August 14th Newsletter

Author

Da Yu
August 14, 2026

As of close on August 13th, 2026

Capital kept flowing into the AI buildout even as markets and policy grew more complicated. July core PPI, which excludes food and energy, was 4.2% YoY, easing near-term pressure for a Fed rate hike and supporting stocks and bonds. U.S. tariff refunds are boosting earnings, with 40+ S&P 500 companies reporting $9.6B in recoveries, led by Apple at $2.2B, Nike at $986M and FedEx at $800M. Alphabet drew roughly $115B of demand for a $25B bond sale. Data-center operator Switch filed confidentially for a U.S. IPO,  Singapore-based data-center operator DayOne filed for a U.S. IPO targeting ~$5B, and Intel launched a $15B share sale. China’s humanoid robot makers captured more than 97% of global shipments in the first half of 2026. AI adoption deepened: Gemini surpassed 1B monthly users, becoming Google’s fastest-growing product. A New Mexico judge, meanwhile, ordered Meta to pay $567M and strengthen protections for minors. Lastly, Silver Lake is in talks to acquire Workday, which has a market value of $43B, in a potential take-private that would rank among the largest software buyouts ever; Workday’s stock rose 18% on Thursday.

We’ve published our Monthly AI Deck for July 2026. Check it out here.

Top 5 AI Highlights

[1] 🛡️ There has been a flood of news from OpenAI this week:
– Astra tests cannot rule out “Critical” cyber capabilities, prompting safeguards (OpenAI)
– Ultrafast delivers up to 14x faster GPT-5.6 Sol inference speeds (X)
– OpenAI repurchased ~$7B employee shares at $852B valuation for liquidity (Bloomberg)
– OpenAI CRO Denise Dresser exits; ex-Wiz COO Dali Rajic takes over (Business Insider)
– Ethics chief Chloé Bakalar departs within one year, extending turnover (Financial Times)

[2] ⚡ Apollo, Blackstone, BlackRock and Brookfield are partnering with Nvidia to invest $500B in AI infrastructure; separately, Nvidia agreed to invest up to $3B in Lancium, the power developer behind Texas' Stargate campus, including $1B tied to performance thresholds. More from Bloomberg and The Information.

[3] 🏗️ Anthropic agreed to a 20-year, $9.1B compute deal with Riot Platforms for 191MW at its Rockdale, Texas campus, sending RIOT shares up more than 15% as Anthropic locks in long-term AI capacity; separately, it is also in talks to acquire Decart for about $6B to add world-model and GPU optimization technology that could lower training costs. More from Compute Deal and Decart.

[4] 🧠 Anthropic widened its Ramp AI Index lead to 43.5% of eligible U.S. businesses versus OpenAI at 39.7%, while premium Fable 5 accounted for just 6% of Anthropic tokens but 11.4% of model spending last month, showing outsized revenue contribution despite limited usage share. More from Ramp.

[5] 🚀 SpaceX is reportedly on track to build roughly 10GW of compute capacity by year-end 2027, including 6GW–8GW during 2027, a scale SemiAnalysis estimates could support a $300B annual revenue run rate; separately, it’s rumored that the $60B acquisition of Cursor could close as soon as next week, with the Cursor brand likely phased out. More from SemiAnalysis and The Information.

✨Open Source AI Development

[1] 💰 Alibaba reportedly plans to introduce revenue share with its heavy commercial users of its next open Qwen model, while Moonshot's Kimi K3 can require up to 30%, signaling that Chinese labs are pairing open weights with new monetization models. More from Reuters.

[2] 🔐 Moonshot's Kimi K3 reportedly left its sandbox during defensive cyber tests and accessed the internet, though researchers said it did not hack external systems; separately, filings show Moonshot converted its Chinese entity into a joint stock company, its first visible step toward a Hong Kong IPO. More from Wired and Financial Times.

[3] 🚀 DeepSeek launched the official V4-Pro and open-source DeepSeek Harness, pushing deeper into agent infrastructure while sharply raising API prices from Aug. 16, with V4-Pro moving from $0.435/$0.87 per 1M input/output tokens to $1.32/$3.96 at peak, signaling a shift from ultra-low-cost model competition toward monetizing a broader developer stack. More from VentureBeat.

[4] 🔓 Meta released Muse Glimmer, a 30B-parameter open-weight model distilled from Muse Spark 1.2 and licensed under Apache 2.0 for local use; Meta also plans Spark 1.2 weights and a $1B community fund, framing the push through its “Future Is for Everyone” campaign for broadly distributed personal AI. More from CNBC, Bloomberg, and Meta.

[5] ⚡ Nvidia released Nemotron 3.5 Lightning, an open 30B-parameter MoE model it says delivers up to 4x faster output, alongside NeMo Switchyard; sources say Nvidia is also developing Nemotron 4 with 1T+ parameters, nearly twice Nemotron 3 Ultra’s 550B, though still smaller than leading Chinese open models. More from SiliconANGLE and The Information.

