One type of corporate expenditure is visible to the marketplace, whereas another type stealthily transforms sectors while no one notices. For the past year, Nvidia has done both simultaneously. Total supply, including inventory, purchase commitments, and prepayment arrangements, has increased to almost $145 billion, according to the headline figure from this week’s most recent results call. That amount was $95.2 billion three months prior. It was only $16.1 billion a year prior to that. That rate of growth in numbers typically indicates one of two things. The corporation is either making a historic wager or it sees something that the rest of us do not.
Although the scope of Nvidia’s approach is indeed astounding, it isn’t the most remarkable aspect. It’s the underlying architecture. Instead of purchasing businesses outright, Jensen Huang and CFO Colette Kress have been discreetly putting together a parallel type of vertical integration by placing strategic checks at almost every stage of the AI supply chain. In January, CoreWeave received two billion. The next week, Nebius received two billion. IREN, a data center operator, could receive up to $2.1 billion. Corning, a 175-year-old glass manufacturer that is now at the forefront of optical networking, might receive up to $3.2 billion. Coherent, Lumentum, Synopsys, and Marvell. a $30 billion pledge associated with OpenAI. With seven months remaining in 2026, the corporation had invested more than $40 billion in equity by early May.
The dollar quantities don’t fully capture the intrigue of the strategic reasoning. In a recent joint event with Dell, Huang stated that memory chips are currently the bottleneck and that there is a real lack of high-bandwidth memory, the HBM stacks that enable modern AI accelerators. For advanced manufacturing slots, TSMC, the company that makes Nvidia’s chips, has been requesting lengthier contracts and higher upfront payments. There are supply constraints in the optical components sector that were unthinkable just two years ago. In response, Nvidia has taken direct investments in the businesses it depends on, locked in capacity through purchase agreements, and structured those agreements to guarantee the partners continue to use Nvidia technology. It’s supply chain management disguised as empire-building.
It’s important to consider how this approach is altering the AI ecosystem as a whole. In only one quarter, AMD’s purchasing commitments more than doubled, reaching $21 billion. In March, Broadcom informed investors that it had “secured the supply chain required” to generate $100 billion in sales from AI chips the following year. In order to match supply with future purchase promises,
Amazon has been discreetly offering warranties to vendors of optical components, such as Fabrinet and Applied Optoelectronics. These actions are not taking place in a vacuum. The whole AI infrastructure sector seems to have come to the conclusion that scarcity is a structural issue that must be overcome rather than a temporary issue to wait out.
Some people don’t think this ends well. Investor Michael Burry, who gained notoriety in 2008 for shorting the housing market, has been publicly likening Nvidia’s present trajectory to that of Cisco around 1999. His comparison focuses on the nature of the obligations rather than the technology, which is real. When the demand for networking equipment fell, Cisco eventually wrote down around 40% of its supply chain agreements. Most investors don’t realize how much of Nvidia’s current stock value is represented by its $145 billion in total supply obligations. Current valuation models do not account for the severe unwind that would occur if AI capital expenditure ever reduced, even little.
Some of these layouts have a cyclical character that is difficult to ignore. Nvidia makes a CoreWeave investment. Nvidia chips are purchased by CoreWeave. Nvidia makes an investment in IREN. IREN constructs data centers with Nvidia hardware. Nvidia gives OpenAI a $30 billion commitment. Through the hyperscalers, OpenAI’s computational expenditures eventually find their way into Nvidia’s revenue stream. Nothing here is fraudulent, and nothing is inherently harmful. However, this does indicate that Nvidia is increasingly contributing to the market for Nvidia chips. This interplay makes typical revenue forecasting more harsh than usual and is more difficult to model than a straightforward supplier-customer connection.
The upcoming quarterly report is the next test. The purchase-commitment line has been monitored by analysts for indications that the growth rate is slowing down or picking up speed. It’s been too esoteric to worry with, so most people don’t even formally forecast that number, but in the last year, it has emerged as one of the most instructive figures in the whole AI industry. Purchase commitments rising once more indicates that Huang and Kress continue to believe that demand is greatly outpacing supply. As Kress has been telling investors, if they plateau, it may indicate that the company has finally secured sufficient capacity for the upcoming quarters. Either reading is important. Both leave open the question of whether this is Wall Street’s most stunning pre-correction commitment to yet or the most prophetic industrial plan of the decade.
