Accounting tricks and sky-high valuations: inside the strange economics of AI startups

“Reality is merely an illusion... albeit a very persistent one,” a certain Albert Einstein once said. Novelist Arthur Koestler added that, “Nothing is more sad than the death of an illusion." Well, current thinking in the high-tech world of business is that we may well be being played into believing in that of infinite growth.

Silicon Valley has always worshipped expansion. When markets saturate, the industry conjures a new “next big thing.” After crypto and the metaverse came AI, dominating the airwaves and aspects of our lives, that keeps share prices soaring. As long as investors believe this story, stock can be used like currency, letting big companies buy talent, parts, or entire firms without spending cash. But as analysts point out, the AI boom is more narrative than profit engine.

According to MIT’s August report, 95% of companies experimenting with AI aren’t making any money from it.

What’s up with the Scale-Meta alliance?

Meta’s $14.3 billion purchase of a 49% stake in Scale AI valued the data-labeling startup at $29 billion and lured its founder Alexandr Wang to Menlo Park. The deal gave Meta privileged access to a crucial pipeline of training data – and left competitors like OpenAI and Google wary of feeding a supplier part-owned by a rival. Scale’s new chief executive, Jason Droege, insists business has grown each month since the takeover. He now wants to shift the firm from simply labeling data toward building AI applications for clients such as the Mayo Clinic, Cisco, and the U.S. Army, with whom Scale recently signed a $99 million contract.

The trouble for most corporations isn’t enthusiasm but execution. “Companies thought it was easier than it actually is,” Droege told CNN. Many run flashy pilots that automate nothing of consequence. Projects that succeed tackle tasks humans handle inconsistently – summarizing documents, processing claims, or flagging anomalies – and keep experts “in the loop” to correct errors. Yet for now, AI remains an expensive experiment whose benefits arrive slowly.

Artificial income driving AI riches

Critics like writer Cory Doctorow argue with few punches pulled that the new economy of “artificial intelligence” relies on artificial income. Cloud giants count pre-paid compute credits as revenue; firms trade infrastructure for stock; investors declare paper profits from illiquid shares. Doctorow and analyst Ed Zitron estimate industry-wide spending of more than $300 billion this year against roughly $18 billion in revenue, a mismatch disguised by exuberant accounting . Even the so-called “Stargate” data-center initiative, touted at $500 billion, has only $40 billion in committed funds. You are reading that correctly.

Unlike earlier tech waves, where scaling reduced unit cost, AI’s expenses rise with every new user. More prompts mean more GPUs, more electricity, more cooling. Microsoft, Meta and Amazon each plan tens of billions in AI investment for 2025, yet most of their reported “AI revenue” is recycled compute time or unverified projections. As one analyst at D.A. Davidson put it: “Once companies finally figure out AI at scale, it will be worth it… but that’s years away.”

Get your game on! Whether you’re into NFL touchdowns, NBA buzzer-beaters, world-class soccer goals, or MLB home runs, our app has it all.

Dive into live coverage, expert insights, breaking news, exclusive videos, and more – plus, stay updated on the latest in current affairs and entertainment. Download now for all-access coverage, right at your fingertips – anytime, anywhere.