Altimeter Capital

Last week on CNBC Halftime, Brad Gerstner (@altcap) broke down the AI selloff — first with Scott Wapner (@ScottWapnerCNBC), then alongside Jensen Huang (@nvidia): With Scott Wapner on markets + AI CapEx: "There's a fog of war when the rate of change goes parabolic." "When the smartest guys in the world — Sundar, Elon, Andy Jassy, Mark Zuckerberg — are all saying we are going to earn a return on this investment, my bet is on them." With Jensen Huang on AI infrastructure + U.S. competitiveness: "This is like building the interstate highway system — $700 billion of investment, all funded by profitable private companies in America." "Could you imagine being large early investors in Google, Amazon, and Meta? Incredible. And that's what we're looking at." As founder of @InvestAmerica24 | Trump Accounts, Brad is focused on ensuring every American has a stake in the future being built. Watch the conversation 👇

Power is upstream of economic value. True on Earth. True in space. Economic value will concentrate in platforms that can deliver unique capability and large amounts of power on orbit. @K2SpaceCo delivering 20kW per satellite is a massive unlock. It creates a new operating regime that expands the frontier of what is technically and economically possible. Higher-power payloads. More compute. Stronger propulsion. Greater maneuverability. Entirely new mission areas. T-Minus ~10 days to Gravitas

What drove ChatGPT from 0 to a billion users? How do they plan to get the next billion? New pod. Sat down with @nickaturley, Head of ChatGPT, for one of the more candid conversations I've had on the show. We cover GPU allocation tradeoffs, retention smile curves, pricing, Code Red as a strategy, bringing on @steipete from OpenClaw, and why Nick believes curiosity is the only perma-skill in the age of AI (with a shoutout to our friend @bgurley's new book). full interview in comments

Awesome stuff from the @Hammerspace_Inc team. Storage is more important than ever. In the age of AI, storage systems will need to be re-invented to keep up with model's insatiable demand for data (both during training and inference). Nvidia today came out with their "AI Data Platform" reference architecture to help enterprises solve these storage challenges. Pumped to see Hammerspace listed as 1 of 12 partners, alongside some legendary storage companies! nvidia.com/en-us/data-cen…

Heard $Druckenmiller on a podcast: "Anybody who believes that(i.e. AI will be very deflationary and it will lead to massive job losses) with conviction suffers from arrogance and not an open mind…every technological revolution since was known to man, it’s been declared, for jobs it’s the end of the world…So let’s say the pessimists are right on AI, it’s possible you get a government response with printing and universal income." That got me intrigued, so I asked $Claude and $GPT to go back through history and review a few cases where societies feared that new technologies or structural shifts would trigger permanent mass unemployment: The Luddites (1811-1850s); American farmers (1800s-1970s); Telephone operators (1920s-1960s); ATMs and bank tellers (1970s-2010s); The Rust Belt (1980s-present); Radiologists and AI (2016-present). Full case study: open.substack.com/pub/robonomics… In none of these cases did we get sustained, economy-wide deflation or permanent mass unemployment. --- What actually feels different this time? Speed. The shift from 41% to 2% of the labor force in agriculture took roughly a century. Telephone operators faded over about 60 years. ATMs played out over roughly 40 years. Even Rust Belt deindustrialization - often seen as shockingly fast - unfolded over 20-30 years. AI could move much faster. --- What happens if AI really does cause deflation or unemployment? The Fed has four main anti-deflation tools: - Cut interest rates - QE - Forward guidance - Coordinate with fiscal stimulus - the 2020 playbook These tools can support aggregate demand. What they cannot do well is determine who benefits. The most likely macro outcome may look something like a Rust Belt at scale: nationally, GDP still grows, unemployment stays moderate, and headline deflation never really shows up. But underneath those aggregates, inequality widens and pain gets concentrated. --- If $Dario is directionally right - say 50% of entry-level white-collar jobs disappear within 1-5 years - the math gets meaningful quickly: - Current unemployed = 168M x 4% = 6.7M - 50% of entry-level white-collar jobs displaced = 5M-7.5M - Unemployment could rise toward roughly 7-9% - Historically, U.S. unemployment has usually ranged between roughly 4% and 10% outside extreme crises. A move from 4% to 7-9% would not be unprecedented - but it would be large enough to reshape politics, wages, and social expectations. --- Long way of saying: I’m trying not to be “arrogant without an open mind.” As $Druckenmiller said: "you just have to always be looking at what other people might not be, and then if you’re prepared for it mentally, you can adjust quickly enough, um, in your portfolio to it as it unrolls." +++ Full analysis: open.substack.com/pub/robonomics…

