional livestreams for teen rides, and 28% of Americans now use chatbots for quick health information. Meanwhile, Unitree’s superhuman-speed robot ambitions underscore China’s physical-AI momentum.
Top 5 AI Highlights
[1] 🤝 Nvidia agreed to acquire Hugging Face for $12.9B, giving the chipmaker control of a central AI-model repository. The deal follows Hugging Face’s unveiling of Microduck, a $400, 1.7-lb bipedal robot; separately, Nvidia is discussing an investment in Perplexity at a $30B+ valuation after considering a technology-licensing deal. More from The Information, Bloomberg, and The Information.
[3] 📈 Nvidia reported Q2 revenue up 106% YoY to $96.22B, including Data Center revenue up 117% to $89B and net income up 126% to $59.7B; it forecast Q3 revenue of $108B versus $104B expected and sees FY2028 revenue growth of about 70%. More from Reuters and Nvidia Newsroom.
[2] 💰 Anthropic’s bankers said it could raise $100B+ in an IPO at a potential $2T valuation and pitch a $30T+ revenue opportunity. The company also agreed to pay Nscale $45B over six years for roughly 460MW at a West Virginia Vera Rubin data center, while Salesforce and Anthropic unveiled Claudeforce, a Claude plugin connecting Salesforce data and workflows through 37 prebuilt sales skills. More from The New York Times, Wall Street Journal, Bloomberg, and VentureBeat.
[4] ⚙️ OpenAI says its Jalapeño chip delivered 1.5x-1.9x more AI work per watt and 1.7x-3.6x lower latency than Nvidia chips across leading open-weight models. Limited deployment is expected by year-end, with broader rollout in 2027. More from The Verge and TechCrunch.
[5] 🌐 Anthropic and OpenAI increasingly control the world’s AI compute, SemiAnalysis founder Dylan Patel argues, as he examines $11T in AI capex from 2024 to 2029, China’s computing position, and the economics of the next infrastructure buildout. More from Dwarkesh Podcast.
✨Open Source AI Development
[1] 🧠 Nvidia plans to use its $6B Poolside licensing deal to build an open-weight model aimed at competing with Chinese systems such as DeepSeek and Kimi. More from The Wall Street Journal.
[2] ⚡ Z.ai released the MIT-licensed, open-weight GLM-5.3-Flash and confirmed it was the model tested as Ox Alpha, which topped OpenRouter’s leaderboard. The 320B-parameter multimodal model scored 84.3 on Terminal-Bench 2.1 and 63.4 on DeepSWE, versus GLM-5.2’s 46.2 on DeepSWE, at one-tenth the price. More from Z.ai, Bloomberg, and Hugging Face.
[3] ☁️ Moonshot AI is in early talks with Microsoft, Amazon, and Google to host its open-weight Kimi K3 model under revenue-sharing agreements. The Chinese lab is seeking as much as a 30% share, a sign that open-model developers are testing cloud distribution as a new monetization path. More from Reuters.
[4] 📊 DeepSeek is set to raise $7.4B at a $74B valuation to fund R&D and computing infrastructure. The Chinese lab generated $70.7M in revenue and posted a $106M net loss in the first seven months of 2026, roughly 10x its full-year 2025 revenue, and is weighing a Shanghai IPO as soon as 2027. More from Wall Street Journal, The Information, and South China Morning Post.
[5] 🛡️ Chinese state-linked groups are increasingly using open-weight models such as Kimi K3 and DeepSeek in cyberattacks, showing how broadly accessible models are also expanding offensive capabilities. More from Bloomberg.
[5] 🔀 Ramp launched Router, the AI model-routing service it has used internally for three years, in the US and will make it free through 2026. It offers models from OpenAI, Anthropic, DeepSeek, Moonshot, MiniMax, Nvidia, xAI, and Z.ai, plus strategies that route requests according to customer preferences. More from TechCrunch.
