Are we in a GPU bubble? Doug O'Laughlin's case that this bull market has further to run — and who gets hurt when it ends

Summary

Doug O'Laughlin argues the AI infrastructure bull market has room to run into end of 2026. He maps the current cycle to 1998 rather than 1999, suggesting the most dramatic gains still lie ahead. In 1998, the September rate cut preceded the true frenzy of 1999. O'Laughlin reads the recent Fed rate cut the same way: a green light for aggressive spending that won't peak for another year or more.

Three structural conditions could sustain the bull market. Productivity metrics are moving in AI's favor. Q2 GDP growth hit 3.8% and labor productivity is accelerating. If new highs in productivity materialize, "there will be almost no limit to the justification of AI spending," O'Laughlin writes. This mirrors the dot-com era, when top-line economic validation justified sprawling capital expenditure.

Data center buildout is economically multiplicative in ways past tech booms were not. GPU investment is geographically distributed across power-rich regions, spreading spending across multiple counties and creating political support across jurisdictions. No natural constituency is unified against the spending.

Most critical is supply chain alignment. Today's boom is driven by the largest, most profitable companies in the world, not unprofitable startups. Meta and Google, both founder-controlled, are willing to go capital-intensive. Oracle is pushing into negative free cash flow to grab GPU share from hyperscalers. The hyperscalers themselves are swimming in cheap capital. Apple, Google, Amazon, and Meta all borrow at spreads of only 50 basis points over U.S. government bonds. Microsoft pays just 5 basis points above Treasuries.

O'Laughlin highlights Sam Altman's strategy to make OpenAI so central to the entire ecosystem that everyone becomes vested in its survival. The more Altman raises and the larger his spend becomes, the more he locks in the incentives of chip suppliers, cloud providers, and memory makers. If a company doesn't participate in the OpenAI coalition, it risks slower revenue growth or irrelevance. The Nvidia deal and recent Korean memory partnerships are visible expressions of this logic.

O'Laughlin compares it to a borrower-lender dynamic: if you owe the bank $1 billion, it's your problem; if you owe $10 billion, it's the bank's problem. Altman has scaled OpenAI's obligations to a point where his success is now embedded in the balance sheets and stock prices of the entire tech sector.

Meta and Google remain wild cards. Both have the tools and capital to escalate spending beyond free cash flow, which O'Laughlin frames as "rocket fuel for the next stage of the AI trade." Amazon is positioned weakly. Its Anthropic strategy is faltering as Anthropic shifts to Google TPUs away from Amazon's Trainium chips. Google is not yet playing at full magnitude but "that changes soon," O'Laughlin predicts.

The vulnerability, if it materializes, may not be fraud but arithmetic. A venture fund underwriting ten AI companies valued in the billions will likely see one wind down or fail. That gets absorbed as normal portfolio loss. Individual collapses won't puncture the broader cycle. The 2008 crisis saw Lehman Brothers and Bear Stearns fail while JPMorgan, Goldman Sachs, and Morgan Stanley survived. The constellation of forces—government policy, founder control, monetary conditions, and alignment of interests—suggests the cycle runs longer than most expect.

One unresolved tension concerns Intel. Sam Altman publicly called for TSMC to expand capacity rather than entertaining a domestic alternative, but the geopolitical logic of building a TSMC-equivalent fab in the U.S. capable of serving both Nvidia and AMD remains compelling. Such a move would require political will and Intel's execution. O'Laughlin does not resolve this but flags it as a potential inflection point.