My AI bill just went way up—and I’m relieved

AI bills – A technologist running a software studio says an Anthropic bill that recently spiked is a sign the “subsidy” era for AI pricing may be ending. The author argues that artificially low inference costs can push businesses toward permanent decisions that hollow ou

The day the Anthropic bill jumped, it didn’t feel like punishment.

It felt like a correction.

I run a software studio as a technologist, and AI is in nearly everything our agency builds and in most of how we work. So when our Anthropic bill recently skyrocketed, I was happy about it—not because I want higher costs, but because I think the current price signal has been misleading.

The core problem, as I see it, is that the price we pay for AI has been fictional.

In 2025. OpenAI booked around $13 billion in revenue and posted a roughly $21 billion operating loss. spending close to $1.60 for every dollar it took in. That pattern, in my view, isn’t a temporary mistake. The frontier labs are deliberately pricing below what it costs to serve customers. trying to win the market before the market figures out what the service is actually worth. The difference is being covered by venture capital and the cloud giants.

Eventually, that subsidy race has to end. Most analysts expect API prices to climb once the funding and subsidization wind down.

So if AI feels absurdly cheap, it’s because it is—temporarily, and on someone else’s dime.

For me, the relief is practical. A subsidized price lies to businesses about one of the most urgent decisions they’re facing: human work or machine work. At today’s “fake-low” rate, the machine can win arguments it shouldn’t. When the price moves closer to the real cost. the calculus shifts. and some of the “obviously cheaper to automate” decisions quietly flip back toward the human side.

That shift matters because the danger isn’t only that AI becomes expensive. It’s what teams might do before the bill tells the truth.

We’re being invited to make permanent decisions based on a temporary price. If a company clears out writers, support staff, and analysts now—while AI is on sale—it doesn’t just trim a line item. It loses people, relationships, institutional memory, and human judgment that AI still can’t touch.

And once those decisions are made, they don’t reverse neatly. You can re-sign a software contract. You can’t always rehire a relationship.

The pattern is familiar enough to feel like a warning from the past. Amazon undercut local retail on price and convenience until alternatives thinned out. and 90% of independent retailers said it hurt their revenue. After the competition was gone, Amazon gained leverage to dictate terms to the sellers who remained.

In another sector, Google and Meta absorbed ad dollars that used to fund local newspapers. A court found Google illegally monopolized ad tech and “substantially harmed” publishers in the process. The shops didn’t reopen. The newsrooms didn’t refill.

The three-step sequence is what sticks: subsidize or undercut, capture the market, then set the terms.

Knowledge work is the next thing in line, and the companies positioned to capture it include OpenAI, Anthropic, and a handful of others—vendors I use as well, and vendors whose invoices I pay.

When the people running the platforms control the terms, the cost question stops being academic. When a few companies own the operations your business runs on, you pay what they decide, when they decide it.

If you’ve built your funnel on a social platform, you’ve already seen how that feels when an algorithm changes and ad rates rise. Now imagine the same kind of shock—except the thing on the meter isn’t just marketing. It’s your analysis, your code, your writing, your actual thinking.

For leadership, that’s the “can we afford it” question, and it has two meanings: can we afford the bill when it lands, and can we afford to gut the human capacity that would have been our fallback?

There are optimists, and I understand why. Inference costs have been dropping fast, and it’s possible AI keeps getting cheaper and the reckoning never fully arrives. But you don’t bet your company’s entire human side on “maybe the subsidy lasts forever.” That isn’t a strategy.

None of this makes me anti-AI. I’d be out of business if it weren’t useful, and I’d be lying if I said I don’t enjoy the work.

We use it everywhere. I just treat it like a brilliant intern who’s working for free this year and might want 10 times more next year. You lean on that intern for leverage, not for replacement. You keep your people pointed at the things that compound: relationships. taste. judgment. and the decisions that someone has to be accountable for.

And you stay smarter than the tool, because you’re the one in charge. Human expertise has to be running the show no matter what AI costs.

So yes, I’m okay with the bill. Let AI cost what it costs. The sooner it does, the sooner businesses stop confusing “subsidized” with “better,” and remember that the humans weren’t the expensive option. They were the resilient one.

The only real question is whether we figure that out before we let them go—or after.

Lindsey Witmer Collins is founder and CEO of WLCM App Studio and Scribbly Books.

AI pricing Anthropic bill OpenAI revenue operating loss API costs inference costs venture capital subsidies human expertise knowledge work automation business strategy