NVIDIA Founder Jensen Huang: Investing in Cutting-Edge AI May Be the Best Opportunity in History, a Once-in-a-Generation Chance
NVIDIA founder and CEO Jensen Huang stated in an interview on CNBC's "Mad Money" that cutting-edge AI labs like OpenAI and Anthropic may be companies that appear once in several generations. NVIDIA aims not only to be a supplier but also an investor and long-term partner.
He noted that these companies are experiencing rapid growth in customer and token usage, and their services are already profitable, with the main constraint on expansion being computing power. With insufficient financial history and difficulty in low-cost financing, NVIDIA is in a position to increase support. He mentioned that when there was an opportunity to invest, more should have been put in; now that the flywheel is turning, helping them expand computing power is also helping them increase revenue.
He even referred to such investments as one of the best ever and described it as a platform shift that occurs once in generations. In the past six months, approximately $400 billion in venture capital has flowed into AI startups. He emphasized that NVIDIA's graphics processors are versatile; if one customer encounters issues, computing power can be transferred to other customers, with residual risk lower than perceived by outsiders.
The company is simultaneously promoting an "AI factory" that can be financed: partnering with large asset management institutions to create an independent financing platform, mobilizing over $500 billion in third-party capital to build infrastructure. OpenAI currently has and plans to commit about 12 gigawatts of NVIDIA computing power, while the Ohio PORTS-Pike project is initially around 4.25 gigawatts, with discussions to expand to about 16 gigawatts, corresponding to an estimated $600 billion computing opportunity by 2030. Each generation of factory systems corresponds to about 1.5 million graphics processors.
Buyers are labs looking to secure production capacity and infrastructure funds seeking rental income, while sellers are platform companies selling chips, equity, and space simultaneously. Funds flow from lab revenues and institutional debt equity to wafers and electricity. Beneficiaries are suppliers who can resell cards to the next customer; those under pressure are labs that have not yet been able to borrow at public offering rates. This represents a capital structure statement of suppliers becoming shareholders, not just an announcement of a single completed investment.
Source: Public information
ABAB AI Insight
Huang Jensen has rewritten NVIDIA from selling cards to selling factory years. Lab tokens are starting to make money, but they do not yet have a long enough financial history to queue in the bond market, which is precisely the gap filled by chip manufacturers with existing scale and supply chain visibility. Investing and supplying locks customers onto the same stack: the more cards sold, the more equity resembles a call option; if a customer switches stacks, the equity is damaged first, while the cards can still be subleased. He expressed regret for not investing more, framing the already realized valuation increase as a reason for the next round of investment.
The capital path involves wafer prepayments, long-term data center contracts, and third-party infrastructure funds. The motivation is to make "computing power equals revenue" a collateralizable asset, extending the cycle from one order per quarter to twenty years of upgrades. OpenAI building its own factory and Anthropic seeking other chips indicate that the largest customer is reducing reliance on a single supplier. NVIDIA is hedging this layer with an open-weight model and full-stack software while also becoming a potential competitor to its customers.
This is similar to Intel investing in OEMs, Qualcomm investing in mobile brands, and Boeing using supplier financing to bind airlines. The industry is transitioning from expansion to control: training demand is still rising, while inference costs are beginning to matter. Those who can prove that cards can be resold and electricity can be upgraded will be able to sell data centers to pension funds, not just to startups.
The essence is capital concentration. The mechanism is to package scarce advanced packaging and power locations into quotas that labs must pre-purchase. Demand is no longer a constraint; quotas are. After suppliers become shareholders, bad debt risk shifts from accounts receivable to long-term equity; general-purpose chips turn single-customer defaults into reconfigurable inventory.
ABAB News · Cognitive Law
- Demand is no longer lacking; what is lacking is operational cards.
- Supplier equity stakes turn orders into call options.
- When cards can be resold, customer risk is repriced.