“We all want to be safe, but we all want to move quickly.” Altimeter Founder & CEO Brad Gerstner (@altcap) joined Scott Wapner (@TheJudgeCNBC) live from the White House today to discuss AI leadership, U.S.-China competition and why Brad believes safety and speed have to advance together. He also discussed the importance of exporting the American AI ecosystem globally and why continued investment in compute could help extend the U.S. lead. Watch: piped.video/8dND0Pj7PA0

The speed at which @USNavy is moving to field Blackbeard and @Castelion is scaling production to meet the need is @DeptofWar and industry at their best, moving with urgency against a critical national priority. 🏴‍☠️🤝🇺🇸

@Goyal_Vivek's example is Scenario C-1 below happening live. There are four things an agent can do to a $1,000 hotel booking: A) Free referral to $BKNG or $EXPE. OTA EBITDA goes from $53 to $98. That's today. B) Paid channel at 4–6%. EBITDA goes to $48. New landlord, same business. C-1) Agent sends you to the hotel's own site and the OTA is left selling rate data. ~$8. C-2) Agent aggregates the supply itself. $0. @Muse checked Booking and Expedia, then booked direct. The OTAs did the price discovery and got paid nothing for it. That's C-1, and it's the scenario the market is pricing. The catch: @VallartaGV has a real booking engine and a member rate. Most of the world's 4.7M independents don't yet, and that gap is the whole fight. @skift @Cloudbeds @MewsSystems

“The world wants a lot more AI than it can get its hands on.” Brad Gerstner (@altcap) on CNBC Halftime on the extraordinary demand for AI, and the other side of the equation: the revenue required to support roughly $1.5 trillion in annual infrastructure investment. At the end of the day, consumers and enterprises have to pay for the products being built.

“The world of AI for consumers is now agents.” Brad Gerstner (@altcap) joined Scott Wapner (@TheJudgeCNBC) discussed Meta’s momentum in AI and what he sees as the next 10X moment for consumers: the shift from chatbots that answer questions to agents that can take action. Brad also discussed Meta’s execution, Muse and why he believes the company is beginning to demonstrate a return on its significant AI investment.

“We’re building over 40 gigawatts of AI compute in 2027.” Brad Gerstner (@altcap) joined Scott Wapner (@TheJudgeCNBC) on the scale of the U.S. AI infrastructure buildout and why he believes continued investment in compute can extend America’s lead in AI. “The more compute you have, the more intelligence you have.”

From 90%+ reliance on China a year ago, to proven domestic production today. American-made magnets powering 1 million American-made drones a year. This is reindustrialization 🇺🇸 Proud to back @VulcanElements

Who's accelerating and who's slowing down. This chart is forward dilution minus blended dilution. Positive means the grants made in the last 12 months are bigger than the run-rate they replace. Negative means the worst of it is already in the trailing window. 📈 The accelerators: $W +2.5% $PINS +1.7% $BRZE +1.5% $GTLB +1.4% $S +1.2% None of the five is a large cap, which is a big part of why the Internet and Software sector lines are climbing. The why is different for each: $PINS is a dollars problem. $202k of SBC per employee, 4.4x its cohort, forward dilution 8.6% against 3.0% trailing. $W is a stock price problem. $26k of SBC per employee which is well below its cohort. The market cap is just small enough that a modest comp budget converts into 7.9% forward dilution. The comp committee can't fix that, only the stock can. $GTLB is an affordability problem. $97k of SBC per employee against $93k of FCF per employee. Every dollar of cash the business makes per head goes back out as equity, and forward dilution is 6.7%. 📉 The decelerators: $AI -2.0% $YELP -2.0% $TRIP -1.2% $U -1.1% $CVNA -1.0% $CVNA is actually fixed. Forward dilution rounds to zero and SBC per employee is $4.8k. The 3.3% trailing figure is old issuance rolling off. $AI is not. It's decelerating from 10.5% to 9.2%, which is still more than double its cohort, with negative FCF per head and grants running at 70% of revenue per employee. Slowing down is not the same as stopping. $YELP is on the decelerator list for the second quarter in a row, which is what you actually want to see.

5 examples of company performance for dilution 1⃣ $NVDA covers its $299k/head of grants 26x over in revenue. 2⃣ $CVNA barely grants at all anymore. 3⃣ $NET looks fine on dilution and terrible on FCF. 4⃣ $PINS grants 80% of its FCF per head. 5⃣ $AI grants $0.70 per $1 of revenue per head and burns $160k of FCF per head. Pure dollar amounts aren't what is important, it's whether the business can actually afford what it's issuing.

