The $7 Doritos Moment: Enterprises Are Repricing AI as Discretionary

There is a line in Vin Vashishta’s April essay that should be printed and taped to every CFO’s monitor this quarter: enterprises have started evaluating their AI subscriptions the way they evaluate a $7 bag of Doritos at checkout. A pause. A wince. A quiet decision to put it back on the shelf.

That is not a metaphor about snacks. It is a diagnosis of how the AI industry gets priced in 2026.

The disappointment number nobody wants to print

Writer’s 2026 AI sentiment survey landed with a thud most vendor decks have quietly decided to ignore. Forty-eight percent of executives now describe their AI adoption as a “massive disappointment.” A year ago that number was 34%. Only 29% report meaningful ROI. Only 5% report substantial ROI.

Read those three numbers in order. The middle is collapsing. Either you are in the 5% that can put a dollar figure on what the tooling did for the quarter, or you are drifting toward the 48% that can’t, and the market for executive patience between those two poles has closed.

This is the shape of a product that is being reclassified. Not as infrastructure. As discretionary.

The unit economics under the hood

The labs are not helping themselves.

OpenAI’s numbers, as best we can reconstruct them, look like this: roughly $25 per weekly active user, a 5.5% conversion rate off a base of around 900 million free users, and a projected $14 billion in losses by year-end. That is a consumer funnel with a consumer funnel’s margins, wearing an enterprise logo.

Anthropic is in a different room of the same house. Roughly $211 per monthly user, $30 billion annualized revenue, 70 to 80% of it from enterprise contracts and API seats against a total user base around 20 million. The economics are healthier because the customer is different. But the customer is being asked the same question every month: is this essential, or is this a $7 bag of Doritos?

Both labs responded to the squeeze the same way. They tightened free tiers. They reduced token allocations. They raised the effective price by shrinking the bag.

The Frito-Lay lesson nobody inside OpenAI seems to have read

Here is the part of Vashishta’s analogy that should be keeping pricing teams up at night. Frito-Lay tried this exact move. Walmart, reading its own shopper data, told Frito-Lay to cut Doritos prices. Frito-Lay said no and chose shrinkflation instead. Revenue went negative for the first time in more than a decade. Roughly $50 billion in market value vaporized. The parent company is now unwinding the decision.

Shrinkflation works until the customer notices. Once the customer notices, the conversation is no longer about value. It is about trust, and trust is a much more expensive thing to rebuild than a price list is to cut.

Every token allocation tightened in 2026 is Frito-Lay reducing the bag. Every executive reading a shrinking context window is a Walmart buyer doing the math at the register.

Why this is a governance story, not a pricing story

The 5% of companies getting substantial ROI are not winning because they bought a better model. They are winning because they can see, at the line-of-business level, what their AI tooling is doing, where it is failing, and which humans and which agents are compounding the advantage. The 95% cannot see any of that, which is why they describe the experience as disappointment.

“Disappointment” is the feeling that precedes a cancellation. It is also the feeling that precedes a downgrade to a cheaper tier, a quiet reduction in seats, and a budget memo explaining to the board why the line item is shrinking. Every one of those outcomes is what being classified as discretionary looks like from the vendor side of the invoice.

The separation between the 5% and the 95% is not talent, budget, or model choice. It is measurement. The 5% built the layer that tells them whether the tooling earned its keep this quarter. The 95% are still guessing, still getting vendor dashboards optimized to make the vendor look good, still unable to answer the only question a CFO is asking in April 2026: what did we get for the money?

A measurement layer is boring until the day a CFO asks the question, at which point it is the only thing standing between your AI budget and the shelf where the Doritos went.

What the repricing actually means for a 2026 operating plan

If you are a buyer, your leverage has never been better. Your vendors know their own retention numbers. They know what happens when the 48% starts churning. Walk into every renewal with the 5%-versus-95% question and ask your vendor to show you, in your own data, which side you are on. If they cannot, that is the answer.

If you are a seller, the SaaS playbook you inherited assumed the category was infrastructure and the conversation was about features. Both assumptions are gone. The category is discretionary and the conversation is about proof. You do not win the renewal by shipping a new model. You win it by making the ROI legible before the customer has to go looking.

If you are an operator inside the company, treat the disappointment survey as a forecast, not a snapshot. The companies that will spend confidently on AI in 2027 are the ones that built a measurement layer in 2026. The companies that cut AI lines to the bone in 2027 are the ones that did not.

The honest headline

Software spent twenty years being priced like infrastructure because recurring revenue looked like a utility bill. AI is being priced like a snack because 48% of the executives paying for it cannot answer whether it is worth the bag.

The path out of that room is not a better demo. It is a measurement layer that converts “we think it’s working” into “here is what it was worth this quarter, by team, by workflow, by outcome.” That is the moat. That is what separates the 5% from the 95%. And that is what the next twelve months of AI procurement conversations will be about whether the vendors are ready for them or not.

The $7 Doritos moment is not the crisis. It is the signal. The crisis is continuing to sell, and continuing to buy, as though the category is still infrastructure when the executive writing the check has already reached for the shelf.

---

This analysis synthesizes The AI Labs Have a $7 Doritos Problem by Vin Vashishta (April 2026), the Writer 2026 AI sentiment survey, and reporting from TLDR Founders’ “The SaaS Reckoning: Stock-Based Compensation Was Never Free” (April 2026). The piece updates arguments previously developed in Post-SaaS Economics: The Premium Is Gone with fresh unit economics and a new diagnostic frame.

Victorino Group helps enterprises build the measurement layer that separates the 5% from the 95%. Let’s talk.