The AI Bubble - Shady Behavior

The AI sector is being celebrated as the next industrial revolution, but a more uncomfortable truth is emerging: much of today’s “growth” is not organic demand, it’s circular financing. Companies are investing in their own customers, who then spend that money right back with them, inflating valuations and creating the illusion of unstoppable momentum.

Nvidia is the clearest example. The company has poured billions into OpenAI and Anthropic, firms that immediately turn around and buy Nvidia GPUs at unprecedented scale. Nvidia’s CEO Jensen Huang has now publicly stated that its investments in OpenAI and Anthropic will “likely be the last,” acknowledging the circularity as both labs approach IPOs.

The dynamic goes deeper. Nvidia explored a $100 billion OpenAI deal structured so that most of the money would flow back to Nvidia through GPU‑backed infrastructure commitments. Analysts flagged the structure as “murky,” noting that OpenAI would deploy massive Nvidia‑powered data centers, effectively guaranteeing revenue from the very company Nvidia was funding.

And this isn’t isolated. Across the ecosystem, Big Tech is simultaneously investor, supplier, and customer. Microsoft invests in OpenAI and Anthropic; both commit tens of billions to Azure. Amazon invests $4 billion in Anthropic; Anthropic becomes a major AWS Trainium customer. Google invests billions in Anthropic; Anthropic buys Google TPUs. Nvidia invests in CoreWeave; CoreWeave spends billions on Nvidia GPUs. The money flows out, and then flows right back as guaranteed cloud or chip revenue.

Companies are also inflating the cycle by blurring the line between generative AI and agentic AI. They market today’s text‑and‑image generators as if they already possess autonomous reasoning and multi‑step execution, even though true agentic systems remain largely experimental. That conflation lets vendors justify ever‑larger compute budgets for “agentic workloads” that don’t yet exist, feeding the same revenue loop: hyperscalers sell capacity for future capabilities, labs raise capital on promises of imminent autonomy, and investors price in productivity gains that are still theoretical.

This is the definition of a bubble: valuations rising not because of real market pull, but because capital is circulating inside a closed loop of hyperscalers and model labs. Analysts now warn that these structures echo the telecom bubble, where vendor financing inflated demand until the system collapsed under its own weight.

AI will transform industries, but today’s financial architecture is not sustainable. Until the sector breaks this loop and proves genuine, external demand, we’re not witnessing an AI revolution. We’re watching a bubble form in real time.