If you run a 15-person law firm or a 40-person shop and you've been happily paying $20 or $30 a month for an AI tool that does the work of a part-time employee, I want you to ask a different question than usual. Not "is this tool good." Ask: why is AI pricing so cheap right now, when the compute behind it is some of the most expensive infrastructure ever built?
Dave Blundin, who runs Link Ventures and has built and sold real companies for three decades, posted something on LinkedIn this week that gets at exactly this. He pointed out that Microsoft, Google, Amazon, Meta, and xAI are spending on AI infrastructure faster than they're earning from it. Their capital expenditure — the data centers, the chips, the power contracts — is outrunning their cash flow. His point is simple and a little uncomfortable: that math only works as long as investors keep funding the gap. It's not a stable business model. It's a bet that growth will eventually catch up to spending.
Why Your $20/Month AI Tool Doesn't Actually Cost $20 to Run
Think about what's actually happening when you use an AI writing assistant, a legal research tool, or a customer service bot. Every query you send runs on a GPU cluster that cost someone tens of billions of dollars to build and costs a fortune in electricity to keep running. The company selling you that $20/month seat is not pricing based on what it costs them today. They're pricing based on what they need to charge to keep you, beat a competitor, and grow fast enough to justify the next funding round. The real cost of serving you is higher than what you're paying. Someone is covering the difference — right now, that's venture capital and public market investors betting on future dominance, not current profit.
This isn't a conspiracy theory. It's how nearly every subsidized-growth business has worked. The gap gets filled by someone with deep pockets who's playing for market share, not margin.
Dave Blundin's Warning: The Big Five Are Burning Cash Faster Than They Make It
What makes Blundin's post worth paying attention to is who's saying it. He's not a pundit. He's spent 35 years building and running companies, including as founding CEO of two firms and chairman of a third. He knows the difference between a company that's investing ahead of demand and one that's structurally dependent on outside capital just to keep the lights on. His read is that the five companies building most of the world's AI infrastructure fall into the second category right now. That's a big claim about the biggest companies on earth. It should worry anyone whose business runs on their tools.
What Happened Last Time a Subsidy Era Ended
We've seen this movie before. Cloud storage was nearly free for years while providers fought for market share — until consolidation happened and prices crept back up. Rideshare pricing was famously subsidized for almost a decade; Uber and Lyft lost billions to train customers into a habit, then raised prices once the venture money dried up and profitability became the mandate. In both cases, the customers who built their operations assuming the subsidized price stayed cheap forever got squeezed hardest, because they'd made the tool load-bearing without a backup plan.
AI tooling is following the identical script, just compressed into a shorter timeline because the capital amounts involved are so much larger.
The 18-24 Month Window Before Prices Reset
Here's the mirror for your business specifically. A 15-person real estate brokerage using an AI tool to draft listing descriptions and follow-up emails. A dental clinic using an AI scheduling and intake assistant. A small manufacturer using an AI system to write quotes and manage inventory forecasts. In each case, the owner is looking at a bill of maybe $50 to $300 a month and thinking of it as a fixed, small cost of doing business — like a phone line.
It isn't. It's a promotional rate. The investors funding your vendor's losses are not going to fund them forever. At some point — Blundin's numbers suggest we're already well into this cycle — growth targets tighten, and pricing has to catch up to real cost. When that happens, it won't be a gentle 10% bump. It'll look like the cloud storage and rideshare corrections: a 2x or 3x jump, sometimes overnight, sometimes buried in a new "tier" that quietly removes the features you actually use unless you pay more.
Four Moves to Make Before Your AI Bill Triples
First, lock in annual contracts now where the vendor offers them, even if it feels like overcommitting. A locked annual rate is effectively insurance against the reset.
Second, avoid building anything mission-critical on a single vendor's proprietary API without a fallback. If your law firm's document review process only works inside one tool's specific workflow, you have no leverage when prices move. Keep your data exportable and your prompts and workflows documented somewhere outside the tool itself.
Third, test at least one alternative vendor every quarter, even if you don't switch. Knowing your second-best option costs you an hour and gives you real negotiating power later.
Fourth, separate the AI tools you use because they save real money from the ones you use because they're a novelty. If a tool isn't clearly paying for itself today, it's the first one to cut when the price triples — don't let it become load-bearing.
How to Build Your Budget Assuming Prices Double or Triple
When you're doing next year's budget, don't model your AI spending as flat. Model it as doubling, possibly tripling, within 18 to 24 months, and check whether the tool still earns its keep at that price. If a $30/month tool that saves your paralegal five hours a week still pencils out at $90, keep it and don't worry. If it only works at today's subsidized price, that's useful information now, not a surprise later.
The businesses that get hurt in every subsidy correction are the ones that assumed cheap was permanent. The ones that do fine are the ones that priced in the correction ahead of time and built their margins to survive it either way.
