The AI boom is usually measured in GPUs, data centers and soaring revenue forecasts. Increasingly, however, another number matters: debt.
This week provided two striking examples.
Nvidia-backed CoreWeave priced an upsized $3.7 billion convertible-note offering, while AI cloud rival Nscale disclosed a $3.1 billion convertible financing, including $1 billion from Nvidia, as it filed to go public.
Neither deal means the AI infrastructure boom is in financial trouble. But together they highlight the extraordinary amount of outside capital required to build it.
AI Compute Is Expensive Before It Becomes Profitable
Modern AI infrastructure requires much more than Nvidia chips. Companies need data centers, land, power, cooling, networking and enormous GPU clusters, often before the facilities generate meaningful revenue.
Coinpaper previously examined who is financing the estimated $3.6 trillion AI data-center buildout, where banks, bondholders, private-credit funds and infrastructure investors are increasingly joining Big Tech in funding expansion.
Nscale shows the scale of that equation particularly clearly.
Its IPO filing says revenue jumped 1,252% to $140.6 million during the first half of 2026, while its net loss reached $1.02 billion. At the same time, Nscale had $103.4 billion in active and contracted total contract value by Aug. 31, supporting roughly 461,000 active or contracted GPUs.
In other words, demand can look enormous while the infrastructure needed to serve it remains extraordinarily expensive.
CoreWeave Shows the Other Side of the Boom
CoreWeave is already public, but it continues raising capital aggressively.
Its latest convertible offering was initially announced at $3 billion before being upsized to $3.7 billion. Coinpaper recently examined the company's latest financing push as investors weighed rapid expansion against financing and dilution risks.
This is the fundamental financial question behind the AI race.
Debt isn't inherently a warning sign when companies have long-term customers and rapidly expanding revenue. But billions borrowed today still have to be serviced tomorrow.
AI data-center debt risk reveals what happens if utilization or AI revenue eventually grows more slowly than the infrastructure built to support it.