Ian buck presentation actually really great and compelling - explaining why we need to push out to pareto frontier (agent swarms), why we need to push higher token /s. Why the hybrid architecture of GPU + LPU is much better than just GPU or LPU standalone… pretty awesome

ChatGPT's retention is in a league of its own. 71% of users still active at Month 10... nearly 2x the next closest AI app. Most consumer apps would kill for these numbers. Full analysis in Part 2 of our State of Consumer AI series 👇 x.com/apoorv03/statu…

Altimeter partner Apoorv Agrawal (@apoorv03) joined @TBPN to share his thoughts on consumer AI, durability, and what it takes to build a lasting platform. A few highlights: 🏆 On winner-take-most dynamics: "History has shown us that these consumer markets have been winner take most. Sometimes winner take all." 📈 On ChatGPT: Apoorv discussed how ChatGPT is the only app in AI that shows what we call a smile curve — where users come back over time after having churned. 🌍 On the next wave of users: Apoorv anticipates that the next phase of utility will come from proactive AI... not waiting for you to log on, but coming to you. ⚡ On the public vs. private market disconnect: Apoorv comments that the public markets are a very efficient mechanism for repricing — but not always hyper efficient in times like now, when the change is so fast. Full episode linked below. 👇

Uber Is Quietly Winning the AV Rideshare Setup If 2025 was the proof point that consumers will actually take autonomous rides at scale, 2026 is starting to look like the year the strategic map gets redrawn. For the last few years the AV debate has mostly been framed around who has the best self-driving technology. That still matters of course. But increasingly that is the wrong question for investors. The more important question now is: who is best positioned to turn AV supply into a scaled rideshare network? That is a different question entirely. To level set: this is no longer just about the best AV stack @Waymo is the only player that has really crossed from demo to scaled commercial reality. The company said in February it was already doing more than 400,000 paid rides per week across its operating markets, and it raised another $1.6 billion while laying groundwork for expansion into more cities. Its new Arizona manufacturing facility with $MGA is designed to produce “tens of thousands” of autonomous vehicles per year at full capacity. That is the most real robotaxi business in the U.S. by a mile. But the leap from “best AV operator today” to “winner of AV rideshare economics” is not automatic. Because scaled rideshare is not just a software problem. It is a supply problem, a dispatch problem, a maintenance problem, a financing problem, and maybe most importantly a utilization problem. That is where $Uber's setup starts to look much more interesting than the market gives it credit for. Uber is not trying to win autonomy. It is trying to win the network. Uber’s strategy now looks pretty clear: let others build the autonomous brain, while Uber becomes the default marketplace, demand layer, and utilization optimizer. That may end up being the smarter economic position. A lot of commentary around AV tends to sloppily bundle “partnerships” together as if they are equal. They are not, some partnerships are real supply, some are geographic options, but Uber increasingly has both.

