Why AI Is NOT the Next Dot-Com Bubble with Andreas Steno and Mikkel Rosenvold | Macro Mondays

Andreas Steno called the comparison between the current AI buildout and the dot-com bubble weak. Compute capacity is fully utilized or insufficient for demand, and rental prices are rising; during the dot-com era, bandwidth prices were falling. He also contrasted revenue-generating companies such as Anthropic and OpenAI with firms such as Pets.com, which he said had little in common with today’s AI businesses. For Steno, the combination of constrained supply and substantial revenue makes the current investment cycle materially different, though it does not by itself guarantee that spending will earn adequate returns.

Anthropic’s S-1, as Steno described it, shows the company still needs more compute capacity, with constraints potentially behind some of its loss of momentum since June and weaker enterprise momentum. Its margin improved from about minus 255% a year earlier to roughly minus 5% by the end of the first half. Steno said extrapolating from those figures suggests the company may be near cash-flow neutral, but does not establish its third-quarter results; he said Anthropic could become cash-flow positive this year. Revenue grew 15 times in a year while compute costs rose four times, and Anthropic has not passed higher costs on to all API clients.

Rising GPU rental prices are another sign of a supply-demand mismatch, Steno said, despite expectations that newer chip generations would make compute cheaper. H100 rental prices were up 30% this year, while B200 prices had more than doubled. Although GPUs and memory chips are generally modeled with roughly five-year depreciation, rental-price gains have outpaced depreciation for GPUs installed since 2020. Steno distinguished those rising nominal compute costs from the declining quality-adjusted cost of AI: accounting for improvements in model capability, he estimated that token prices had fallen about 95% over two years. If compute costs continue to decline, adoption could accelerate through what he called a “Jevons paradox” dynamic.

The main risk, in Steno’s view, is not that AI has no demand but that capital providers may doubt whether the buildout will generate returns. He described credit markets as having a dot-com-like attitude toward AI financing and said companies must persuade lenders that their spending can pay off. Amazon has issued bonds and Alphabet has tapped capital markets, so he argued that neither can rely solely on free cash flow. The strongest evidence for continued capital expenditure would be specific returns on money already invested. If earnings deliver, he said, credit stress should be manageable; if they disappoint, financing pressure could worsen.

Steno said Meta’s use of a new equity category to fund capital expenditure diluted existing shareholders and helped explain the stock’s decline. He would be surprised if Alphabet, Amazon or Microsoft followed suit, and said they should provide specific return-on-investment figures. Third-party data-center providers, including neoclouds, could add capacity, but he said they face an even greater challenge accessing capital. The financing question therefore applies beyond the largest technology companies: additional capacity may be available, but building it depends on investors and lenders believing that demand will support attractive returns.

Steno said markets were pricing the AI buildout as if it were over, with chip growth expected to slow sharply, an assumption he considered debatable. He saw data centers and power-grid investments as relatively well supported, while other areas faced capital constraints and high bond yields. If the Federal Reserve signaled that the rate-hiking cycle might not last long, he expected a sharp broad-market rebound or short squeeze; if the AI buildout continues, he expected AI-related shares to rally sharply. A low-float Anthropic IPO could also lead index funds to absorb shares and prompt repositioning in the Magnificent 7 and other major index constituents, though he expected less IPO-related pressure than with SpaceX.