You know that moment when you open your banking app, stare at a stack of tiny subscriptions, and think, wait… when did basic life become a monthly payment plan?
Now zoom out. One of the most famous AI companies on Earth can feel “everywhere” and still be bleeding cash. That’s the weird part about openAI right now. The product looks unstoppable, the headlines sound infinite, the partnerships look enormous. But the math underneath is loud, and it’s not pretty.
This post breaks down why openAI can grow fast and still look broke, why regular people are tapped out, and why ads and “shop inside the chatbot” deals suddenly start to make sense (even if they make users uneasy).
How can openAI have huge growth and still be running out of money?
The simple explanation is kind of boring, which makes it easy to ignore: revenue and costs are both rising, but costs are rising faster.
Reports going into January 2026 describe openAI as projecting very large losses in 2026 (around $14 billion), even while it reportedly reached more than $20 billion in annualized recurring revenue by late 2025. That combo sounds impossible until you remember one brutal truth: with AI, more usage can mean more cost.
Every prompt has a real price. Not in a poetic way. In a “chips, electricity, servers, and cloud bills” way. If 800 million people show up and ask for help writing emails, coding, tutoring, planning, and generating images, the bill doesn’t politely wait until you figure out pricing.
Older estimates floating around the last year painted openAI as having already burned roughly $8 billion in a prior year, with projections climbing into the tens of billions by 2028 if nothing changes. Some analysts even warn that the cash runway could get tight by mid-2027 without a major shift. When the spending curve looks like that, big-sounding fundraising or giant infrastructure talk can still feel like pouring water into a bucket with a crack.
If you want a snapshot of how extreme the expectations have gotten, TechCrunch framed it as turning $13 billion into $1 trillion. That’s not a normal corporate challenge. That’s a moonshot with monthly invoices.
The simple math: it is not a profit problem, it is a cost problem
Think of it like running a restaurant where every new customer also forces you to buy more expensive ingredients every night. You’d love more customers, but each extra table isn’t “free growth,” it’s higher food cost, more staff, bigger kitchen, more rent. AI works like that, except the kitchen is a data center.
The biggest cost buckets are straightforward:
Compute to serve users (inference), meaning GPUs, cloud contracts, networking, power, cooling.
Training new models, meaning huge bursts of compute and long research cycles.
Top-tier staffing, because the labor market for elite AI talent is still expensive.
Infrastructure commitments, which can be massive even before the first server rack is turned on.
So yes, revenue growth is real. But the core problem isn’t that openAI doesn’t make money. It’s that it can spend money faster than it can collect it, especially when it keeps trying to push the tech forward while also serving a planet-scale user base.
This is also why you see internal urgency around speed and cost. If you want more context on that pressure, this breakdown of OpenAI’s Code Red response to Google Gemini 3 captures the vibe: improve the product, cut the cost per interaction, and do it fast.
Why ads and shopping deals look tempting when subscriptions stall
Here’s the part that gets awkward: a huge user base doesn’t automatically mean a huge paying user base.
The chatter around openAI’s finances keeps circling one simple pattern: only a small slice of users pay, and that slice drives most recurring subscription revenue. In one widely repeated framing, it’s something like low single-digit percentages paying, while those payers account for the majority of recurring dollars. That’s not crazy for consumer apps, but it’s scary when your costs scale with usage.
So when subscriptions stall, two levers start glowing in the dark: ads and commerce.
Ads are often called a “last resort” because they can misalign incentives. If the model makes money when you click or buy, people start wondering what they’re really being shown. And that’s not paranoia, it’s learned behavior from the last 20 years of internet business models.
Commerce integrations are even more tempting because they’re measurable. If a chatbot helps you buy something, the platform can take a merchant fee. Typical affiliate or merchant-style fees in shopping flows are often discussed in the rough range of about 0.5 percent to 6 percent depending on volume and category. And yes, those costs usually show up somewhere, higher prices, fewer discounts, or extra fees that feel “mysteriously standard.”
I keep picturing a simple diagram: your request goes in, the recommendation comes out, the purchase happens, and a tiny cut flows back to the platform. It’s clean. It’s trackable. It’s also exactly where trust can break.
Everyone else is broke too, and that changes what AI companies can charge
openAI’s pricing problem doesn’t exist in a vacuum. It sits inside your grocery bill.
Consumer data out of the 2025 holiday season tells a rough story: about 37 percent of shoppers took on holiday debt, averaging around $1,223 in new debt. A lot of people rolled that debt forward while still carrying last year’s balance. Credit card debt numbers floating around in public discussion often land in the “this is not fine” zone (think thousands of dollars per household on average), and even if you don’t love any one number, the mood is obvious.
People aren’t refusing to pay $20 a month because they don’t value AI. They’re refusing because life already feels like a subscription trap. Rent rises, car insurance jumps, electricity climbs, groceries do that slow, annoying creep. It’s death by a thousand little renewals.
