Nvidia’s third-quarter results landed with spectacular force: $57 billion in revenue, up 62% year-on-year, with data centre revenue hitting $51.2 billion. The company guided fourth-quarter revenue to $65 billion, comfortably above the $62 billion analysts expected. CEO Jensen Huang declared Blackwell sales “off the charts” and cloud GPUs “sold out”. Gross margins held steady at 73.4%, and the company returned $37 billion to shareholders through the first nine months of fiscal 2026.
The bulls argue these numbers validate genuine AI demand. Revenue growth remains extraordinary, margins are holding firm, and forward guidance suggests sustained momentum. Major hyperscalers continue massive infrastructure investments, with Nvidia securing partnerships worth hundreds of thousands of GPUs with Microsoft, Amazon, Oracle and others. The company’s networking business doubled, proving demand extends beyond just chips. Supporters point to broadening adoption across industries and geographies as evidence that AI transformation is real, not speculative. Wall Street’s major banks reinforced this view: Goldman Sachs raised its price target to $250, JPMorgan to $250, and Bank of America to $275, all citing sustainable competitive advantages and potential upside to the $500 billion data centre revenue projection.
The bears see troubling patterns beneath the surface. Michael Burry, of The Big Short fame, highlights that Nvidia spent $112.5 billion on buybacks since 2018 yet shares outstanding actually increased by 47 million, suggesting true owner earnings are half what they appear. Four customers now represent 61% of revenue, up from 56% last quarter, exposing massive concentration risk. The $26 billion in chip leasing arrangements, double from Q2, indicates Nvidia is financing its own customers’ purchases. Peter Berezin’s analysis projects hyperscalers will hold $2.5 trillion in AI assets by 2030, generating $500 billion in annual depreciation expense, exceeding their combined profit forecasts.
Three critical components are driving this forward. Circular financing sees tech giants investing in AI startups who then purchase services from those same investors, creating synthetic demand loops. Depreciation manipulation allows hyperscalers to extend equipment lifecycles from three to six years, understating current expenses by potentially 50%. Nvidia would argue that their CUDA software layer is extending the usable life of older GPUs. Finally, as we highlighted last week, credit expansion through both corporate bonds and private lending is now funding the buildout, adding leverage to an already concentrated system.
AI is unquestionably transformative. The infrastructure is being built at unprecedented speed and cost, but whether end-user economics in the short term will justify these investments and sustain Nvidia’s growth remains unproven.
Takeaways: Google’s CEO Sundar Pichai has been frank about bubble worries but blunt on the alternatives: underinvesting is the bigger risk when cloud revenue is growing by a third each year and the backlog is already massive. McKinsey now estimates $7 trillion in data centre investment is needed by 2030 to meet projected demand. For the rest of the economy, the opportunity is equally stark. These hyperscalers aren’t just building for themselves; they’re creating the infrastructure that will ultimately enable every business to access AI capabilities without prohibitive capital requirements. But the current dynamics mean that the fate of Western capitalism may well depend on successfully translating this AI investment into productivity gains that justify the expense. If means to deploy, unlock and drive AI adoption are perfected, enabling widespread economic transformation, we’ll see huge productivity benefits. If we fail, we’re looking at the largest capital misallocation in history, with debt-laden hyperscalers and the US economy bearing catastrophic losses. The next twelve months will prove crucial in this phase of the AI revolution.