Trailing vs Forward Dilution Forward dilution (what was granted in the last 12 months) is above trailing (what was actually issued) in Software, Big Tech, and every cohort except mega caps: Software: 2.4% → 3.0% Big Tech: 0.9% → 1.2% Small caps: 3.2% → 4.4% Internet is the only sector where the forward number is lower than trailing. Mega Cap is a bit impacted by the continued $TSLA awards to @elonmusk

📈Tech dilution is creeping up again. After 1.5 years of companies taking their medicine, run-rate dilution has now risen for 4 straight quarters with Internet companies back to 3.3% and software to 2.7%. 🤼 I've added per employee metrics that says this is a dollars problem, not a stock price problem. 💸 TTM net grant $ per employee, 2Q26 vs 2Q25: - Software: $88k vs $68k (+29%) - Internet: $75k vs $63k (+19%) - Big Tech: $69k vs $41k (+68%) The AI talent market is pulling comp up across the rest of tech but not every company can afford that. $NVDA grants $186k per employee and it's fine: $7.8M of revenue and $3.3M of FCF per head. $PINS grants $202k per employee, 4x its cohort, with forward dilution at 8.6%. $NET ranks 29th at 0.8% dilution and looks disciplined. SBC is 162% of FCF. The stock price is doing the work. $AI has 11% dilution, negative FCF, and $0.70 of stock granted for every $1 of revenue per employee. $SNAP has 10%+ dilution while cutting headcount but it's not enough to fix their issues. 🔴Generous comp is fine, but generous comp the company hasn't earned the right to give is the problem. Every company page on the tracker now has revenue, $ grants, SBC and FCF per employee, benchmarked to cohort and sector. Charts below.

TLDR: Do or do not, there is no try

Asked for his favorite AI use case, @karpathy: give a model an arbitrary topic, ask for a one hour video explainer, walk away, come back 10 hours later to a finished hour of video. “why would you read text? Reading text is prehistoric. Just get it to make a video.” 🎥

Brad Gerstner (@altcap): @elonmusk's idea for independent peer review of AI risks is a "great idea." "We had @elonmusk come here & say, 'I tend to agree with Dario. The risk is augmenting & we need to have independent peer review.' I thought that was a great idea."

I built an AV fleet tracker. Every robotaxi argument hits the same wall: it needs a fleet number, and every source gives a different one. So I went bottoms up. TX DMV VIN registry + CA CPUC filings + every press release and news story I could crawl, cross-referenced into one number. 🤖 5,574 autonomous vehicles operating or testing in the U.S. right now. @Waymo is 4,252 of them. That's 76% of every robotaxi in the country. $TSLA is second at 621, @Avrideai third at 344. Everyone else is a rounding error, for now. Texas is the real story. California still leads at 2,153, but Texas is right behind at 1,896, and it's the only state with a VIN-level public registry. Under SB 2807 every car has to be listed on a TxDMV authorization. The tracker crawls it daily, so when Tesla adds Cybercabs they show up immediately. Also on the site: every Waymo CPUC filing for California in one place through Q2 2026 (trips, passengers, miles, deadhead %), a searchable AV news feed that links straight to the sources, and a map of fleet per 100k people. Early days on that last one, but give it 12 to 24 months. We don't have formal numbers on @nuro but should in the CPUC Q3 update. Caveat: the numbers aren't perfect. They include test cars, safety-monitor cars, and cars registered but not yet deployed. It's my best attempt at ground truth, if you see something wrong, let me know.

Next up on First Pass: A conversation with @adijayaprakash from @useblacksmith Everyone knows about the GPU crunch, but the CPU crunch is getting equally painful. In this episode we talk about the nuances of CPU clouds and how agents are driving a surge of CPU demand

Brad Gerstner (@altcap) joined the @theallinpod for a wide-ranging conversation on the state of AI. They cover: → Can AI revenue support the massive CapEx buildout? → The gigawatt gap and power constraints → Why semis could take a larger share of the Nasdaq → AI’s impact on hiring and the workforce → Regulation, rates and other risks ahead → Trump Accounts and “Every Child a Capitalist” Watch the full conversation 👇

Energy is upstream of prosperity. Nuclear is foundational to energy abundance. @generalmatter is building the enrichment capacity to scale nuclear.