The AI Cost-Cutting Narrative Is Loud. The Spending Data Tells a Different Story.

Over the past few months, a new narrative has taken hold across the tech world: the era of unconstrained AI spending is over, and businesses are starting to tighten their belts.

The anecdotes have been piling up. Microsoft reportedly banned Claude Code internally. Uber's CTO noted that the company had burned through its entire annual budgeted token spend before the year was even halfway over. And on OpenRouter — the routing service that much of the developer community relies on to move between models — inexpensive Chinese open-weight models have been steadily gaining share.

For the first time, the conversation around AI has shifted. The models keep getting more capable, but capability is no longer the only thing decision-makers are weighing. Cost has firmly entered the equation.

‍

Then Came Kimi K3

The launch of Moonshot.ai's Kimi K3 model sent a shock through the internet. According to some reports, it rivals the power of Anthropic's Fable model — while coming in at a notably lower price point. Demand was so overwhelming that the company temporarily paused new subscriptions, unable to muster the compute needed to keep up.

For anyone watching the cost narrative build, Kimi K3 felt like a potential inflection point. If a cheaper, open-weight model could genuinely go toe-to-toe with the frontier labs on performance, wouldn't the more price-sensitive corners of the market be the first to make the switch?

That question is exactly what we set out to answer.

‍

The Frontier Labs Aren't Going Anywhere — At Least Not at the High End

There's ample evidence — both in our data and in public commentary from enterprise decision-makers — that the largest enterprises will always demand frontier models. For these organizations, performance, reliability, and vendor relationships tend to outweigh incremental cost savings.

But the more interesting question sits one tier down. What about tech-focused small and mid-sized businesses (SMBs)? This cohort is typically more cost-conscious, more nimble, and — in theory — more willing to experiment with cheaper, open-source alternatives. If a spending shift toward open-weight models were happening anywhere, this is where we'd expect to see it first.

So we looked.

‍

What the Data Actually Shows

We examined our SMB cloud spending panel of nearly 3,000 customers, capturing roughly $3 billion in spend across AWS and Google Cloud.

The result was unambiguous: we're not seeing a shift play out whatsoever.

Gemini and Anthropic dominate spending across the panel. Meanwhile, spend on Kimi, DeepSeek, Qwen, and other open-weight models remains minimal — a rounding error relative to the frontier incumbents.

Despite all the headlines about cost discipline, all the buzz around cheaper open-weight challengers, and all the anecdotes about budgets being reined in, the actual dollars are still flowing overwhelmingly toward the frontier labs — even among the cost-conscious SMB cohort where we'd most expect defection.

The Takeaway

The gap between narrative and behavior here is striking. It's a powerful reminder of just how dominant the frontier labs have become — and how deeply entrenched their customer relationships are as they compete head-on for share.

‍The cost narrative is loud. The spending data is louder — and the frontier labs still own the room.

‍Learn more about how YipitData helps customers across the AI, Cloud, and Software industries here.

‍