Agentic Commerce Update | 1Q2026 Recap - 3 most interesting developments in agentic commerce over the past few months… 1/ @ChatGPTapp scaling back from an end-to-end shopping destination and pivoting to a shopping entry point that routes transactions to specialized commerce apps (source: The Information). 2/ All commerce platforms embedding agents directly in their products ( $BKNG $EXPE $AMZN). The most interesting example to me may be AMZN’s “Buy for Me” - agent can complete purchases from other brands inside the Amazon Shopping app when Amazon itself does not carry the product. 3/ B2B commerce is seeing early agentic adoption on the sourcing side, where the workflow is highly research-intensive. --- Guiding principles - what likely holds regardless of how agentic commerce evolves 1/ AI is all about democratizing information and resources Commerce has historically been closer to a seller’s market. Consumers are largely shown what platforms choose to surface - through ads, rankings, and distribution channels. Each layer adds another “tax” in the form of marketing spend, platform fees, or payments. AI agents should gradually shift power toward the buyer. Instead of navigating layers of marketing and platform incentives, consumers can theoretically access the best product for their needs directly. This dynamic may be even more powerful in B2B commerce, which historically has been opaque, fragmented, and relationship-driven. 2/ Faster, cheaper, better rules Faster & cheaper: Price competition should intensify across retailers and even payment providers; Scale advantages may matter more, not less - AMZN’s logistics and fulfillment advantage still remains structurally difficult to replicate Better: more personalized and context-aware - agents matching products to a buyer’s specific needs, constraints, and preferences. 3/ Control of the top of the funnel matters (still) Today $AMZN generates roughly $80B+ (source: Visible Alpha consensus number) in ads revenue, largely because shopping intent starts on $AMZN. If discovery shifts upstream toward AI interfaces (e.g. ChatGPT or GOOGL’s Gemini), the economics of the top of the funnel could shift meaningfully. --- h/t Bernstein analysis: More analysis on each company in here: open.substack.com/pub/robonomics…

Altimeter partner Erik Kriessman (@ekriessmann) spoke on a panel at iConnections on investing in cutting-edge defense technologies — alongside investors and allocators from 137 Ventures, ICG Advisors, and the University of Utah. A few of Erik's key takeaways: 🎯 On talent and company formation: "Few companies will ever matter in any sector. I think that's even more true in defense. Talent's going to concentrate against a couple of opportunities that are actually worthwhile to go after." 🏗️ On what success looks like: Teams that understand customer requirements, can deliver on them, and can build trust with government partners — whether that means shaping requirements or meeting them. 🔩 On the industrial base landscape: Beyond the platform plays, Erik highlighted the importance of companies supplying irreplaceable inputs across the defense ecosystem — like @VulcanElements, which is onshoring rare earth magnets critical to every satellite, drone, submarine, and fighter jet. 🤝 On collaboration: The ecosystem is more collaborative than people expect. Companies like @anduriltech, @HadrianInc, @Castelion, and @K2SpaceCo work together to deliver full solutions — and the best founders understand that building the best product alone isn't how you win. You have to know how to sell. The discussion highlighted how the current wave of defense innovation has been building for well over a decade. Watch Building the Best Product Is Not How You Win in Defense... youtu.be/BLJcFOWlmGE?si…

Consumer markets follow a power law: winner take most, if not winner take all. Google took 90%+ of search. Facebook took social. Apple took mobile profits. But user #s are not conclusive. Remember Clubhouse? It hit 10M downloads in its first month. Or BeReal? was #1 on the App Store. Neither built a habit. Usage makes headlines. Habits build franchises. In this Part 2, I dig deeper into how AI engagement and retention. Data suggests ChatGPT is not just the largest AI app. It is also the stickiest. x.com/apoorv03/statu…

I built out an MCP server and linked it up to a CustomGPT that right now anybody can use to ask questions and query transcripts: chatgpt.com/g/g-69a7bf0d3c… If you get a bad response or it feels lacking reply below with a screenshot of it and I can tweak. It's a good learning method for how all this works on the back end. It's not an OpenAI app because it's super arduous to go through the approval process. But you should be able to pin it in your ChatGPT app to save to your sidebar. Also it's unlocked so anybody can share and utilize the link right now but if people start abusing it or it looks like it's costing me a ton on the MCP hosting side I'll probably switch it to oauth for free subscribers only.

