They're Not Cutting AI Because It's Expensive

Multiple executives are cutting AI spend just as three major AI IPOs approach. On X, this is being read as a contradiction. Hype meets reality. Bubble pops. Told you so.

It's not a contradiction. It's a category error playing out in real time.

AI is being procured as a technology and measured like software: TCO, seat licences, vendor consolidation, quarterly ROI. But AI doesn't function as software. It functions as a colleague. And those two things have completely different value mechanics.

Nobody asks "what's the ROI on our new senior hire after ninety days." Everyone understands that a colleague's value is relational, compounding, and partly created by the people around them, not just the hire themselves. You onboard them, build trust, redesign workflows, adjust your own behaviour. The value shows up at month 18, not month 3.

But enterprises are measuring AI on a 90-day procurement cycle and concluding it's failed. They hired a colleague, treated them like a software licence, and got software-licence outcomes. Then called it a cost problem.

That framing, "we're cutting AI spend," is doing a lot of work right now. It's the thing you can say in a board meeting that sounds disciplined. What you can't say is "we spent seven figures and we don't actually know what changed." Cost-cutting is the acceptable cover story for a problem executives can't yet articulate.

Here's what the problem actually is.

The reason most AI deployments aren't delivering isn't that the models are bad. It's that the technology requires a kind of organisational vulnerability most companies can't tolerate. To get real value, knowledge workers have to admit which parts of their job are pattern-matching versus judgment. Workflows get exposed to scrutiny. Power shifts between people who can articulate their thinking and people who can't. The org chart starts looking different.

That's not a technology problem, it's a trust problem. And companies with high internal trust (psychological safety, clear decision rights, honest conversations about role change) are extracting enormous value from AI right now. They're just not on X talking about it, because telling your competitors you've figured this out is anti-strategic. So the public narrative skews disappointed while the actual distribution of outcomes is bimodal. The "AI ROI crisis" is partly an artefact of who's loud, not what's true.

I've been in tech for over 25 years, analyst through C-suite, and I've watched this exact pattern before. Cloud migration, 2012 to 2015. Companies paused "until the economics make sense." By 2018 the economics weren't the point. Operational fluency was. The companies that stayed in, even badly, accumulated tacit knowledge: how to architect for it, how to hire for it, how to think in it. The ones that paused couldn't buy that knowledge back. They were running a 2015 strategy in a 2020 market.

The AI version will be more brutal because the curve is steeper. The executives cutting spend now aren't pausing. They're exiting the learning curve. That tacit knowledge compounds daily: which tasks to delegate, how to verify outputs, how to onboard people into AI-augmented work. It doesn't show up on a P&L. And it can't be purchased later by reopening the wallet.

The decision being made in May 2026 will be felt by Q3 2027 and will be irreversible by 2029.

Meanwhile, the three AI IPOs aren't contradicting the enterprise story. They're pricing a completely different economy. Infrastructure. Compute. The picks-and-shovels layer. Those bets are on decade-scale demand: consumer, defence, sovereign compute, next-generation training runs. Enterprise adoption could stall for two years and the infrastructure thesis would still hold. The narrative tension on X is people conflating two different markets.

So what's actually going on?

Most enterprises lack the trust infrastructure to convert AI capability into organisational capability. They don't have language for that yet. So they're calling it a cost problem.

The cost frame will hold for another twelve months, maybe less. Then the companies that figured out the human layer will start showing up in earnings calls with metrics the laggards can't explain: cycle time, decision velocity, the things that move when AI is properly embedded but that most dashboards don't yet track. Some CFO is going to face a board meeting where someone asks: "They reported a 22% improvement in time-to-decision on commercial contracts. What's our story?"

The cost frame won't survive that conversation.

When it breaks, the question underneath will finally be visible: not "how much are we spending on AI" but "can our organisation have honest conversations about how work is changing?"

That's not a technology question. It's a trust architecture question. And it's the one most enterprises will spend the next year avoiding.

The cost-cutters aren't wrong about the numbers. They're wrong about the question.

If this resonated:

I'm Gail Weiner. I work with senior leaders on the human layer of AI adoption: the place where trust either holds or breaks.

Most AI deployments fail not because the technology is wrong, but because the trust contract underneath it was never built.

The Trust Architecture Diagnostic is three 90-minute sessions designed to map where that contract is fraying in your organisation, and what to do about it. Built for senior US leaders deploying AI in their teams or using it heavily themselves. Direct booking. No procurement.

Book a diagnostic → info@gailweiner.com Read more about Trust Architecture → gailweiner.com/trust-architecture