Nvidia doubled its revenue. OpenAI just showed a chip that beats it.

Hi AI Futurists,

Nvidia just posted the biggest quarter in its history, and the most interesting part happened the day before. OpenAI walked into a chip conference with silicon that beats Nvidia at its own game. Let's take a look.

Our agenda.

Best,
Lex Sokolin

P.S. Hit reply with suggestions or feedback!

Leadership is asking: are we getting value from AI? Which tools are worth the spend? Where are we exposed? Right now, most teams have no idea.

Harmonic Security Usage Explorer changes that.

You get a complete picture of how your organization uses AI, automatically categorized into custom tasks and use cases.

You’ll see the projects being worked on, who’s using what tools, where AI investments are driving value, and where employees are engaging in risky behavior.

CIOs can rationalize spending and cut wasted licenses. CISOs can pinpoint where risk exists and neutralize it. AI committees can show exactly how their efforts are paying off.

See it in action

💰 Nvidia is in talks to back Perplexity at a $30 billion valuation (PYMNTS). Perplexity's revenue run rate passed $750 million, triple where it started the year. Huang calls it his go-to chatbot, and now he wants a receipt.

🤖 Amazon is shutting down Mechanical Turk after 21 years (The Next Web). Half a million people doing micro-tasks for cents ends September 30. A 2023 study found nearly half of them were already using AI to do the work.

📊 OpenAI is gaining on Anthropic with business users (TechCrunch). Ramp data across 70,000 businesses puts Anthropic at 44% and OpenAI at 40%, and OpenAI is closing fast. Enterprise loyalty lasts exactly one model release.

🧠 Claude's memory now spans chats and Cowork sessions (Engadget). One memory across chat and agents, visible and editable by topic. Health, religion, and politics stay out unless you opt in.

🎵 Universal, Sony, and Warner all bought into Stability AI (Music Business Worldwide). A $76 million round, the first time all three majors took equity in one AI company. Cheaper than three lawsuits.

⚡ Emerald AI hit unicorn status making data centers grid-friendly (SiliconANGLE). $150 million at a $1.05 billion valuation for software that flexes a data center's power draw with the grid. Power is the new GPU.

🚕 Waymo is taking its robotaxis to Munich (TechCrunch). Its first European market, with testing in weeks and service targeted for late 2027. Straight into BMW's home town.

🚚 Gatik raised $200 million for driverless trucks (Forbes). Qatar's sovereign fund led the round after 85,000 fully driverless deliveries for Walmart and Tyson. Freight got boring, which means it works.

🎓 OpenAI gave free ChatGPT to 100,000 more teachers (9to5Mac). Another 55 school systems across 20 states, free through June 2028. Free until 2028, sticky forever.

🚀 Musk told Cursor staff he is "not used to losing" (Gizmodo). At his first all-hands since SpaceX's $60 billion Cursor deal, Musk admitted Grok fell behind OpenAI and Anthropic. Forty-five staffers have already left.

Nvidia reported $96.2 billion of second-quarter revenue on Wednesday, up 106% from a year earlier and about $4 billion ahead of Wall Street's estimate. Data center sales made up $89 billion of it, earnings came in at $2.22 a share against $2.09 expected, and the company handed $26 billion back to shareholders in a single quarter. Jensen Huang told analysts that "AI has reached its inflection point" and demand is accelerating, then went further than any guidance the street had modeled: roughly 70% revenue growth for fiscal 2028, where analysts had penciled in 44%. For a company this size, that forecast is the story. It says the spending has years left to run.

The awkward part happened a day earlier at the Hot Chips conference. OpenAI unveiled Jalapeño, its first custom chip, designed with Broadcom and built into systems by Celestica, and independent benchmarks from SemiAnalysis showed it beating both Blackwell and the newer Rubin at inference. The numbers were not close: 1.5 to 1.9 times more work per watt at peak throughput, latency up to 3.6 times lower, and 1,400 tokens per second on GPT-OSS 120B. There are real caveats. Jalapeño only runs inference, the chips shown were engineering samples, and volume production does not start until 2027. Still, the design work began in mid-2024, which means OpenAI built a Nvidia-beating chip on its first try in about two years.

Here is how both stories fit together. Training built Nvidia's monopoly, but inference, the everyday work of running models for customers, is where AI spending goes as products mature, and inference is exactly where every big customer is building an exit. Google has TPUs, Amazon has Trainium, and now OpenAI has Jalapeño. Nvidia's response this week was to go deeper into its customers instead: those Perplexity talks would value the company above $30 billion. If your product runs on someone else's model, this is the best pricing news you have had all year. Chips competing at the inference layer is how API prices keep falling, and the fight officially has a second contender.

X1: describe an iPhone app, ship it to the App Store

For founders with an app idea and no engineer.

Nimbia: an AI that runs live onboarding calls for your product

For SaaS founders and PMs who cannot put a human on every trial signup.

ChatCut Desktop: a video editor you direct in plain English

For marketing teams shipping talking-head video and ads without an editor on staff.

This 121-page research report provides the first comprehensive taxonomy of public companies, private ventures, and tokenized protocols building the infrastructure for autonomous AI systems. Compiled by Lex Sokolin, former Chief Economist at ConsenSys, fintech strategist at Autonomous Research, and current Managing Partner at Generative Ventures, this report delivers institutional-grade analysis of 100+ companies across 14 critical infrastructure layers. Learn more here.

That’s all for today, folks!