Nvidia recently announced agreements with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR that could mobilize more than $500 billion for AI infrastructure. The initiative would provide financing to companies that need Nvidia computing systems but cannot easily afford the enormous cost of acquiring them. The equipment purchased would secure the financing used to acquire it, making collateralized compute a new asset class.
The arrangement resembles automobile financing. A customer borrows money to purchase an expensive asset, the asset secures the loan, and the manufacturer helps make the transaction possible. Nvidia sells more computing systems, its customers acquire infrastructure that would otherwise be beyond their reach, and financial institutions receive assets capable of producing long-term revenue.
The risk is that the arrangement could perpetuate the circular financing that already characterizes the AI sector. Nvidia invests in companies that use the money to purchase systems containing Nvidia chips and extends credit to companies doing the same. AI developers purchase services from cloud providers that borrow money to acquire Nvidia equipment. Companies invest in one another, purchase from one another and use the resulting contracts to support still more borrowing and investment.
This does not make the industry a Ponzi scheme, but it does pose a risk. These companies are building data centers and producing products and services that people use. The financing may prove extraordinarily profitable and work out for everyone.
Nvidia’s announcement may eventually be remembered as an ingenious innovation that accelerated the construction of a transformative technology’s infrastructure, much as GMAC and Ford Motor Credit expanded automobile ownership. It could also unravel as GE Capital did and become a source of vulnerability throughout an intertwined industry. No one knows how it will turn out.
The common criticism of the growth and the financing that fuels it, corporate greed, only looks at half of the equation. AI companies pursue more because someone wants more. If people and companies didn’t demand more, there wouldn’t be a need to supply more. The more interesting issue is what the Nvidia announcement reveals about the unrelenting need to grow.
If we want companies to exercise restraint, we must first determine what restraint requires of us. The AI sector is responding to demand for more, better and faster products. Employers expect workers to use AI, and workers use it to meet those expectations and relieve their workloads. Consumers use it to answer questions, write emails, create images, plan vacations and complete a growing number of ordinary tasks. Governments, universities, hospitals and nearly every major industry are trying to discover what the technology can do for them.
The fundamental question is therefore not whether Nvidia’s financing will succeed. It is: When is enough, enough? When have we automated enough work or achieved enough productivity? At the individual level, when have I scrolled enough, streamed enough, consumed enough or queried enough?
Aristotle recognized the problem almost 2,500 years ago. An activity directed toward a purpose contains the basis for determining its proper measure. We can know how much food is enough because food serves the purpose of nourishment. We know when we’ve had enough because we are full, too much and we’re sluggish, too little and we’re still hungry. But the appetites for wealth, speed, convenience, entertainment and productivity contain no comparable limit. Satisfying them expands the appetite and establishes a new baseline.
Their limits must therefore be established through moral judgment. Aristotle called the capacity to make such judgments practical wisdom which requires deciding what kind of person one should become and whether a particular use of technology contributes to that life. Does it strengthen contemplation or replace it with quick reaction? Does it support craftsmanship or displace it with cheap production? Does it deepen human interaction or substitute an interface for another person?
We cannot delegate those judgments to companies. A company survives by meeting demand and if it refuses, a competitor will satisfy the demand. We cannot demand limitless convenience and then regard the limitless growth of the companies supplying it as an unacceptable moral failure. Companies retain responsibility for their conduct. But there is no institutional substitute for the individual responsibility to impose limits on desires that contain none of their own. If we want companies to exercise restraint, we must first determine what restraint requires of us.