Qualcomm said Wednesday it's acquiring Modularin an all-stock deal worth about $3.92 billion. Qualcomm makes chips. Modular makes none. What Modular makes is software that runs an AI model across whatever silicon you have — Nvidia, AMD, Apple, Qualcomm — without rewriting the code for each one. Qualcomm just paid four billion dollars for the ability to not be locked into anyone. That price tag is the whole story.
Nvidia's lead in AI was never only the chips. CUDA, its software layer, made Nvidia the default place developers build, and porting all that code to another vendor is expensive enough that most teams never bother. The chips are fast. The lock-in is what keeps the customer. Modular, started in 2022 by Chris Lattner — the engineer behind Apple's Swift language and the LLVM compiler — built the counter: a platform called MAX and a language called Mojo that let a team write once and run anywhere, no CUDA required. Buy Modular and you buy the way out of everyone else's lock-in, with your own hardware suddenly a real option for the customer.
Cristiano Amon, Qualcomm's CEO, framed it as a bet on “developer-friendly, horizontal platforms that can run across diverse compute environments and give customers real choice in how and where they deploy AI.” Put plainly: the customer wants out of the trap, and Qualcomm will sell them the door. The money behind that is inference — running models in production, the bill that grows every month as usage climbs — where buyers are hunting for cheaper options and resent being unable to switch. Qualcomm is reportedly also circling chip startup Tenstorrent for $8 to $10 billion, so this isn't a one-off.
What this says about your own moat
Every seller has a moat, and it's one of two kinds. Either customers stay because leaving is painful — the data's trapped, the integrations run deep, the contract's long — or they stay because the thing is worth keeping. The two look identical on a renewal report. Both show retention. The difference only surfaces when someone offers the customer a clean way out. Nvidia is about to learn what its number really was.
For years, switching costs were the safest asset a software company could own, and investors paid up for them. Now there's a buyer willing to spend four billion dollars to manufacture an exit for the customer, and AI keeps lowering the cost of rebuilding what used to be too expensive to touch. Anything you sell whose stickiness rests on “it's a hassle to leave” is a target. If your renewal conversation leans on the migration being awful, you're renting that customer, and the lease is getting shorter.
If you sell
Find out which moat you have before a competitor runs the test for you. Ask the uncomfortable version of the renewal question: if a rival offered to move your customer off you for free this quarter, would they go? If the answer is “they'd stay because switching is a nightmare,” you have a lock-in moat with a clock on it. Spend the next year converting it into the other kind — make the product worth keeping on its own merits, get adoption deep enough that people would miss it, tie your price to an outcome the buyer can name out loud. None of that is new advice. The deadline is what's new.
If you buy
The flip side is a gift, and you should take it. The whole point of a deal like this is to hand you leverage you didn't have. Anywhere you're stuck with a vendor because moving is too expensive, the cost of moving is falling, and you can use that at the next renewal whether or not you ever actually leave. Ask your stickiest vendors what it would take to port off them today, and watch how they answer. The ones building real value will tell you and not flinch. The ones who've been coasting on switching costs will get nervous, and that tells you where to push on price.
Qualcomm just demonstrated, at four billion dollars, that lock-in has a price and someone will pay it to set your customers free. The sellers who sleep fine are the ones whose customers could leave tomorrow and choose not to. Build that, and the next acquisition aimed at prying open your category does nothing to your number.