Proud to announce that @Principal_VC portfolio company @Infinity_ai has raised $15 Million in seed capital, with participation from Touring Capital, researchers from @OpenAI and @AnthropicAI, and executives at leading chip companies. We deeply believe in Infinity's mission to make powerful AI affordable and accessible by building the software layer that lets any hardware run frontier models at top performance. This funding will fuel Infinity's growing engineering team, accelerate its work with chip partners like d-Matrix, and allow the team to scale Ignition, Infinity's autonomous AI agent that writes and optimizes inference code for new hardware, compressing years of engineering work into days. Founder Jeremy Nixon (@JvNixon) put it best: "The next era of AI will be defined not just by who makes the best chip, but by who can make any chip run state-of-the-art models at blazing speeds." Thrilled to be backing Jeremy and the whole Infinity.inc team, and excited for what's to come. Learn more about today's news here: businesswire.com/news/home/2… Full story from @TechCrunch here: techcrunch.com/2026/07/20/in…

Starting a company has never been easier. Building one that lasts hasn't gotten any easier at all. AI has collapsed the cost and friction of company formation. A founder today can ship a working product, set up infrastructure, run marketing, and handle ops with a fraction of the team (and a fraction of the capital) it took five years ago. The result is a surge in new companies. More launches. More demos. More "AI-native" pitch decks than any investor can read in a lifetime. (I say this from experience.) But here's the part that's getting lost in the noise: Easier to start ≠ easier to build something long-lasting: *Lower barriers to entry are inflating startup volume, not necessarily startup quality. *Many of these companies exist because formation is cheap, not because they're solving lasting problems. *Easier formation doesn't expand the market. The number of real, durable problems worth solving hasn't grown just because the tooling did. *Durability still depends on the same things it always has: real customer pain, defensibility, distribution, and the discipline to compound over years. *The next great sorting mechanism in AI won't be novelty. It'll be endurance. The launch phase is being democratized. The build phase isn't. And that gap is where the real winners of this cycle will be decided.

Most of the conversation around @SpaceX's IPO is fixated on the numbers: a record $75 billion debut, rising to $85 billion once demand pulled in additional shares. But the dollar amounts miss the larger story. SpaceX's IPO is a marker of where public market appetite is moving next. The last great public market wave was built on the internet: search, social, mobile, and commerce. The next will be built on intelligence, compute, and the physical infrastructure that makes AI possible. SpaceX is an early example of the shift, but its success shouldn’t be seen as directly comparable to @OpenAI or @Anthropic. Those labs build models, but SpaceX builds the scaffolding underneath: satellites, launch capacity, networks, power, data centers, and compute. That's a scarcity layer – more capital-intensive, harder to copy, and potentially more durable. The numbers make it clear. SpaceX generated $18.7 billion in revenue in 2025, mostly from @Starlink and connectivity. But the forward revenue mix looks different: if fully realized, cloud and compute deals with Anthropic and @Google could amount to $26 billion in annualized revenue just by themselves. That reality changes the way to think about the company. SpaceX isn’t a rocket company or a satellite company – it’s quickly becoming infrastructure for the entire AI economy, and the market is already pricing the shift. At 100x trailing revenue, SpaceX is trading as critical infrastructure with enormous unbuilt upside. That multiple also becomes strategic currency, giving SpaceX richly valued stock to reach up the AI stack and acquire compute, models, and applications and build an intelligence flywheel. The Cursor deal is one example: compute at the base, AI coding product at the application layer, with data flowing back down to improve models that need more compute. SpaceX captures value at both ends. It’s still early days, and the risks for SpaceX are real: one indispensable founder, weak shareholder control, heavy capital needs, a stretched multiple, and a thin 4-5% float that can create meme-like trading dynamics. But a step back from the trading noise reveals a company sitting at the center of two extraordinary stories: humanity's reach into space, and the infrastructure powering the next era of intelligence. SpaceX may be the first company public markets are pricing not as a business, but as foundational technology for an entire economy. That’s a big idea — and a very high bar.

Always a pleasure, @RThorDonovan — thanks for having me. And thanks to @StackOverflow for sharing.

Venture capital is not always patient capital. But these days, many VCs are funding Neolabs — AI organizations that operate more like private research institutions than startups, doing foundational, open-ended work where the breakthrough isn't certain and the timeline isn't predictable. It's a topic and point of tension I was excited to explore on stage at yesterday's @Forbes Iconoclast Summit. Think Bell Labs or PARC, not Series A. While this moment for AI is genuinely different — the capital, talent, compute, and data are coming together in ways they haven't before — it's still early days. These organizations are being funded by LPs with an actual time clock that need liquidity at a certain point. Yes, there's real potential for breakthrough, but investors should still be asking: is our capital structure the right fit for this kind of research? Thank you to Maneet Ahuja (@WallStManeet) for including me in this year’s programming, to Katherine Schwab (@SchwabKatharine) for the sharp moderation, and to fellow panelists Eric Wilmes of GIC and Michael Anders of @ICONIQCapital for a conversation that didn't shy away from the hard questions on AI valuations, the AI IPO race, and what it means to invest responsibly in this moment.

Honored to have joined @NYSE Live at #HumanX2026 to talk about what's actually different this time in AI (interview starting at 53:58). After 30 years of hype cycles, the data, compute, and talent are finally arriving together. That's the shift worth paying attention to.

