NVIDIA data centres drive sales surge and US$16 billion forecast

NVIDIA reported US$13.51 billion in quarterly revenue on 23 August and forecast another increase to US$16 billion, as spending on AI infrastructure changed the shape of its business. Its data-centre business supplied almost all the growth since the previous quarter. The accompanying cash-flow disclosures show why sales, profit and cash still need to be read separately.

File photo: NVIDIA Sign – Santa Clara – California, © 2020 Will Buckner, photographed 31 October 2020. CC BY 2.0, via Wikimedia Commons. Resized and cropped.

The results cover fiscal 2024’s second quarter, which ended on 30 July 2023. Revenue rose 88% from the previous quarter and 101% from a year earlier, exceeding the US$11 billion midpoint of the forecast NVIDIA gave in May by about 23%.

Where the increase came from

Data Center revenue reached US$10.323 billion, compared with US$4.284 billion in the preceding quarter and US$3.806 billion a year earlier. Calculated from NVIDIA’s figures, its US$6.039 billion quarterly increase accounts for roughly 96% of the company-wide US$6.315 billion increase. It now represents about 76% of total revenue.

That category covers more than AI processors. In her written commentary, chief financial officer Colette Kress attributed demand for Hopper- and Ampere-based HGX systems mainly to large language models and generative AI. Data-centre compute revenue increased 195% year on year, while networking grew 94%, primarily from InfiniBand infrastructure supporting those systems. The figures describe a wider infrastructure buildout, without isolating revenue from generative AI alone.

Gaming also recovered, reaching US$2.486 billion, up 11% sequentially and 22% annually. Automotive revenue moved the other way from the previous quarter, falling 15% to US$253 million, although it remained 15% above the year-earlier level.

Profit and cash have different comparisons

GAAP net income was US$6.188 billion, against US$2.043 billion in the preceding quarter and US$656 million a year earlier. Diluted earnings were US$2.48 a share. NVIDIA’s adjusted, non-GAAP figure was US$2.70, excluding items including stock-based compensation and acquisition-related costs.

Gross margin rose to 70.1% from 64.6% sequentially and 43.5% a year earlier. NVIDIA attributed the improvement principally to the larger data-centre business. The annual comparison also includes a depressed starting point: the year-earlier quarter carried US$1.34 billion in inventory provisions and related charges.

Operating cash flow reached US$6.348 billion, up from US$2.911 billion sequentially and US$1.27 billion a year earlier. Customers paid approximately US$1.25 billion ahead of invoice due dates, benefiting accounts receivable in the quarter. NVIDIA also expects to pay about US$3.81 billion in previously deferred cash taxes during the third quarter, following California disaster relief. These disclosures limit what one quarter’s cash generation says about a recurring run rate. The early payments concern collection timing; the tax bill is an expected outflow next quarter.

More supply, with an unquantified ramp

The next-quarter revenue forecast spans US$15.68 billion to US$16.32 billion. Its midpoint would mean another increase of about 18.5%. NVIDIA expects a GAAP gross margin of 71.5%, plus or minus half a percentage point.

On the 23 August earnings call, Kress said data centres would drive most sequential growth, with gaming and professional visualisation contributing. She expected supply to increase each quarter through next year but didn’t give a percentage when asked. The forecast therefore comes with a production expansion still under way.

Kress also put China demand within its historical range of 20–25% of data-centre revenue, including networking. She said worldwide demand meant additional GPU export restrictions, if adopted, weren’t expected to have an immediate material financial effect. Her longer-term warning was different: a prohibition on sales to China would permanently reduce opportunities for US industry in the Chinese market. Both assessments were management’s expectations about potential restrictions, rather than evidence that export exposure had disappeared.