🌊 What a16z’s Speedrun Demo Day Reveals About the Future of AI Startups

Traction-heavy AI startups, OpenAI’s new platform play, and how capital is shifting.

Oct 09, 2025

I’m Ivan. This is a newsletter about startups and investing—for founders and investors. Each week I cover how top 1% startups use ai, raise, and grow.

Hello there - I’m in SF for Tech Week and the energy is… different. Whatever your bubble take, a visit 2x/year is a great way to reset your quality bar.

Here’s what went down in ai + vc world:

🌊 AI

I spent the day at a16z’s Speedrun Demo day (thanks

Ryan K. Rigney

). The level of talent was off the charts. YC demo day was impressive, this was next level. Many of the 50 that came on stage were hitting $1M+ revenue numbers in c.6 months. The selection was almost entirely AI startups. A few of my favourites:

Evercurrent - alignment for hardware dev.
Dispatcher - agentic interface for drones + robotics.
Azimov - engine bringing AI to life.
Nexxa.ai - specialized AI for heavy industry.
Atrios - network intelligence for GTM.
Pencil.dev - ai designer for engineers.
Rork - lovable for mobile apps.
And a few super interesting consumer plays like Taya (wearable AI), Zingroll (ai streaming), Pelago (voice-first wellness), Daedalia (virtual friends).

OpenAI is angling to turn ChatGPT into the new App Store. They launched an App SDK that lets developers build apps inside ChatGPT, meaning you’ll soon say “Canva, make a deck” or “Spotify, make a playlist,” without leaving chat. They also teased a no-code Agent Builder, letting anyone design AI workflows like Zapier, and dropped Sora 2, their video generation model that feels straight out of Pixar. Meanwhile there’s an interesting $1T web of investments happening between OpenAI and the big dogs.

Mercury dropped a report showing where startups are putting their AI budgets: mostly into horizontal apps (60%), core stacks (OpenAI, Anthropic, Perplexity), creative tools, meeting copilots, AI “employees,” and GTM tools. I also found this data on OpenAI customers that have used >1T tokens.

I’ve been thinking a lot about moats in AI. On one hand, it’s incredible how much is being built, faster and cheaper than ever before. On the other, it feels like a seller’s market. Defensibility is messy. Too early and it’s pointless, too late or ignored and it’s a problem. And “execution is the moat” is only half true, and mostly a cop-out.

The guy who built Google’s TPU dropped some spicy takes on AI. His new company Groq just raised $750M to build faster, cheaper AI chips. His view is that compute is the new oil and whoever controls chips + energy controls AI. Labs like OpenAI aren’t limited by users, they’re limited by silicon. And every millisecond faster = more trust, more revenue.

The guys at Acquired deep dived into the history of Google, the same week yet another Googler receives another Nobel prize. They’ve produced 7 to date (wild).

Also, Google is basically the Berkshire Hathaway of tech. It owns 14% of Anthropic and 8% of SpaceX, runs X (its moonshot factory) that birthed Waymo and Wing, and through CapitalG has backed Stripe, Databricks, Duolingo, and more. Plus their M&A record is legendary: DoubleClick ($3.1B) became the backbone of Google Ads, YouTube ($1.65B) is now worth hundreds of billions, Android ($50M) powers most of the planet’s phones, and DeepMind ($500M) turned into a world-class AI lab.

The AI-native ERP space is heating up with 14 startups aiming to use AI as a wedge to attack this (massive, slow, old) market. Check out DualEntry*, one of the latest to raise a big Series A round ($90M).

*Sponsored

🌊 VC

Pitchbook published data on how AI deals are dominating the VC market.

Sourcery just dropped Carta’s new VC fund report. IPOs are reopening, M&A is picking up, and funds are sending money back to LPs. The AI wave is inflating returns, but cash reserves are drying up fast. In fundraising, tiny and huge funds are thriving, and the middle is getting squeezed.

Stanford’s Ilya Strebulaev just shared new data on which VCs actually pick unicorns early. Firebolt Ventures tops the list with a 10% hit rate, Thrive follows at 9%, and Paradigm, Coatue, and Kleiner sit around 7%.

Palle Broe shared a dataset of founder dilution at Seed. Seed valuations in the US typically range from $10M–$15M post-money. Founders usually give up 15–25% ownership, with the median at ~19.5% (Carta).

Europe has quietly sent ~100 unicorn founders to the US. The incentive structure for startups there works. VC-backed companies drive 42% of IPOs, 85% of R&D, and 63% of public market cap in the US. Until we build EU Inc., a true single market with real startup incentives (especially fiscally) our best founders will tend to keep crossing the Atlantic.

In Europe, Sweden keeps punching above its weight, consistently producing more unicorns per capita than almost any other country including Spotify, Hexagon, Klarna, Lovable and Tink.

I published a list of 27 startups from Spain that raised over €250M in September.