A company just raised $800 million without shipping a single AI model. It didn't sign a customer contract you've heard of. It didn't invent anything new. In about a year, its valuation went from $3.3 billion to $8.3 billion.
The company is Together AI. What it sells is access to someone else's chips.
What happened
Together AI runs what the industry calls a "neocloud." It buys or leases GPU capacity, then rents it out by the hour to companies that want to train or fine-tune AI models without buying hardware themselves. This week, it closed an $800 million Series C, backed by Nvidia, Aramco Ventures, Vista Equity Partners, and General Catalyst, among others. Its valuation more than doubled from its Series B.
The pitch is simple. AI labs and enterprises need huge, unpredictable bursts of compute. Buying that capacity yourself means big upfront cost and the risk that today's chip is outdated in 18 months. Renting it removes that risk. Together AI absorbs it instead.
On paper, this looks like a classic "picks and shovels" story: don't dig for gold, sell the shovels. Investors have repeated that logic for two years of the AI boom.
Why it actually happened
Here's what the picks-and-shovels story leaves out.
Picks and shovels only make you rich if picks and shovels are hard to get. In the 1849 gold rush, they weren't scarce, and most shovel sellers didn't get rich either. The ones who did control something scarcer: the only road into town, the only bank, the only store for fifty miles.
GPUs aren't scarce in that sense. Nvidia sells them to anyone with money. Microsoft, Google, and Amazon already run compute fleets bigger than any startup could build. If "we have GPUs you can rent" is your only asset, that asset has no fence around it. A well-funded competitor can copy it in a single fundraising cycle.
So why did investors write an $800 million check into a business built on a commodity?
Because the real bet isn't the hardware. It's whether Together AI can turn a rented commodity into a sticky platform before bigger players decide to compete on price. The chips are the bait. The prize, if it exists, is the layer on top — the tooling and workflows a customer would have to rebuild from scratch to leave.
That distinction is the whole game. Most founders selling "picks and shovels" in their own market never stop to check which side of it they're actually on.
The founder principle
Position is not protection.
Sitting inside a hot layer — GPUs, no-code tools, marketplaces, "AI wrappers," distribution platforms — feels like leverage. Money is flowing past you, so some of it should stick. But money flowing past you isn't the same as money you control.
The market doesn't pay you for being close to demand. It pays you for being far from substitutes.
A company gets real pricing power only when leaving costs the customer more than staying ever did. Everything else is a rental arrangement with good timing — valuable while the timing holds, worth nothing the moment a bigger player decides to undercut you.
The Toll Bridge Test
A simple way to check which one you're building. Three questions, answered honestly:
Replication speed — If a well-funded competitor copied exactly what you sell, how many months would it take them?
Switching pain — If your best customer left tomorrow, would it cost them real rework, or would they barely notice?
Price authority — In your last few deals, who set the price? You, or the market?
Fast replication, painless switching, market-set price: you're running a rental car business. You can still make money, but expect the margin to shrink every year a bigger player enters your lane.
Slow replication, real switching pain, you set the price: you're a toll bridge. That's rare, and worth defending at almost any cost.
Most infrastructure businesses — including most "picks and shovels" plays in any boom — start as rental cars. They only become toll bridges if they deliberately build switching costs on top of the commodity they started with.
Do this today
List what a rich competitor can't buy this quarter. Everything else on your list is rented position, not owned advantage. Plan around that difference.
Time your best customer's exit, in hours of rework, not feelings. Under a day means treat every account as month-to-month, no matter what the contract says.
Check your last three deals. Who proposed the price? If it's always the customer, you have access, not pricing power.
Ask what happens to your margin the day the biggest player in your category competes with you at cost. If the honest answer is "we don't survive," you're renting a moment, not running a business plan.
Name the part of your company that would still have value if the current hype disappeared tomorrow. If nothing would, that's the thing to fix before your next raise, not after.
Closing thought
A hot narrative can fund a rental business for years. It rarely lasts long enough for anyone to notice the difference. Only a real moat gets to find out what happens after the narrative moves on to someone else.