Wall Street's Problem, Not Yours
Every few months a new report drops warning that AI is overvalued, overhyped, or heading for a correction. The headlines get shared, the think pieces multiply, and somewhere in every leadership meeting, someone raises a hand and asks: "Should we slow down? What if it's a bubble?"
Let me be direct about this, because I think the bubble debate is doing real damage to real businesses. The question of whether AI company valuations are sustainable is a question for venture capitalists and public market investors. It is not a question for the leader of a 40-person professional services firm who is trying to figure out whether to automate their client onboarding process.
Those are completely different questions. Confusing them is costing you.
What the Bubble Argument Actually Says
To be fair, the concern is not entirely without basis. In mid-2025, Sequoia Capital published an analysis — widely circulated in tech circles — estimating that the AI infrastructure build-out had created roughly a $600 billion revenue gap: the difference between what had been invested in AI infrastructure and what was actually being generated in revenue by AI applications. The implication was that the sector was running far ahead of monetisable value, and that a correction of some kind was inevitable.
That analysis has merit as a capital markets observation. Sequoia was right to flag it. When you build massive infrastructure ahead of demand, you typically see consolidation, write-downs, and a thinning of the competitive field. We've seen versions of this in every major technology cycle — from the dot-com era to the cloud buildout to the mobile app explosion. Infrastructure overshoots, then demand catches up, then the market normalises around the players with durable business models.
That process is messy for investors. It is largely irrelevant for operators.
The Productivity Data Doesn't Care About the Bubble
Here is what is not in question: the productivity gains from AI tools at the point of use are real, measurable, and accumulating fast.
GitHub's internal research showed that developers using Copilot completed coding tasks 55% faster than those who did not. McKinsey's work on knowledge-worker productivity found that AI-assisted workers handled 14% more client interactions per hour with measurably higher satisfaction scores. Researchers at MIT and Stanford have independently documented productivity gains in the 20-40% range for writing-intensive tasks when workers use AI assistance versus not.
These results don't depend on Nvidia's stock price. They don't change if OpenAI's valuation corrects by half. They exist at the workflow level — in the time it takes someone to write a proposal, summarise a meeting, analyse a dataset, or draft a client communication. The value is in the doing, not in the financing of the infrastructure.
This is a crucial distinction that gets lost in the bubble discourse. The AI bubble debate is about whether AI companies are overvalued. It has almost nothing to do with whether AI tools are useful. The internet bubble of 2000 didn't make email less effective. The overbuilding of fibre-optic cable in the late 1990s didn't make broadband less transformative once the market cleared.
What a Bubble Correction Would Actually Mean for You
Let's play it out. Suppose the pessimists are right and a meaningful correction happens. What does that look like for an SMB operator?
Some AI companies consolidate or shut down — particularly the ones burning capital on infrastructure with weak monetisation. The large, well-capitalised players — Microsoft, Google, Anthropic, OpenAI — likely get stronger, not weaker, as the competitive field thins. Pricing on AI tools may actually come down as the market matures and competition for enterprise customers intensifies. The tools themselves continue to improve regardless of who is funding the research, because the underlying capability gains are driven by model architecture and training, not by venture rounds.
In that scenario, the businesses that built genuine AI-powered workflows during the growth phase come out ahead. The ones that waited for the market to settle come out behind — having ceded a year or two of compounding productivity gains to competitors who moved earlier.
Waiting for certainty is itself a strategic choice, and it has a cost.
The Right Question for SMB Leaders
The question worth asking is not "is AI overvalued?" It is: "What are the two or three workflows in my business where AI could save five hours a week, and what would it take to build them?"
That question has a concrete, actionable answer. It doesn't require a view on monetary policy, GPU demand forecasting, or the capital efficiency of large language model training runs. It requires honest knowledge of your own operations and a willingness to run a short experiment.
The macro debate about AI investment cycles will resolve itself the way all such debates do — slowly, then all at once, with plenty of noise in between. In the meantime, the businesses building real AI capability into their day-to-day operations are doing something valuable and durable regardless of what happens to the Nasdaq.
That's where your attention belongs. Not on the bubble. On the workflow.