Founder’s Corner
Top Model by Task
Source: X
What We Read This Week
- Google DeepMind is piloting double-blind AI evaluations that keep external tests in a cryptographic box to reduce benchmark contamination and protect intellectual property. (Google DeepMind)
- Tang Jie and Yang Zhilin, once teacher and student at Tsinghua University, illustrate how China’s AI gains reflect years of scientific development rather than a sudden breakthrough. (Wall Street Journal)
- A profile traces Leopold Aschenbrenner’s path from SBF and OpenAI to launching Situational Awareness and examines his ties to Anthropic’s leadership circle. (Wall Street Journal)
- OpenAI built Jalapeño with Broadcom in just 16 months, then reportedly beat chips from Nvidia, AMD, and Google across leading open-weight models; the results suggest custom silicon could challenge Nvidia’s grip on AI inference. (SemiAnalysis)
- Nvidia and Xiaomi introduced new edge and physical-AI chips: Jetson Orin Nano 2 delivers 78 TOPS, while Xiaomi’s Xring O100 targets MiMo models and its 3nm D100 targets autonomous driving. (The Robot Report and South China Morning Post)
- China’s YMTC is seeking $5B in an IPO and told investors it aims to become the world’s largest NAND flash supplier by the end of 2027. (Financial Times)
- SoftBank is in talks to buy a majority stake in humanoid robot maker 1X at a $6B valuation; 1X sought $1B at a $10B valuation in 2025 but raised less than half. (The Information)
- Amazon agreed to acquire DuckLabs, the company behind open-source database DuckDB, for an undisclosed sum; DuckDB will remain open source. (GeekWire)
- HP signed a multiyear agreement with Huawei to license standard-essential Wi-Fi patents, signaling adoption of Huawei technology outside China. (CNBC)
AI Fundraising News (Aug 21 — Aug 27)
- Agentrys: AI chip-design workflow automation company, raised a $24.5M in pre-seed and seed
- Alice: AI model safety and security company, raised $140M
- Arga Labs: AI enterprise software simulation company, raised a $10M seed
- Cloverleaf AI: AI government sales intelligence company, raised an $8M Series A
- Deep Cogito: AI specialized open-weight models company, raised a $43M Series A
- DeepSeek: AI frontier model development company, in talks to raise $7.4B
- Devoted Health: AI healthcare coordination company, in talks to raise at a $25B valuation
- Digs: AI homebuilder operations software company, raised a $25.3M Series A
- Embedd: AI physical infrastructure software company, raised a $2.3M pre-seed
- Emerald AI: AI data-center energy orchestration company, raised a $150M Series A
- Euler: AI B2B partner operations agents company, raised a $4.3M seed
- Faro AI: AI clinical development infrastructure company, raised a $37.3M Series B
- General Intuition: AI foundation models for autonomous agents company, in talks to raise at a $6B valuation
- Generalist: AI robotic learning models company, raised $200M
- Instinct: AI personal assistant company, reportedly raising a $250M Series B
- Itoflow: AI investment research agents company, raised a $2.5M pre-seed
- Keenable: AI web retrieval infrastructure company, raised a $26M seed
- Lambda: AI GPU cloud infrastructure company, in talks to raise up to $3B
- Legato: AI hearing assistance glasses company, raised $12M
- Lica: AI personalized visual communication company, acquired by Gamma for an undisclosed amount
- Liner: AI search engine company, raised a $36.1M Series C
- Mem0: AI memory infrastructure company, raised a $20M Series A
- NeoSoul: AI decentralized infrastructure platform company, raised an $11M seed
- Newcode: AI legal and government workflows company, raised a $12.8M Series A
- OmicsBank: AI clinical data infrastructure company, raised a $2.3M seed
- Perplexity: AI search engine and chatbot company, in talks to raise at a $30B+ valuation
- Primero: AI legacy software modernization company, raised a $12M seed
- Quintessent: AI data-center photonics company, raised a $40M Series A
- Rezolv: AI borrower collections software company, raised a $12.5M Series A
- Ringg: AI enterprise voice agents company, raised a $10M Series A
- Runable: AI small-business growth agents company, raised a $21M Series A
- Rundoo: AI supply-store operations software company, raised a $30M Series B
- Stability AI: AI generative creative tools company, raised a $76M Series B
- Supermemory: AI application memory infrastructure company, raised a $2.6M seed
- Wrtn: AI services platform company, raised $72.4M
- Xpeng’s robotics business: AI humanoid robotics company, raised $900M+