09 / 08 / 25 – Eddi Weinwurm
Today’s AI’s Fragile Foundations
Choosing an AI product for your company and wiring it into your workflow isn’t a casual decision. It’s closer to a marriage. You’ll go on a lot of “dates” first, demos, PoCs, and presales calls. Each product shows up in their best outfit, promising you incredible useful solutions, savings, and the future on a plate.
But not all potential partners you date are equal when it comes to the long run. The one who rolls up in a flashy sports car might look impressive – until you discover the car isn’t theirs. It’s leased, the bills are piling up, and the whole lifestyle is financed on borrowed money. Unsustainable.
That’s exactly what’s happening in today’s AI market. Many cloud solutions impress with slick SaaS products and attractive pricing, but behind the shine they’re running on investor subsidies, like an ICU patient on infusion. They burn through huge amounts of cash at a loss just to keep the engines running.
Level 1: The Shiny SaaS Layer
Take, for example, a typical AI product, a cool SaaS tool that offers to make your media assets searchable, or to help you generate stories, summaries, and drafts at scale. It looks polished, the pricing seems manageable, and the pitch is simple: unlock productivity.
So you commit. You join the ecosystem. You pay to index terabytes of data, pay to keep it alive, and assume you’re investing in a long-term partner. You adapt your production workflows to fit the new tool – and it all works amazingly ever after. End of the story?
Here’s the hidden truth: the company selling you this service doesn’t make money on you. In fact, they probably lose money every time you upload another hour of content.
That’s not a bug; it’s the business model. Investors aren’t paying for profitability; they’re paying for growth. Revenue today is a proxy for valuation tomorrow. If the bet works, the investors win big. If not, they shut the doors and write it off.
But even worse, beneath this glossy SaaS wrapper lies another surprise: often your provider isn’t even running the AI at all. They’re just a gateway. Your requests get enhanced, wrapped in business logic – and routed to a bigger model provider, OpenAI, Anthropic, or Cohere, through carefully engineered API calls. One more concern, one more possible point of failure.
Level 2: The API Giants Running on Losses
That’s almost common knowledge by now, but even the giants themselves, the ones powering your provider, are not profitable either. At all.
In 2024, OpenAI generated roughly $3.7 billion in revenue but faced estimated costs of around $9 billion. Analysts warn that if costs keep scaling faster than revenues, those losses could grow into the tens of billions in the near future.
The company as a whole bleeds cash, and each API request adds weight to an already unprofitable structure. So when your SaaS provider pays OpenAI for the compute, that money doesn’t cover OpenAI’s costs either. It’s other investor cash making up the difference.
In other words, you’re standing on the shoulders of a giant that might also be losing balance at one point.
Level 3: The Infrastructure Beneath
Dig one layer deeper. OpenAI doesn’t own its own vast server farms. Instead, it rents massive compute capacity from Microsoft’s Azure. OpenAI gets favorable pricing compared to retail cloud rates, but whether Microsoft makes real margins on those deals isn’t public. What’s certain is that the costs are astronomical, and they only go in one direction: up.
Level 4: The Only Valid Business
And at the very bottom of this chain, holding up everyone else, sits Nvidia. Unlike everyone above, Nvidia isn’t running at a loss. It sells GPUs with fat margins, and the scarcity of its hardware has only increased its leverage. In this upside-down economy, the supplier at the root is the only consistently profitable player.
The Consequence?
The chain will break at its weakest link. And all the links between you and the final hardware anchor are glowing red – in numbers. You are betting on a horse, betting on a horse, betting on a horse.
But there is an alternative. You don’t have to depend on this chain of weak links. You can run AI yourself, on-premises, anchored directly to the hardware you own. It may take a bigger upfront investment, but it rests on something far more solid than borrowed money.
Most importantly, you free yourself from the endless inflow of investor cash that keeps today’s cloud AI economy alive. That supply of money is not infinite. A single downturn can flip investor sentiment overnight and shut off the thick hose of funding.
And when that happens, the consequences are catastrophic. You may have already paid your SaaS provider huge sums to index your media assets, an investment that can evaporate from one day to the next. If even one link in the chain snaps, the product at the core of your production goes dark. Instantly. You don’t just lose your investment. Your entire production comes to an abrupt halt.
Therefore, instead of betting in the big game, lured by low-effort, cheap entry, invest in on-prem AI. Own the horse you ride. It won’t betray you.
Discover More
Why Building for Privacy Matters: The Case for Resident AI
A New Urgency Around Data Privacy
LLMS: Start to Think Small
Why Huge LLMs Aren’t the Solution for Everything
AI Agents Are Not the Future of AI
The challenge is to rethink AI entirely!
Discover More
