There are four things that can go wrong for Nvidia, but only end-user demand really matters. 1. Direct customers stop buying compute. Not going to happen, the hyperscalers are exploding.  2. Worries around roundtripping cause problems. Also not happening, Nvidia is kicking off so much cash that they can’t spend it fast enough.   3. Competition. In theory, one or five other startups could take significant market share. It’s probably already happening, but when you start at 90%+, it takes a long time to show up in the numbers. 4. End-user demand. This is all that matters. Everybody gets to sell chips to hyperscalers, provided hyperscalers can sell compute to OpenAI and Anthropic, provided they can sell intelligence to end customers. Right now that’s happening, so everything down the line is plus or minus fine. If it ever unravels, it won’t be because of roundtripping. It will be because the frontier models forecasted 5x growth in end-user demand and only got 3x.

Imagine the VC industry consists of 2 funds: one is $99.9B, the other is $100M. Let's say after 10 years, the first fund is worth $105B. The second is worth $500M. The first fund has generated 93% of the industry's profits. But no rational person would rather be an investor in the first fund than the second fund.

Lucky enough to have @DroneDeploy and @fin_ai close this week. Happy for both teams but genuinely sad to no longer work with @mikewinn and @eoghan. Two examples of savvy big company SaaS acquirers (@procoretech and @salesforce) leaning into AI with acquisitions that can meaningfully move the needle. Congrats to all.

Today Fin officially joins @Salesforce. It’s a huge moment for us. Fin was the very first AI Agent for CX and started an entirely new category. Today Fin is the highest performing Customer Agent on the market, running on our custom AI models trained specifically for CX, and resolving over 2 million customer issues every week. Thank you to our customers, teammates, partners, and supporters who’ve been a huge part of our journey so far. Now with Salesforce, Fin will make perfect customer experiences possible for millions more people. You can read the full press release here.

It's official: DroneDeploy is now part of @procoretech. Today we closed our acquisition by Procore, joining a team building the operating system for construction. Read more from Procore CEO Ajei Gopal:

One of the most exciting (and challenging) applications of physical AI is thinking about how it can accelerate scientific discovery. Imagine a robot that can pipette, handle stem cells, operate existing lab machinery, 24/7 - a physical AI scientist. This is what the new episode of @GreylockVC Change Agents is all about, featuring @michellearning, founder and CEO of @medra_ai. We talk about building an AI-powered lab, how accelerating physical AI lab experimentation is the key to unlocking drug development, and how Medra is automating experimentation today inside pharma companies. 00:00 Teaser 00:56 Medra’s founding mission 04:17 Hypotheses are not enough 13:27 Medra’s differentiation 20:22 The experimentation bottleneck 27:06 Refining the agent harness 29:05 The role of the physical lab 33:23 Managing context across different customers 34:59 Working with models 40:26 Towards personalized medicine and drug discovery 43:15 AWS for life sciences Thank you for joining @michellearning

Last week, I sat down with @nicckopp, CEO of @RilletHQ, and Ali Hussain, CEO of @get_tabs, to discuss the next generation of AI-enabled financial software. What wins deals: While AI has hit the CFO stack, replacing an ERP is too consequential for an AI story alone. Application merit still wins. The depth of the accounting workflows, controls, integrations, and overall product must stand up against systems that have been entrenched for decades. What's next: Auditability may be the key to unlocking agentic finance. The ability to not just execute work but also show exactly how it was done is what allows teams to trust AI with progressively more responsibility. This means that instead of automating one workflow completely end-to-end, the opportunity lies in breaking complex work into two- or three-step workflows where every action, assumption, and output is visible and verifiable. Thanks to all who joined us, and to Nick and Ali! It's always a pleasure to discuss the future of the market with the founders building it.

Big acquisitions are driven by paranoia, but “just because you are paranoid doesn’t mean they’re not out to get you.” When the frontier models are claiming TAMs the size of US GDP, every other tech company has no choice but to respond. The recent spate of big acquisitions (Hugging Face, Cursor, and OpenRouter) is about extracting value away from the frontier models, either by enabling the open weight ecosystem (Hugging Face and OpenRouter) or by taking the largest frontier model market: coding (Cursor) and building a viable competitor to Claude. They won’t be the last.

When the AI IPOs come, the stock based comp numbers are going to be nothing like we’ve ever seen. None of that shit will matter. The only thing that matters is the growth rate and the 27 and 28 projected revenue. The reason you normally worry about SBC is because in a steady state business like Workday it really is a cash number. If you’re giving someone 500 grand every year to show up and be a middle manager, they are mentally putting those RSUs into their comp. If you stop giving them RSUs, they’re going to want cash. In a mature business, it’s totally correct to worry about SBC. A hypergrowth AI company is not that. Someone hired with a million dollar package in 2023 ended up making 51M four years later. That doesn’t mean you have to pay the next guy 51M. He would have signed up for the million he was offered. That’s the real economic stock based comp. The other 50M is just dumb luck. So it’s okay in a hypergrowth company to look past a good slug of the SBC and normalize it out. And conversely it’s not okay in a mature company because that’s real money that people are spending. It might be a little unfair that the hypergrowth company gets a free pass and the mature one doesn’t, but they get a free pass anyway, provided that revenue goes up. Once revenue stops going up, all bets are off.

