Nvidia's data centre surge makes AI capacity a financing decision
Nvidia's sales surge showed extraordinary demand for computing infrastructure. For young companies buying or supplying capacity, returns depended on utilisation and contract terms.
Nvidia's results on 23 August 2023 put a financial scale on the rush to build AI infrastructure. Data Center revenue reached $10.32 billion in the quarter ended 30 July, up 141 percent from the preceding quarter and 171 percent from a year earlier. Total company revenue was $13.507 billion. These were sales recognised by an infrastructure supplier, rather than a measure of the revenue generated by applications running on its products.
Venture investors had to examine who was carrying the cost of that expansion. A supplier could sell systems while a customer was still developing the service intended to pay for them. A young cloud operator could secure equipment before securing steady demand. An application company could reserve computing capacity while its customer contracts remained uncertain. Rapid infrastructure growth increased the stakes for each business.
A quarter that outgrew the previous forecast
Nvidia had reported Data Center revenue of $4.28 billion for the preceding quarter, ended 30 April. It forecast total revenue of $11 billion, plus or minus 2 percent, for the following quarter. The $13.507 billion result exceeded that range. This was a substantial increase against the company's own guidance, rather than a comparison with analyst consensus.
| Period or category | US$ billion |
|---|---|
| Q1 FY2024: ended 30 April 2023 | |
| Q2 FY2024: ended 30 July 2023 |
Reported quarterly revenue for Nvidia's Data Center market platform, rounded as in the earnings releases. Both observations cover one fiscal quarter and measure revenue for this platform, rather than GPU shipments or total company sales.
The two quarterly Data Center revenue figures show how quickly sales grew. Revenue reflects the products and systems sold, their prices and the timing of recognition; it does not reveal GPU shipment counts, installed computing capacity or the number of paying AI users. A startup planning deployment could take the growth as evidence of infrastructure demand while separately verifying delivery dates and usable access.
Supplier profits and customer economics
Nvidia's GAAP gross margin across the company rose from 64.6 percent in the first quarter to 70.1 percent in the second. The latter was not a disclosed standalone Data Center margin. It demonstrated substantial gross profit at the supplier, but did not reveal the economics of a cloud operator renting out the resulting infrastructure or an application business consuming it.
Those businesses faced different costs. A capacity provider had to recover equipment, hosting, power, operations and financing expenditure through customer payments. An application company had to cover computing alongside product development, support and selling costs. Applying Nvidia's margin to either business would skip those obligations. For investors, the question was whether a company could improve throughput, reliability or distribution enough to retain value after paying the infrastructure bill.
Making rented capacity productive
Nvidia's DGX Cloud launch offered a model based on monthly cluster rental, with instances starting at $36,999 per instance per month. Each instance featured eight H100 or A100 80GB GPUs, providing 640GB of GPU memory. The service bundled software, networking, storage and access to Nvidia expertise. The starting price covered that package. It was distinct from the purchase price of a GPU or an hourly rate for computing from another provider.
For a young company, renting could move expenditure from acquiring and operating equipment into a recurring service commitment. That could suit a bounded training project or an uncertain development schedule. It did not ensure that every month produced equally valuable work. Data preparation, debugging and idle intervals still affected how much useful output a paid instance delivered. Comparing providers required a defined workload and comparable service terms, rather than dividing a headline price by the number of accelerators.
Utilisation connected the engineering plan to the cash plan
A practical budget could begin with a specified result: a completed training run at an agreed quality, or an inference service responding within the time a customer required. Engineers could then record the paid capacity, retries, storage and operating effort needed to deliver it. Fast hardware still needed to produce an economical service. A failed run consumed money even when the successful run looked efficient in a demonstration.
US$ million · net accounts receivable
View data
| Date | US$ million · net accounts receivable |
|---|---|
| 29 January 2023 | 3,827 |
| 30 April 2023 | 4,080 |
| 30 July 2023 | 7,066 |
Net accounts receivable across NVIDIA as a whole, reported at three balance sheet dates. These balances record outstanding assets rather than quarterly sales or cash collected. They do not show whether invoices were overdue or collections had slowed, nor do they isolate the Data Center business.
For an infrastructure startup, the corresponding question was whether customer commitments matched its own capacity obligations. A long equipment or hosting commitment financed by short, cancellable usage left the operator carrying the gap. Reserved customer capacity reduced one uncertainty but could introduce others, including concentration in a small number of buyers. Investors needed to examine those contracts with the company they were funding; Nvidia's results could not establish the economics of its customers.
Growth left room for products that made capacity useful
Nvidia's earnings release highlighted networking, server designs, software and cloud availability alongside processors. That breadth reflected the work required to turn components into functioning infrastructure. A startup improving scheduling, observability or deployment could address a valuable problem if it demonstrated better results on the customer's systems. It also needed to assess which functions the platform supplier might provide itself.
Nvidia forecast $16 billion of company revenue, plus or minus 2 percent, for the next quarter. That outlook remained unproven when announced. A young company's financing depended on its own measurable contribution: more useful work from committed capacity, a dependable service customers would pay for, or lower deployment effort. The supplier's extraordinary quarter established the scale of infrastructure demand. Each downstream business still had to establish how it would earn its return.
Investment implications
- Data Center revenue measured Nvidia's sales into that market, not customers' utilisation or application profitability.
- The relevant computing cost depends on completed, useful work and the capacity commitments needed to deliver it.
Sources
- NVIDIA Announces Financial Results for Second Quarter Fiscal 2024 (opens in a new tab)
NVIDIA
- NVIDIA Announces Financial Results for First Quarter Fiscal 2024 (opens in a new tab)
NVIDIA
- NVIDIA Launches DGX Cloud, Giving Every Enterprise Instant Access to AI Supercomputer From a Browser (opens in a new tab)
NVIDIA
Published