OpenAI / @sama at MS TMT: Scaling + progress: “Most important thing to know is we have been on accelerating exponential of progress; been on it much longer than 3 years; since 2019 - reason it is accelerating is algorithmic, compute, more smart ppl working on it, more economic input letting us buying GPUs faster; these are compounding exponentials… does not seem the exponential is stopping anytime soon… end of 2028, maybe sooner, will likely be more intellectual capacity inside data centers than outside of them” Codex / evolution of SWE: 2M+ users growing 25% per week. “Seeing teams of 5 ppl / 1 person + tens / hundreds / thousands of GPUs doing work of entire company engineering department. Codex can write large and complex code but now also do things like chip design where it is super human…” software engineer is now “manager of agents” Scientific discovery + new research: “Seeing beginning of AI discovering new science - still small but seeing early signs across mathematics, theoretical physics…" Models are not just learning distribution of data, we are "teaching them generalized ability to think and understand data and in general purpose way, how to reason about what comes next. It appears models have crossed this threshold and can make discoveries." Company of the future: “We will all want person making final decisions for accountability of company for a long time…" but "AI CEO can [talk to every person at company, be in every meeting, customer call, be an expert in every function]. The % of decisions relied on AI will go up [at your company] or a competitor who does rely more on AI will rise up.” Capital raise + strategic partnerships: “NVDA we have partnered with for decade - continue to lead with best chips. All training on NVDA. With partnership we get allocation to new chips we are excited about. AMZN we will take some Trainium in 2027 and 2028 which will further help us meet inference demand. Will work with them to distribute stable runtime environment for agents in the enterprise” Defense + government: "Really love and support our country, AI will be be important to defense" Did not rush into military work on classified networks but believes it's critical. "If I were running military and gov and had a bunch of AI companies saying we are about to build super intelligence which will be important for geopolitical power, but we are going to stop working with you - I would say (i) I need that tech and (ii) gov supposed to be more powerful than private companies." OAI effort was to de-escalate. DoW "very understanding and great partner" on red-lines re survelliance and autonomous weapons. Will deploy FDEs with clearance to make sure tech properly used. Three flash points of superintelligence: "(1) Are the companies developing AI or is the government more powerful? (2) How do you reconfigure economy when no one can outwork a GPU? (3) Who gets to decide the values we align superintelligence to? Have seen beginning of #1 - we believe we need to trust in democratic process which has gotten us so far over last 250 years” What people misunderstand about AI: "In general, I have found that the smarter someone is, the more they want to believe AI is going to hit a wall. Basis of this psychological flaw is understandable, the more you define yourself by intelligence, the more you are incentivized to say it can't do this." AI-native competition + adoption urgency: Companies used to only compete with competitors who adopted slowly too / they had same constraints like a “slow security org”. "Now you have to compete with not only current competitors but new companies that are mostly AI that do not have slow security org… “Next year I predict we'll talk about companies with 1/10th or 1/100th of the people and a ton of GPUs building entirely new types of competitors.”

Takeaway from Sam Altman's talk at MS Conf today is that Sam and Dario agree on far more than they disagree on. Both say: AI is on an accelerating exponential with no wall in sight. Both say: code is the breakout use case. Both say: the future of work is managing swarms of AI agents. Both support working with DOW in principle. Both say: companies that don't adopt fast will lose to AI-native competitors. Other highlights from Sam's talk: On defense/DOW:"Cannot tell government that superintelligence is critical to geopolitical power but we won't work with you." Found DOW "extremely understanding" on building safely. Last week when things became a "fight" — "increasingly clear things could go very badly and we wanted to de-escalate." On Codex:2M+ users, growing 25% per week. Teams of 1 person + hundreds of GPUs doing the work of entire engineering departments. Now doing chip design where it's "super human." On partnerships:Nvidia: all training on Nvidia, new chip allocation coming. Amazon: Trainium in 2027-2028 for inference + distributing agents to enterprise. On those underestimating AI: "The smarter someone is, the more they want to believe AI is going to hit a wall." The more you define yourself by intelligence, the more you're incentivized to believe it can't be replicated. On the AI CEO thought experiment:Will always want a human making final decisions for accountability. But the % of decisions relying on AI will go up and up — "or a competitor who does will rise up." "Humans will stay in charge but have to delegate and supervise."

