WeWork raised $22 billion, peaked at a $47 billion valuation, and filed for bankruptcy in November 2023. SoftBank alone lost $14.4 billion. The coworking company didn't fail because coworking was a bad idea — it failed because the money propping up its growth never connected to a sustainable business underneath.

The AI industry has a version of this problem, and it's getting harder to ignore.

Bloomberg recently mapped the circular deal structure connecting Microsoft, OpenAI, and Nvidia. The pattern is striking. Nvidia committed up to $100 billion to OpenAI. OpenAI's CFO Sarah Friar acknowledged that the money "will go back to Nvidia" in GPU purchases. Nvidia also backs CoreWeave, which buys Nvidia GPUs to build data centres, then sells capacity back to OpenAI. The money moves. Whether it actually goes anywhere is a different question entirely.

Tom Tunguz drew an explicit comparison to Nortel's vendor financing during the telecom bubble — a company that lent money to its own customers so they could buy its products. Nortel's revenue looked real on paper. Until it didn't.

WeWork had the same circularity, just cruder. SoftBank invested billions. WeWork used those billions to sign long-term leases on buildings it didn't need yet. The expansion justified the valuation. The valuation justified more investment. Adam Neumann called it a "community company" and a "state of consciousness." The market called it a $47 billion technology company when it was a landlord with a beer tap.

The AI version is more sophisticated. The companies involved are profitable elsewhere. Microsoft and Google have cloud businesses generating hundreds of billions. Nvidia sells real products to real customers beyond the AI startup loop. And unlike WeWork — which was locked into leases it couldn't escape when demand fell — data centres have repurposing options. You can run cloud workloads, render farms, scientific computing. I keep reminding myself of this whenever the parallel starts feeling too neat.

The differences matter. I'm not arguing this is WeWork reborn.

What I am arguing is that the circular financing pattern should alarm anyone who watched a bubble before. When revenue from Company A depends on investment from Company B, which depends on revenue from Company A, the system is more fragile than the topline numbers suggest. The spending gap — $527 billion in, $51 billion out — looks especially precarious through this lens.

OpenAI is projected to lose $14 billion in 2026 while seeking another $100 billion in funding. The company that started the whole frenzy still can't make the economics work, even after turning to advertising despite its CEO calling the idea "uniquely unsettling" barely a year earlier.

WeWork's original sin wasn't ambition. It was the gap between the story and the balance sheet — the willingness to let growth narratives paper over unit economics that never worked. SoftBank kept writing cheques because the alternative was admitting the previous cheques were wasted. The AI industry hasn't reached that point. But the circular deals, the vendor financing, the ever-growing commitments justified by ever-larger projected returns — the architecture of the bet looks familiar.

The hardware is different. The founders are different. The technology does more real things for more real people. But money that goes in circles still ends up back where it started.

Sources: