Selling compute beats buying it
Money now flows to firms that sell compute, not firms that buy it. The compute bottleneck keeps growing worldwide, so the providers - the pick-and-shovel sellers - earn more than the companies digging for the gold. Cash goes to those who supply the services everyone needs, over those who keep building AI models that rivals can match.
When does AI spending stop being an investment and start being a risk to hyperscaler balance sheets? The shift is already visible in who gets rewarded. Alphabet (GOOGL) posted negative free cash flow in its latest earnings, a real concern. Its core business, strong as it is, can no longer pay for these research plans on its own.
If the research pays off later, isn't a bit of leverage fine? Yes. These companies are taking on more debt to fund the work. The bet is a race to AGI - true artificial general intelligence. Whoever builds it first sees profits and cash flow soar far above every rival. Alphabet took out about $25 billion in senior note loans. The math: pay the interest through around 2060, and by the time the loans come due, the company expects to already be making huge profits. Amazon (AMZN) and Microsoft (MSFT) reported a week later and told that story better than Alphabet did.
NVIDIA as kingmaker
NVIDIA (NVDA) is helping customers finance purchases of its own chips while outside investors carry much of the risk.
What should investors worry about in that model? NVIDIA, Goldman Sachs (GS), KKR, Blackstone (BX) and other large finance firms signed a memorandum of understanding. It lets NVIDIA back AI labs, frontier companies, and startups. Instead of raising equity, these firms take out a loan for the chips, then use them to build their labs. The deal also lets pension funds and large endowments join in, not just private equity or venture capital. It opens up the balance sheets of big corporations in ways old private equity and venture routes could not.
Does this make NVIDIA the true kingmaker of AI? It makes a strong case. NVIDIA has led in GPUs and compute for about a decade and through most of the AI race. Now it can pick winners and losers - it is more likely to finance a company working on something it wants to see. Supplying the wood of the shovel, the shovel itself, and finding gold in the same process means huge profits. A $500 billion credit line lets NVIDIA steer where big financing deals go. Its very high market cap and strong credit rating are what made that $500 billion deal possible. AMD is a good business but unlikely to close the gap, given NVIDIA's lead and brand strength.
There are worries about circular financing. Setting that aside, if the result is real efficiency gains and new capital streams, the risk may be worth it.
CoreWeave and infrastructure as a service
CoreWeave's (CRWV) backlog passed $100 billion and it raised full-year guidance - does that build confidence, or is converting it to revenue still in doubt? Both. Large backlogs are common in the AI race; Micron (MU) is another example. A backlog is future revenue and profit, but turning it into cash quickly is hard, so it is a long-term play. CoreWeave is a GPU cloud provider. As NVIDIA expands hardware, cloud providers like CoreWeave can keep filling backlogs because demand is so strong. As long as CoreWeave supplies the goods and services AI companies need, it can likely meet the backlog and post strong earnings quarter after quarter.
Is the rise of infrastructure as a service a response to the uncertainty of software as a service and a wish for steady, annuity-like revenue? That makes sense. A second driver is domestic demand - many data centers are being built inside the United States. The goal is keeping infrastructure in the US to help US-based businesses, deglobalizing the tech frontier, and backing American companies. That fits the American dynamism sentiment pushed over the last couple of years.
