By David Sherman, AI and Financial Inclusion Strategist at io.net
Wall Street has discovered the GPU. For years, cloud companies and data centre operators bought chips, installed them in racks and rented out the capacity. Nvidia now wants to turn that process into a repeatable investment product. The company is working with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on financing platforms designed to mobilise more than $500 billion in third-party capital.
The chips were already assets sitting on corporate balance sheets. The new model turns the income generated by those chips into something large investors can underwrite at scale. A pension fund or private equity firm can finance an AI data centre based on years of expected usage revenue, much like a toll road, airport or block of flats.
That should produce more infrastructure. It also gives the people financing compute a strong preference for predictable returns, giant customers and long contracts. Startups seldom bear those things.
The little guy gets the leftovers
The compute market already has a pecking order. Nvidia supplies the hardware, and hyperscalers control the cloud infrastructure. Frontier AI labs and global enterprises can reserve huge amounts of capacity years ahead. That leaves startups, independent researchers, and smaller businesses bringing up the back of the line.
A large company can reserve thousands of chips based on a forecast for 2028. A startup needs to show revenue, customers and a viable product before it can afford the technology required to create those things. Compute has produced one of technology’s catch-22s: prove the idea works, then receive the tools needed to prove the idea works.
The consequences go deeper than a larger cloud bill. The SaaS era allowed founders to build a basic product on cheap rented servers, test it with customers and spend more as demand grew. AI brings the infrastructure bill forward. A meaningful product may require expensive GPU capacity before its first customer arrives or its technical approach has even been proven. Founders choose models based on available GPU hours. Researchers trim experiments until they fit the budget. Investors quietly move away from products that look expensive to test. The market filters ideas before customers, science or usefulness get a vote.
Nvidia gets paid twice
Nvidia has spotted the financing problem and developed an interesting answer. Its revenue-sharing and credit-support programme helps AI cloud providers finance Nvidia systems and sell the resulting capacity to emerging companies. The arrangement brings infrastructure online faster and gives smaller cloud providers access to capital. It also extends Nvidia’s economics further into the life of every GPU. The company earns its usual revenue from selling the hardware, then takes a share of the cloud revenue generated by supported capacity. Selling shovels was already an excellent business. Nvidia now gets a piece of the gold. The programme may create more supply, but supply routed through the same handful of companies preserves the same pecking order.
The strange abundance of scarce compute
The current compute shortage has a fairly curious feature. Much of the capacity we already own is sitting idle. Cast AI analysed tens of thousands of production clusters across AWS, Google Cloud and Microsoft Azure and found average GPU utilisation of just 5%. We have the GPUs. Mostly, they sit there.
The companies buying them have enormous power over where that capacity goes. Amazon, Microsoft and Google account for 63% of enterprise spending on cloud infrastructure. The five largest hyperscalers are expected to spend nearly $800 billion on infrastructure this year. That purchasing power allows them to reserve entire data centre buildouts years ahead. Everyone else shops from whatever capacity remains.
The industry has perfected the art of financing, buying and reserving compute. Its next job is to make that capacity move. Decentralised networks have already shown that GPUs spread across different locations can become useful AI infrastructure. A broader clearing house could bring those networks together with data centres, specialist clouds and capacity sitting unused under existing agreements.
Imagine a founder submitting a workload to one marketplace. The system finds suitable GPUs across dozens of suppliers, verifies their performance, compares prices and sends the work wherever it can run best. Spare capacity trapped inside individual providers becomes part of a live global market. Startups buy a few hours today and thousands tomorrow. Suppliers earn money from machines already sitting in their racks. Prices become visible, workloads become portable and hyperscalers have to compete for business.
Opening that capacity would change who gets to build. A university team could run an ambitious experiment during a weekend dip in demand. A startup could test a new model before raising millions for infrastructure. A business in Lagos, Lisbon or Lahore could draw from the same pool as a company in Silicon Valley. Ideas would compete on their merits, with compute arriving when they need it.
Who gets to build?
Every major computing shift expanded the group of people able to participate. Mainframes gave way to smaller machines. PCs put serious computing power on ordinary desks. Cloud services allowed startups to rent servers by the hour. Each transition lowered the price of trying an idea.
AI threatens to reverse that pattern. A company can still begin with two people in a small room, but those two people may need the resources of Microsoft before their first product reaches a customer. More investor money will build more data centres, but granting access will decide whether that investment produces a broad economy that benefits everyone.
The next generation of AI companies needs flexible contracts, portable workloads and a market that makes capacity available to new buyers. Compute is becoming an asset class, but its more important job is becoming a resource that the world’s ambitious outsiders get their hands on.
##
ABOUT THE AUTHOR
David Sherman’s work sits at the nexus of strategy, ethics, and the social impact of technology. With a PhD in Religion and Philosophy, he brings both an analytical and humanistic lens to shaping narratives, helping technology companies articulate not just what they’re building, but why it matters, and its broader impact. Currently the AI and Financial Inclusion Strategist at io.net, he has spent over a decade leading strategy, storytelling, brand, and communications across AI, Web3, and SaaS.
