Artificial intelligence companies are racing to secure one increasingly valuable resource: computing power.
That race has resulted in massive investment in Nvidia (NVDA) processors, data centers, and power. It has also spawned a new breed of specialist cloud providers promising to provide the infrastructure necessary to train and operate ever more advanced AI models.
Now firms are putting a stunning amount behind that demand.
U.K.-based AI infrastructure firm Nscale is telling potential investors that it has about $103 billion in total contracted revenue, according to documents reviewed by The Information and cited by Reuters. The Nvidia-backed startup might launch an initial public offering as early as September.
However, the headline figure comes with a crucial caveat.
Nscale’s contracts are around 5.7 years on average, which implies about $18 billion in yearly contract income. One person involved with the negotiations emphasized that the numbers were “illustrative” and not meant as official revenue projections.
That difference might matter if Nscale looks to public markets to value one of the rising infrastructure providers of the AI boom.
Nscale’s $103 billion figure shows the scale of AI demand
Nscale said it had $103 billion in committed revenue that it will earn over the next few years, not income the business has already booked.
The size is more obvious if you look at one big consumer.
Anthropic will pay $45 billion to lease AI cloud-computing capacity from Nscale’s data center in West Virginia, Reuters reported in late August.
As part of that partnership, Nscale will use Nvidia’s new Vera Rubin processors to meet the processing needs of Anthropic.
Related: Nvidia stock flashes unusual signal for investors
The disclosed Anthropic commitment alone accounts for almost 44% of the $103 billion in contractual income stated for Nscale, based on the two figures provided by Reuters.
That comparison doesn’t necessarily mean Anthropic accounts for exactly 44% of the contracts in Nscale’s investment documents, because the actual makeup and timeframe are not fully detailed.
But it shows how fast AI infrastructure contracts are getting into the tens of billions of dollars.
For Nscale, long-term deals may provide something very significant ahead of a prospective IPO: proof that big AI clients are ready to commit billions of dollars to processing power years in advance.
They also provide an execution challenge.
Nscale still needs to install the processors, networking, power, and data-center capacity to deliver on those contracts.
Nvidia sits at the center of Nscale’s expansion
Nscale’s collaboration with Nvidia adds another spin to the situation.
The startup is backed by Nvidia and wants to employ its Vera Rubin processors for Anthropic’s processing requirements, according to Reuters.
This puts Nscale in a much bigger ecosystem that is growing around Nvidia’s technology.
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
Nvidia doesn’t need to run all the AI data centers to profit from the infrastructure boom. Specialized cloud companies may acquire finance, lock up customers for long-term contracts, and put a lot of money on Nvidia systems to give that capacity.
As such, firms like Nscale might be another signal of future AI demand for Nvidia investors.
The chipmaker’s largest clients are huge tech corporations, but infrastructure experts might be another source of demand as AI developers seek alternatives to developing all of their computer infrastructure themselves.
The Anthropic transaction also highlights why the next generation of technology from Nvidia matters beyond the company’s own quarterly performance.
Those deals may provide investors a view into demand for Nvidia hardware that might be used years from now, since AI cloud businesses are already signing multiyear contracts around systems powered by Vera Rubin.
The number you can see: Nscale’s $103 billion contract haul.
The deeper story is how much infrastructure may need to be built to fulfill it.
Nscale’s massive number comes with a major caveat
Contracted revenue is not the same as revenue that has already been reported on an income statement.
If Nscale goes public, it may be one of the most significant differences for investors.
The contracts are for an average of nearly 5.7 years, or around $18 billion in annualized contractual income, according to records seen by Reuters.
A source familiar with Nscale’s investor discussions also said the figures were illustrative rather than formal revenue guidance, The Information reported.
That proviso applies because Nscale still has to convert those long-term commitments to recognized income.
AI infrastructure is very capital-intensive. Before a provider can provide computing power to consumers, it requires modern processors, energy, networking equipment, cooling systems, and data-center architecture.
The optimistic case for Nscale’s $103 billion number is simple. AI developers are investing enormous sums to ensure future processing capacity, and a supplier funded by Nvidia has found itself at the center of that expenditure.
The more cautious reading is also significant.
Much of the economic value in these contracts remains in front of us.
If Nscale goes public, investors won’t merely be wagering that demand for AI will continue to be robust.
They’ll be wagering that the startup can leverage a huge collection of long-term promises into real income, while constructing enough Nvidia-powered infrastructure to actually deliver what consumers have pledged to purchase.
Related: Nvidia’s $96 billion quarter revealed a surprising constraint
