The AI Bubble Is a Big Tech Problem. You Have a Business to Run.

Every few weeks, a new op-ed surfaces warning that the AI bubble is about to crash and take everything with it. Some of these writers are serious economists. Some are tech journalists with a slow news week. All of them are, to varying degrees, writing about a problem that isn’t yours.

Australian small businesses are operating at the application layer. You’re using tools that sit on top of AI infrastructure, not building the infrastructure itself. Whether OpenAI’s burn rate is sustainable is a question for Sequoia Capital and Microsoft’s CFO. Your question is whether the tool you paid $20 a month for helps you do something faster or better than you could without it. Those are different questions.

Where the Money Is Actually Going

Yes, the numbers are enormous. Amazon, Google, Microsoft, and Meta are spending around $725 billion combined on AI infrastructure in 2026, up sharply from last year. UBS flagged that hyperscalers are spending approximately 102% of cloud revenue on capital expenditure. These are real numbers and they describe a real dynamic in the infrastructure market.

They do not describe your business.

The infrastructure investment is happening at a layer several degrees of abstraction above what you’re doing. You’re using SaaS products (ChatGPT, Copilot, Gemini) that sit on top of APIs that sit on top of model infrastructure that sits on top of compute investment. A correction in the infrastructure market doesn’t automatically cascade down to your subscription. There are multiple commercial relationships and contracts between your $20 a month and Nvidia’s capex cycle.

The Cloud Computing Lesson Is Worth Revisiting

We went through this conversation before. Between 2010 and 2012, serious people made serious arguments that cloud computing was overinvested, that AWS and its competitors were burning money to subsidise artificially cheap storage and compute, and that the eventual correction would leave businesses exposed if they’d built on cloud infrastructure.

The businesses that held off on cloud adoption because of those concerns didn’t protect themselves. They fell behind. And AWS pricing has gone in one direction since then: down. The infrastructure investment that looked like a bubble created commodity compute that every small business now benefits from at prices that would have seemed impossible in 2008.

The AI trajectory is likely the same. The capex going into AI infrastructure right now is building something that will get cheaper and more capable as it matures. The companies writing those large cheques are doing so partly because they expect the returns to compound over a decade, not a quarter.

Waiting Is the Expensive Choice

Framing AI adoption as a question of “should I wait until the bubble resolves?” treats inaction as a neutral position. It isn’t. While you’re waiting for clarity on OpenAI’s financing structure, your competitors who adopted early are building institutional knowledge about what works for their specific workflows. That knowledge compounds. You can buy the tool later; you can’t buy the time your competitor spent figuring out how to use it well.

The CEPR AI Bubble Monitor has noted that aggregate productivity data doesn’t yet show a clear return on AI investment across the economy. That’s a macroeconomic observation, not a business-level one. What happens at the macro level and what happens at the level of a specific business using a specific tool for a specific workflow are different things. Plenty of individual businesses were getting measurable value from cloud tools during the years when aggregate cloud productivity numbers were ambiguous.

Use What Works. Measure What Matters.

The legitimate version of the bubble concern (and there is one) isn’t about whether you should adopt AI. It’s about how you adopt it. Building workflows around a single provider, treating AI output as finished work without human review, spending on AI tools without tracking the return: those are genuine risks. But they’re risks of poor implementation, not risks of the infrastructure economics.

The answer to implementation risk is better implementation, not avoidance. Use the tools that work right now at current prices to solve current problems. If pricing shifts meaningfully in the future, you’ll have institutional knowledge to adapt. If it doesn’t shift (if the trajectory follows cloud), you’ll have years of compounding advantage.

The bubble is not your problem. Sitting on the sidelines waiting for someone else’s financial drama to resolve? That one’s on you.

More Sunday Specials: Sunday Specials →

📬 The SmallBizAI Brief

One practical AI tip for Australian small business. Every Tuesday. Free.

Join business owners getting smarter about AI every week.

Subscribe free →

Sources

🐂 Bull case: The AI Bubble Is a Small Business Problem Whether You Think So or Not