AI is moving us toward a societal “ideal state” of frictionlessness. But not all friction is equal. While some friction does point to pointless busywork, other types of friction push people to actually develop judgment, intuition, and expertise. Cognitive psychologists call it desirable difficulty. The pitch for most AI startups is: "We eliminate the low-value work so you can focus on the high-value work." But a lot of what gets labeled low-value is where expertise forms. The lawyer drafting routine contracts is learning to spot risk. The analyst building models by hand is developing intuition for how businesses work. But it doesn't stop there. These implications go beyond the workplace. Democracy runs on friction too — deliberation, disagreement, sitting with competing ideas rather than receiving easy certainty. When everything around us optimizes for ease, civic participation can start to feel unreasonably demanding by comparison. Wrote on the @Principal_VC Substack about what this means for the AI products being built right now. You can check it out here: principalvc.substack.com/p/f….

Voice assistants in cars usually fail the moment you lose signal. Our portfolio company @liquidai and @MercedesBenz are changing that. Liquid just announced a multi-year partnership to embed foundation models directly into next-gen MBUX systems in North America, with much of the speech and language processing running on-device. Faster responses. Better performance in low-signal areas. More privacy. Less reliance on the cloud.

The Korean government just made its second-ever direct investment from its National Growth Fund. The recipient: @upstageai, our @Principal_VC portfolio company building the future of enterprise AI. Big couple of months for them: • $120M Series C → unicorn status • Named to @FastCompany's "World's Most Innovative Companies of 2026" list • Korea's first recognized frontier model Their flagship, Solar Pro 2, ranks #12 on @ArtificialAnlys's global intelligence index, trained on just 31B parameters. For comparison: @Grok, which tops the list, was trained on an estimated 1.7T. You don't need to out-scale the giants to beat them.

@Allbirds' stock jumped over 580% yesterday after the company announced it was pivoting from eco-friendly shoes to AI infrastructure. If that reminds you of the meme stock era, you're not alone. The market tends to reward companies that align themselves with the most powerful narrative of the moment. Right now, that's AI. The market dynamic is often the same: narrative over fundamentals. GameStop and AMC in 2021 were the clearest examples. Their fundamentals were essentially unchanged, but their equity value was reset by narrative and coordinated buying. From a venture perspective, I can give Allbirds' rebrand the benefit of the doubt, but the inevitable questions are: Where is the founder-market fit? What's the differentiated technology stack? What's the path to competitive advantage? Of course, not every company pivoting to AI is created equal. Some companies, such as Bitcoin miners, are better positioned to make moves into AI. Miners have always been in the business of pointing compute at whichever workload pays most. We saw the same behavior in the MMORPG era, when operators redirected infrastructure to bots and in-game currency farming because they paid better than mining. Those companies are built to dynamically reallocate compute based on economics. As a former footwear company, Allbirds has a very different starting point. Narrative can reprice a company overnight. But only real capability and customer value sustain it over time.

When something gets cheap, we don't just do it a little more. We do it in volumes and for purposes no one imagined. In the early 2000s, every photo cost money to shoot and develop. Humanity took about 80 billion a year. Then the iPhone made photography free. Today we take over 1.4 trillion. A 17x explosion. People didn't just take the same photos cheaper. They photographed everything: meals, parking spots, random whiteboards. Photography stopped being deliberate and became a reflex. A replacement for memory itself. AI is now doing this to software. For 50 years, the real cost of software wasn't servers. It was people: skilled engineers, long dev cycles, a hard wall between those who could build and those who had ideas. Most useful software never got built because it couldn't justify the cost. That wall is falling fast. A nurse, a logistics manager, a shop owner can go from idea to working prototype in hours for nearly nothing. The world is about to fill with software the way it filled with photos after the iPhone. Compute demand is still in its early innings. A trillion photos needed towers, bandwidth, and cloud storage. A flood of new software will need the same. But abundance creates real risk. When only trained engineers built software, human judgment was baked in. As building becomes accessible to everyone, that layer thins. More applications means more attack surface, more ungoverned code, more things that can fail quietly until they don't. The answer isn't to slow down. It's to build governance alongside the technology, not after something breaks publicly and expensively. Are we ready for what's coming?

I've spent over a decade investing in technology, and the single best filter I've developed sounds like an insult. I don't fund the slickest pitch in the room. I fund boring. Here's what I mean. The flashiest startups tend to follow a pattern: big launch, tons of buzz, rapid hiring, then a quiet fade when the unit economics never work out. I've watched it happen dozens of times. The companies that actually returned capital? They looked boring from the outside. They solved unglamorous problems. They built infrastructure: energy-efficient chips, model compilers, coordination layers, the plumbing everything else depends on. They had paying customers before they had a pitch deck. They grew steadily instead of explosively. The founders behind those companies are boring too, and I mean that as the highest compliment. They're the ones still obsessed with the same problem years later, long after the hype cycle moved on. Recently, a founder told me after our meeting, "You were the only VC that made me think critically during the pitch." That stuck with me. Too many pitch meetings are performances. Both sides actually need a thought partner who pressure-tests assumptions, not applause. A few signals I've learned to look for over the years: *The founder talks about margins before market size *The product exists because customers asked for it, not because a trend report predicted it *Growth is coming from retention and referrals, not paid acquisition *The business has a real moat, proprietary data, defensible infrastructure, not just a flashy app sitting on top of someone else's model The AI bubble forming right now isn't about the technology. The underlying science is real. The bubble comes from human behavior: pattern recognition driving investors to rush in, demand to "be in the game" outpacing the number of companies with truly transformative technology and talent. In that environment, "boring" is a sorting mechanism. And the returns belong to those willing to embrace it. See the entire breakdown of my thesis here: principalvc.com/blog-post-en….