AI skeptics have long been concerned with the losses and small revenues of AI software companies. In June 2024, almost three years ago, Sequoia partner, David Cahn, estimated that the AI industry needed to generate roughly $600 billion in annual revenue to justify the money being spent on AI infrastructure including data centers and Nvidia GPUs.
Last month, market research company, Gartner, said that AI companies need close to “$2 trillion per year in revenue by 2029”, token consumption of between 50,000 and 100,000 times its current rate by 2030, and “a 10% profit margin per token.” With huge losses and small revenues, it is not likely that AI companies will achieve these goals on time.
What’s going on here? Usually, companies charge their customers enough money for them to pay their suppliers, and for those suppliers to pay their suppliers. This isn’t happening in AI, however. OpenAI and other AI companies have set prices much lower than their costs to spur demand under the hope that more companies will use AI and then the AI companies will gradually increase their prices, or at least they can.
Those hopes have led to huge losses for OpenAI, which are regularly discussed by Ed Zitron (here is a recent blog). He also covers Anthropic’s losses, which aren’t as big as those for OpenAI, but are still big enough to worry (See here).
These failed hopes have also led to financial shenanigans such as circular financing. For instance, Nvidia invests money in an AI company such as OpenAI and as part of that investment, OpenAI is required to spend that money on Nvidia’s chips.
Other chip and cloud companies are also doing this for OpenAI and other AI software companies to the extent that attempts to show all the investment and purchase connections on a single chart were incredibly complex by late last year.
The latest earnings reports highlight new hype and contradictions from circular financing. The new investments by Google in Anthropic this year are further subsiding Anthropic’s service and that has led to an increase in Anthropic users, an increased valuation for Anthropic and thus increased earnings for Google. These companies are creating higher valuations out of thin air! As long as investors play along, of course.
The bottom line is that circular financing is intended to confuse not enlighten investors and there is less real investment going on than the AI companies are disclosing. Furthermore, please remember that Nvidia’s huge profits have more to do with OpenAI subsidizing its users than Nvidia’s great product, which unfortunately, is lost on many investors.
A new solution that has been pushed by AI companies for more than a year is to use a metric called ARR, or Annual Recurring Revenue, which is a simple way for startups to look like their revenues are bigger than they really are. This is one month of revenue multiplied times twelve, and many companies update ARR every month. Remember that when a company announces their ARR in August of 2026, they are not saying that it will have those revenues in 2026 but actually between July 2026 and June 2027.
Why is this a problem beyond it forcing us to do some mental gymnastics? Simply put, revenues are wildly volatile because of promotions that are constantly made to spur revenues. For example, a startup might count a free three-month “pilot” as three months of real revenue when the user likely will unsubscribe before the payments begin.
I don’t know about readers, but I do this all the time with websites. A website gives me a big discount for a few months, and I mark the unsubscribe time on my calendar, which then sends me a notification the day before I must unsubscribe to the paid plan. Or a startup might write in a contract that the customer will start paying for a certain feature after it’s built. The startup then counts revenue from the months during which the feature is being built. But there’s just no guarantee the feature—or the revenue—will ever come to fruition.
More recently, a Wall Street Journal article entitled “Can Investors Trust AI Sales Figures” said that “joint ventures by OpenAI, Anthropic and others look as if the companies could be paying partners to use their software, not selling it.” These JVs involve OpenAI and top private equity companies such as TPG, Bain Capital, Advent International, Brookfield and Goanna Capital purportedly selling OpenAI software subscriptions to companies owned by PE companies.
OpenAI is also putting money in; it is giving PE companies about $1.5 billion to distribute the software and PE companies are giving those companies they own about $4 billion to purchase the software. It all looks fishy.
Behind this fishiness, OpenAI’s CFO revealed last week that OpenAI: 1) missed multiple monthly revenue targets; 2) missed its internal target of 1 billion weekly active ChatGPT users by end of 2025; 3) ChatGPT’s share of generative AI web traffic fell from 87% a year ago to 65% in January while Google’s Gemini rose from 6% to 22%; 4) and OpenAI is losing ground to Anthropic in coding and to enterprise customers.
Fraud is nothing new to new technologies. Numerous founders have been convicted for fraud in the last five years. Elizabeth Holmes defrauded investors in Theranos as did Trevor Milton in Nikola, Charlie Javice in Frank, Christine Hunsicker in CaaStle, and Do Kwon, in Terraform Labs.
The difference with AI is that the fraud is likely the biggest ever because the AI bubble is the biggest ever, $35 trillion by one estimate. The size of the bubble has invited every type of huckster or grifter, both big and small, and much of the fraud will likely not emerge for years.