I froze for a few seconds when I saw this news last night

I'd been using AI trained by Poolside. When I saw NVIDIA gutted them for $12 billion, I froze.

A 115-person startup built a model that could go toe-to-toe with top-tier AI. But they couldn't handle the pressure of not raising $2B for GPUs. This got me rethinking — what's our actual relationship with AI as non-dev indie operators?

What is Poolside? Why did NVIDIA poach them?

Poolside is an AI brain-training startup. Founder Eiso Kant was bragging on the Latent Space podcast just last month: "We built top-tier models with under 70 people. 115 people total covers all engineering and research."

Reality was brutal: at the end of 2024, they had 6 weeks to raise $2B for 40,000 GPUs (the chips that train AI). The money didn't come through. The cluster collapsed. Training stopped. The founder's internal email said: "Three and a half years of being right on direction — but the capital requirements became a ladder to heaven."

Jensen Huang's play: license your model factory to me, have 109 employees sign on with me, value the whole deal at $12B. The founder stays for a new direction. Investors get their money back. Employees get golden parachutes. First time the industry's seen this — they call it a "reverse execuhire." Normally executives leave, employees stay. This time employees leave, founder stays. The direction got flipped.

What does this have to do with your work today?

Replication cost (this is an industry signal, not a tool):

  • Money: $0
  • Time: 5 minutes to read this news
  • Barrier: knowing that an "AI model" is the brain that makes AI smart is enough
  • First step: search "Poolside NVIDIA $12B" to find the original reporting

But what's more worth thinking about: honestly, I got stuck here too — I never seriously considered how many people are behind Cursor, Claude. The tools you use daily might be this kind of tiny team. What if they get poached tomorrow, training stops, or your tool's price jumps 20%? Do you have a backup? Giants are grabbing talent and compute. Where do we stand? That's the real question I asked myself after reading.

Advice by stage

Just starting / still figuring it out: If you haven't made a dollar from AI yet, this doesn't directly affect you. Cursor, Claude, Notion AI — their core teams are stable. Keep using the existing tools. Don't stress.

Got 1-2 clients / stable side income: I'd suggest you build an "AI dependency list" — which work steps are locked to which tool? If prices jump 20%, can you switch to an alternative within a week? That's a low-cost moat you can build right now.

Scaling up / want to grow bigger: Poolside's 115-person top-tier model story teaches me one thing: we can't build models, but can we learn from them — make "using AI to deliver content/services" as efficient as a small factory? Your personal brand is your model factory. Don't just be an AI user. Start being a small "factory owner" of AI.