ML in Production · 3 min read

NVIDIA's $115.8B Quarter Asks a Question Sales Can't Answer

NVIDIA's Q3 FY2027 revenue hit $115.8B, up 103%, but the real question is whether customers — not investors — are paying for the GPUs. CoreWeave's $6.3B backstop looms.

By Grace Kim674 words

Key takeaways

  1. NVIDIA reported Q3 FY2027 revenue of $115.8 billion, up 103% YoY, with data center revenue of $107.6 billion (up 110%) and net income of $65.3 billion.
  2. NVIDIA agreed to buy up to $6.3 billion of CoreWeave's unused GPU capacity through 2032.
  3. Hyperscaler capex is estimated at ~$730 billion for 2026 (Amazon $220B, Alphabet $200B, Microsoft $170B, Meta $140B), up from ~$413.6 billion in 2025.
  4. NVIDIA and partners including BlackRock, Apollo and KKR target over $500 billion in third-party capital for AI infrastructure, with residual-value support up to 25%.
  5. Vera Rubin NVL72 systems have started shipping to CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.
NVIDIA's AI Funding Loop: Is the AI Boom Creating an Artificial Demand for GPUs?
NVIDIA's AI Funding Loop: Is the AI Boom Creating an Artificial Demand for GPUs? — Pexels / pixabay

NVIDIA booked $115.8 billion in revenue for its fiscal 2027 third quarter, ended October 25 — up 103% year over year — and the number still doesn't tell you whether AI pays for itself. That's the uncomfortable part. Data center revenue hit $107.6 billion, up 110%. Gross margin: 74.8%. Operating income: $77.5 billion. Net income: $65.3 billion. Next quarter, NVIDIA guides to nearly $130 billion, with a 3% margin flex.

The sales are real. What the sales don't show is whether the companies buying those GPUs earn enough from AI to cover them.

What is the funding loop?

The mechanics are simple enough to sketch on a napkin. A chip supplier or large investor puts money into an AI company, a cloud provider, or a data center project. That company raises more capital, builds facilities, buys NVIDIA GPUs, and sells compute to AI labs and business customers. NVIDIA books the sale. The cloud provider, meanwhile, depends on what customers pay to cover hardware, power, leases and debt.

The cycle holds while customers pay for AI services they find useful. It gets risky when investor money replaces customer revenue. That, in a sentence, is the bubble argument.

CoreWeave is the cleanest example. It buys GPU systems and rents capacity to AI companies; NVIDIA has agreed to purchase up to $6.3 billion of CoreWeave's unused capacity through 2032. For CoreWeave, a guaranteed buyer for whatever other clients leave unclaimed. For NVIDIA, fortunes tied more tightly to a major customer's health. If demand fell, CoreWeave could sit on idle GPUs, falling rental prices and debt payments that are hard to meet.

How big is the spending wave?

Hyperscaler capital expenditure is the loop's fuel line:

  • 2025 actuals: Amazon ~$132B, Microsoft $118B, Alphabet $91.4B, Meta $72.2B — roughly $413.6 billion combined.
  • 2026 estimates: Amazon $220B, Microsoft $170B, Alphabet $200B, Meta $140B — about $730 billion in all.
  • First nine months of the current period: $508 billion in capital costs already.

These figures cover buildings, networks, power systems and other equipment, not only AI chips. Spending this much doesn't mean revenue or profit rise by the same amount. Ask anyone who financed fiber in 1999.

Where does the $500 billion fit?

NVIDIA has announced a plan with BlackRock, Apollo, Blackstone, Brookfield, Goldman Sachs and KKR to bring in more than $500 billion of third-party capital for AI infrastructure over time, with project-specific residual-value support of up to 25%. Read the fine print: that's a target, not money raised, and not NVIDIA sales. Financing like this can pay for data centers before they earn anything, and spread losses among lenders, investors, suppliers and cloud companies if projects fail to draw paying users.

The late-1990s telecom boom is the obvious cautionary tale. Telecoms laid fiber fast, equipment makers collected enormous orders, and some networks ended with far more capacity than customers could use. AI data centers could face the same mismatch if supply outpaces paid demand.

The differences matter, though. AI already has real users and clear uses in software, research and business. NVIDIA's software base in CUDA, its networking systems and complete servers give it a moat telecom equipment makers lacked. Its Vera Rubin NVL72 systems have started shipping to CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.

What should a data or finance team actually track?

The story says the signals, and they're the right ones:

  • GPU utilization and rental prices — falling prices plus weak utilization means too much supply.
  • Cloud revenue and debt costs — rising borrowing costs squeeze cloud operators.
  • Free cash flow and repeat customers — sticky AI sales point to demand that lasts.

NVIDIA could stay profitable even if parts of the AI industry borrow too much or overbuild data centers. The real risk isn't whether companies buy GPUs today. It's whether those GPUs earn enough to pay for themselves. That's what separates a lasting boom from an expensive pile of unused capacity.

Analytics Insight (Source)

Price tags

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering companies, deals and strategy in Analytics Digest is a professional daily at Analytics Digest.

18 articles

Read also

« Previous articleNext article »