Founder’s Corner

“Feb 6, 2026 - potentially the last day in history when humans consumed more tokens than agents“

Source: X

What We Read This Week

  • Apple is testing CXMT memory chips for iPhones and MacBooks and has held early supply talks as the AI-driven memory shortage pushes up component costs, potentially opening a major U.S. customer to the Chinese DRAM maker despite political and export-control risks; CXMT has also recently become China’s most valuable company at roughly $520B. (Bloomberg and WSJ) 

  • Situational Awareness invested $500M in Source Foundry, including $400M this week, backing the stealth startup's plan to develop new AI chip-manufacturing tools. (Wall Street Journal)

  • IBM and Together AI signed a $240M, multiyear agreement to build an Nvidia HGX B300-powered inference cluster on IBM Cloud for serving open-source models. (Reuters)

  • DDN's annual revenue is expected to grow from $300M before the AI boom to $1B by year-end as the 28-year-old managed data storage company benefits from AI partnerships and surging infrastructure demand. (Forbes)

  • Big Tech's AI boom increasingly resembles the 1870s railroad buildout, while Nvidia's shift of infrastructure risk to institutional capital could leave investors exposed if AI revenue fails to materialize. (Stratechery)

  • Manus says it will soon resume operating independently, signaling Meta is close to unwinding its acquisition; it says some users must back up and later restore their data during the transition. (The Information and Manus)

  • Legal AI startup Harvey is reportedly in talks to raise $500M+ at a $15.5B valuation, up from $11B in March, while generating $350M+ in annualized revenue. (The Information)

  • OpenAI researchers reconstructed the OpenAI-Hugging Face incident at Black Hat USA 2026, detailing how cyber-capable models escaped a sandbox, chained zero-day vulnerabilities and compromised Hugging Face while seeking answers to the ExploitGym benchmark. (YouTube)

  • Uber divested its stake in Serve Robotics after disagreements over robot deployment ended their long-running alliance, marking a setback for Uber’s push to bring autonomous delivery to its platform; Serve was spun out of Postmates after Uber acquired it in 2020. (Bloomberg)

AI Fundraising News (August 7— August 13)

  • Acrab: AI infrastructure company, raised a $130M Series B

  • Alloy Robotics: AI robotics data platform company, raised an $8M seed

  • Beyang Therapeutics: AI drug R&D company, raised a $30M Series A

  • Blacksmith: AI code-testing platform company, raised a $45M Series B

  • CodeRabbit: AI code-review company, raised a $143M Series C

  • Cognition: AI coding company, in talks to raise $1B+

  • Corma: AI defensive cybersecurity lab company, raised a $60M seed

  • Databricks: AI data analytics and agent platform company, raised $5B

  • DeepSeek: AI foundation model company, in talks to raise nearly $8B

  • Discovered Materials: AI chip-materials discovery company, raised a $9M seed

  • Discovery Loop: AI science and engineering automation company, in talks to raise $1B

  • Edgify: AI edge loss-prevention company, raised a $9M Series A

  • Ema: AI enterprise agent company, in talks to raise $80M

  • Firmus: AI data center infrastructure company, raised $2B

  • Fisent Technologies: AI process-automation company, raised $4.3M

  • FriskAI: AI runtime intelligence platform company, raised a $3.6M pre-seed

  • Genesis AI: AI robotics foundation-model company, in talks to raise $500M

  • Gravity: AI chatbot advertising company, raised a $30.5M Series A

  • Harvey: AI legal software company, in talks to raise $500M+

  • Kevin Weil's startup: AI science company, in talks to raise $150M

  • Lambda: AI cloud computing company, raised $917M

  • Lovable: AI coding company, raised a $400M Series C

  • Mindgard: AI security and red-teaming company, raised a $30M Series A

  • MStack AI: AI chemical R&D and manufacturing company, in talks to raise $30M+

  • Multiplier: AI accounting consolidation company, raised a $35M Series B

  • Naïve: AI business operations infrastructure company, raised a $28.5M Series A

  • Noosphere Labs: AI physical intelligence company, raised $10.25M

  • Palette: AI-native team operating system company, raised a $3M pre-seed

  • Pathway: AI post-transformer model lab company, raised a $30M seed

  • Point2 Technology: AI data-center interconnect company, raised a $60M Series B extension

  • QuantHealth: AI clinical trial simulation company, raised a $45M Series B

  • River AI: AI local model company, raised $1.1B

  • Rocket: AI app-building company, in talks to raise $40M–$50M

  • Silicon Data: AI compute-pricing data platform company, raised a $30.5M Series A

  • Simplismart: AI inference infrastructure company, in talks to raise $50M

  • Skan AI: AI enterprise workflow intelligence company, raised $63M

  • Thrive Holdings: AI-enabled legacy services holding company, raised $2B+

  • Trajectory: AI continual-learning model company, raised $40M

  • Visoid: AI architectural visualization company, raised $2.5M

  • Wippi: AI voice products for children company, raised a $1.2M seed