The most important number in the market today is the OpenAI Q3 ARR number. The trailing numbers had GAAP revenue of $6.7B in Q2, growing 18% q/q. This would point to sub 100% Y/Y growth, rapid deceleration and GAAP revenue under $30B for 2026. Amazing numbers in any normal world but a company with those numbers will not buy $750B of compute by 2030 as OpenAI has promised to do. The result would be “bad” for the AI trade. That is why the company is (wisely) leaking strong July and Aug reacceleration numbers and analysts are crawling through CFO comments like Kremlinologists of old trying to figure out what “35% quarter to date growth” means. Are we back to 3x plus Y/Y growth?

Companies are starting to think about next year's AI budgets, and the big question is what happens to overall spending. There are two forces at play. 1. There are maybe 5% of companies who tokenmaxxed earlier this year and are now “tokenminimizing” like crazy. Coinbase is the poster child here and is pointing to a 50% reduction in spend from the peak. 2. And then there are the toedippers, the 95% of companies who have just started on AI. If even 25% of those companies start to expand, their growth will swamp the reduction from the tokenmaxxers. Tokenminimizers vs toedippers will be what decides FY27 budgets.

Today we're announcing our $100M Series C at a $1B valuation, led by @ICONIQCapital. It's our third round in 14 months and brings our total funding to over $200m The round came together in less than 48 hours. We weren't planning to raise. Then new ARR doubled last quarter and we crossed 600+ customers including public companies and enterprises with $2B in revenue. Rillet AI agent usage is growing 70% each month and CFOs are replacing Oracle Fusion, SAP and Workday to run their companies on Rillet. We also announced our alliance with EY and officially partnered with more than half of the top 20 CPA firms. @RilletHQ is creating an entirely new category of finance infrastructure. With Rillet, the ERP becomes the harness for the modern finance organization, where people and agents share the same financial data, apply the same accounting policies and controls, and work from a single, continuously updated view of the business. Agents perform increasingly complex financial work, while finance teams retain visibility, approval authority and a complete audit trail. Thank you to our customers for believing in the mission and building the future of finance with us. We could not do this without you. Thank you to the Rillet team. As I say often, building Rillet is our own version of sending rockets to Mars. It is incredibly difficult. You make it look easy. Finally, thank you to our investors. To ICONIQ for leading the round and to our existing and new investors for building with us. @sequoia , @a16z, Sequoia Global Equities, @BainCapVC, @oakhcft, @BatteryVentures, @FirstMarkCap, @scalevp, and @creandum. Back to work :)

A platform shift is an ever expanding crevasse between the before and after. If you’re going to make that jump, you have to make it fast, because there comes a point where the gap gets too big to cross. Palantir saw it with AI in 2023 and went from 18% growth to 98%, meanwhile, many SaaS companies are only now nerving up to make the jump. There’s going to be a lot of companies paying the bill in 2026 and 2027 for the hesitancy they showed in 2023 and 2024.

Whatnot raising at a 20 billion valuation doing live shopping shouldn’t surprise anyone. Shopping + Fun equals big bucks. QVC and the home shopping network ran the same exact model on TV for decades, and people loved it. eBay auctions were the same thing online in the 90s and it’s still worth 45 billion today. There is more to life and venture than LLMs.

If you look at the last 3 years of AI progress, technologists have largely been right about the (massive) increase in capabilities, and economists have largely been right about the economic impact. I don’t expect that to change. Ricardian comparative advantage ftw!

I don’t completely agree with Harry’s take around A-tier talent because it's all about what kind of A-tier talent you want. Nobody building application software companies in the PC era had A-tier chip talent, because they weren’t building chips. They were taking advantage of all those smart Intel engineers and Microsoft operating system engineers and building on top. The same logic applies today. If you’re actually building a frontier model and you don’t have A-tier model talent, then yes, you’re toast. But if you’re not, you don’t need A-tier model building talent. You need A-tier talent in USING frontier models to build great apps.

Congratulations to the @coderabbitai team on their $143 million Series C! Scale led CodeRabbit’s Series B less than a year ago. Since then, the company has grown revenue more than 5x year-over-year. Co-Founder & CEO @harjotsgill talks with Scale investor @avitus about CodeRabbit's work "continuously shipping the future" of coding.

We’re excited to announce DiG-bench, a new benchmark for discovery! Over the last few weeks we’ve been testing frontier AI models on our novel discovery games and seeing how they score. Each game is a text-based environment, so they probe discovery capabilities in the natural domain of language models, rather than requiring additional, potentially confounding, visual understanding. TL;DR frontier models have improved a lot over the last few months. But they are still stumped by some surprisingly simple problems, even in their native text domain. With @cocosci_lab (@Princeton) @MITCoCoSci (@MIT) @SchmidhuberAI (@KAUST_News) @misovalko (@Inria) @tri_dao (@PrincetonCS) @RMBattleday @zebkDotCom @FraserGreenlee @akaijsa @ClareMaguire @TimMuller1 @kubicek_ales @physicscat0x7d @SukritSumant @thoughtchannel_ (1/5)