We compare AI products against each other. But the more interesting comparison is against the broader internet / software landscape. I plotted ChatGPT and Gemini alongside Internet Utilities (Chrome, Whatsapp, YouTube) Social (Instagram, TikTok, FB), and niche verticals (Spotify, X, Amazon) on usage: This framing (utility app vs niche app) matters because it defines the natural ceiling for how large these AI user bases can get. Full article State of Consumer AI Part 1 (Usage) : x.com/apoorv03/statu… Part 2 (engagement) is coming soon. I think the data there is even more telling. If Part 1 is the Q, Part 2 will be the P in a PxQ revenue equation.

Dario at MS TMT Conference today: On defense / DOW:"We really believe in defending America." Anthropic has been working with the national security community for 2 years. "We are the most lean forward." On AI acceleration:"We do not see hitting a wall. This year will have a radical acceleration that surprises everyone." Exponentials catch people off guard. "We are at the precipice of something incredible. We need to manage it the right way." On where markets are wrong:"It's already big and it will get 1 million times bigger." The underestimation of exponential growth is the key thing people need to understand. On revenue scale:Anthropic was at ~$100M run rate 2 years ago. Now at $19B run rate. On culture — Dario says he spends 40% of his time on it:"Anyone who is CEO of a growing firm needs to realize they are chief culture officer. My job is to make sure everyone is on same page and believes in what we are doing. That's the most important thing." He does a vision quest with the whole company every couple weeks. "I want them to hear it directly from me. If I tell the CTO, who tells the VP Eng, who tells the manager — that's too long of a game of telephone." "Politics and infighting are a cancer to companies as they grow." On talent retention vs Meta:"We lost 2 people to Meta. They lost several dozen. Normalized by size, they lost 10-20x more people vs us." Attributes this to unified culture generating "super linear returns — by working together vs working against each other." On code as the breakout use case:Code has "exceeded our high expectations." Why? Devs adopt fast, code is verifiable, and gains compound — you build software to build software. "Didn't realize it would go so fast even at traditional enterprises." Frustration is around regulated industries where legal/compliance slow things down. "That's how fast everything could be going if not for non-AI barriers." On Anthropic's own AI usage:Top internal use cases: 1) writing code, 2) the process around writing code (SWE), 3) managing servers and controlling clusters. "If we were paying ourselves for our usage, we'd be one of our largest customers." On Claude Code:"You can supervise an army of 100 Claudes. It's closely analogous to a management skill." The people who are best at it keep the big picture in their head. Higher return to finding people who can handle more complex tasks. On platform vs apps:"We are primarily a platform, but there are places where we have expertise to make something directly useful." Claude Code emerged as a tool they built for themselves — thousands of internal users before shipping it externally. "Code is a prelude for what we will see in everything else." On societal implications:"Human history — lots of muddling through. We found ourselves in this comedy of errors and figured it out eventually. It's happening so fast that we need to do better than that this time." The market will deliver positive benefits — "I see that as priced in." What's not priced in: the choices we make around externalities. Jobs, national security, ensuring the benefits reach everyone. On chips & compute:Anthropic uses multiple chip suppliers. "We find that actually using different chips is useful to us. Chips aren't just a speed number — we gain benefits from heterogeneity." Also standard business logic of having more than one supplier.

If my AI personal agent is truly unbiased and is working “for me” - won’t it always go to the platform with the lowest price and fastest delivery. Won’t all the leading ecommerce players further increase their share. Maybe ad revenue slowly finds its way to higher take rate. But won’t their multiple increase vs decrease as they become more dominant?