This is why “convert free users to paid” is harder than it looks on a spreadsheet. A tech CEO may see 800 million users and think, if we convert just 10 percent… A normal person sees one more charge and thinks, not this month, maybe never.
Subscription overload meets a paycheck reality
A standalone chatbot subscription was easiest to sell when it felt special and scarce.
Now AI is bundled into everything. Google bakes it into search and phone experiences. Microsoft pushes it through workplace tools. Plenty of competitors offer strong free tiers. The more AI becomes a default feature, the harder it is to convince someone to pay for one separate app, even if it’s the best one.
And honestly, people do math in their heads now. If an AI tool saves me 30 minutes once a week, maybe it’s worth it. If it’s “nice,” it’s gone. That’s the new bar.
The trust problem: if AI recommends products, who is it really serving?
Picture a normal use case: you ask for the best laptop under $1,000. Or the best hotel near downtown. You’re not just asking for information, you’re asking for judgment.
If ads or merchant fees are in the background, the fear is simple: is it ranking what’s best for me, or what pays best? If a “sponsored” label is missing or unclear, users will assume the worst. And once that suspicion settles in, it’s hard to scrub out.
Ethical safeguards don’t need to be fancy to help:
Clear labels when money influenced placement.
Opt-outs for sponsored results.
Basic transparency on whether recommendations were affected by partnerships.
It’s not about perfection. It’s about not turning the model into a smiling salesperson pretending to be your friend.
What this means next: cheaper AI, more bundling, and a fight over who pays
In 2026, the fight isn’t only model quality. It’s also power, chips, and who eats the cost.
The big theme showing up in current reporting is constraint: compute supply, energy availability, data center build-outs, and the simple fact that training and serving advanced models costs a lot, even for giants. At the same time, enterprise buyers are getting stricter. They want ROI, not vibes.
So pricing models are likely to keep shifting. Not because companies are confused, but because they’re experimenting in public.
I keep imagining a photo-style scene of a data center with heavy power lines and transformers nearby, not glamorous, just physical reality. That’s where a lot of this story ends up.
The new business models we are about to see everywhere
Expect more of these patterns:
Free tiers that get tighter, with more limits and more nudges.
Ads in places that used to feel “clean.”
Usage-based pricing that charges by task, output, or tool call.
Bundling into devices, browsers, and workplace suites, because distribution wins.
Transaction partnerships where AI gets a cut when you book, buy, or subscribe.
Big players with existing cash flow can subsidize longer. That’s the quiet advantage of companies that already print money from ads, cloud, or enterprise software. Smaller labs will feel more pressure to partner up, take on debt, or chase huge funding rounds.
If you’re watching the fundraising side, TechCrunch’s note on openAI reportedly trying to raise $100B at an $830B valuation shows how high-stakes the capital story has become.
What to watch for as a regular user (without becoming an AI finance nerd)
You’ll feel the business model before you read about it.
Free-tier limits suddenly get stricter. Features quietly move behind paywalls. Prices jump. Ads get “tested.” Shopping suggestions start feeling oddly pushy, or weirdly specific. “Too good to be true” bundles show up, then change terms later.
My practical rule: keep a backup tool, export anything important, and treat AI recommendations like ads until the product proves otherwise. Also, if you’re curious about the pressure across the whole field, not just openAI, it helps to see competitors’ expectations too, like this TechCrunch report on Anthropic’s revenue projections. The entire sector is trying to grow into its costs.
What I learned using openAI while trying to keep my own budget sane
I use openAI in a very unglamorous way. I use it to compare options, rewrite emails so I don’t sound harsh, summarize long docs, and study things I should’ve learned years ago.
And it works. It saves me time. It helps me avoid dumb mistakes when I’m tired.
But I also felt the mental load of “one more subscription.” Not just the money. The feeling of keeping track. The feeling of, am I using it enough to justify it? Then the free tier changes, a feature I liked gets limited, and I’m back doing the little internal budget debate in my head. It’s small, but it adds up.
The takeaway I keep coming back to is simple: AI is most valuable when it reduces friction in your day, not when it adds a new bill. If a tool saves you real time or prevents costly errors, you feel it. If it’s just fun, it’s the first thing you cut when groceries spike again.
(And yeah, sometimes I still use it for fun. Just… not every month.)
Conclusion
openAI can look rich and still be cash-hungry because compute at global scale is expensive, and scaling usage can scale costs. At the same time, households are stretched, so selling another subscription is harder than Silicon Valley seems to expect. That’s why ads and commerce deals keep creeping into the plan, even though trust gets shakier when money influences answers.
In 2026, expect pricing and limits to keep changing. Use AI carefully, protect your data, keep alternatives nearby, and don’t assume the business model you see today is the one you’ll have tomorrow.
0 Comments