Nvidia is now worth its high price

Sitting at the intersection of multiple breakthrough technologies means Nvidia’s share price has been particularly volatile in the last two years. In 2021, low interest rates coupled with the Covid-19 lockdown tailwind to its gaming business led investors to bid up the stock by 130 per cent, to a peak valuation of 71 times forward earnings.

In 2022, it has been a different story. Rising interest rates, tough comparators for the gaming industry and geopolitical uncertainty have lowered the share price by 47 per cent. Nvidia now trades on a saner looking price/earnings ratio of 40.

With inflation still at multi-decade highs and discount rates climbing, that still qualifies as very expensive. But like a lot of technology stocks, it is substantially cheaper than this time last year. Given Nvidia’s explosive growth potential in the years ahead, it is also an entry point worth considering.

Data drives demand

Before the turn of the millennium, computers ran on the central processing units (CPU) that were first made commercially available by Intel in 1971. Although CPUs’ power improved rapidly in their first few decades after their adoption, by the end of the century this progress had started to slow. Then, in 1999, Nvidia invented the GPU.

GPUs allow faster computing because they use matrix calculations rather than linear calculations. This is known as parallel computing. To process this much information the GPUs need hundreds of 'cores' as opposed to the handful that sit on CPUs. CPUs are still used in computers for simpler tasks but nowadays all the really heavy tasks such as graphics generation, machine learning and autonomous driving are done by GPUs.

Nvidia started out designing GPUs for PCs to improve gaming graphics but then expanded into data centres, professional visualisation, and the automotive industry. The company is fabless, which means it outsources the production, mostly to semiconductor giant TSMC (TW:2330) in Taiwan. While this model leaves the company dependant on third parties, it also makes Nvidia a capital light, high margin and supremely cash generative business. In 2021, the operating margin stood at 40 per cent and free cash flow hit $8.13bn (£6.8bn) from just $10.7bn of operating profit.

In the third quarter of this year, data centres made up 65 per cent of group sales. Gaming comprised a further 26 per cent, followed by automotive and professional visualisation with around 4 per cent and 3 per cent, respectively.

Despite this, data centres have only just become the dominant division in the business. In the third quarter, data centre revenue increased 31 per cent from a year ago, to $3.83bn, thanks in part to Amazon Web Services’ growing use of the NVIDIA A100 Tensor Core processor in its servers. Nvidia also announced new two-year partnerships with Oracle (US:ORCL) and Microsoft (US:MSFT) during the quarter, the latter covering a contract to “build an advanced cloud-based AI supercomputer to help enterprises train, deploy and scale AI”.

Almost all enterprise software is now being run over cloud servers rather than on-premises. The cost of running these servers is so substantial it makes little economic sense for any mid-sized business to not outsource. In the most recent quarter Amazon, Alphabet (US:GOOGL) and Microsoft all grew their cloud businesses by over 30 per cent on a constant currency basis compared with last year. As demand for cloud services grows, so will Nvidia’s revenue.

The benefit of investing in Nvidia as opposed to the cloud computing companies themselves is that Nvidia has little competition in the GPU market. Intel (US:INTC) and AMD make CPUs for the cloud servers, but neither can yet challenge Nvidia in designing GPUs used for machine learning and AI. Only last year did Intel launch its first GPU for data centres.

An AI inflexion point

Until now, AI has almost exclusively been used by enterprises for tasks, such as improving cyber security or customer data optimisation. However, pathways to more AI consumer applications have appeared in the from of text-to-image generators like DALL-E 2 and language model GPT-3. DALL-E 2 can turn text prompts into eerily accurate images. It does this by scraping all images on the internet, a task which requires so much computing power that it needs to be run on the cloud rather than a laptop. Last month, Microsoft’s cloud computing division, Azure, said its clients would be able to run DALL-E 2 through its Open AI service.

Although Microsoft only provides these services to enterprise clients, Nvidia says AI consumer applications – such as “large language models, recommendation systems and generative AI” – are already driving growth alongside the major cloud players.

This is an encouraging development. While new technology is often first adopted by enterprises, demand often kicks off when consumer applications appear. It happened with the personal computer and now it is starting to happen with machine learning. This seems to have been a turning point.

A tough backdrop

While the data centre business goes from strength to strength, gaming has been a weakness. Gaming sales in the three months to September were down 51 per cent year on year. This was because Nvidia’s partners needed to “align channel inventory levels with current demand expectations as macroeconomic conditions begin to weigh on consumer demand”. In other words, when the economy is doing badly people spend less money on expensive gaming laptops.

Gaming GPUs are also used for crypto currency mining which drove sales this time last year but is now dropping off. That has meant a 31 per cent fall in gaming revenue on a two-year basis, during which time data centre sales have doubled.

The consumer-facing arm was always more likely to be affected by the downturn, especially after the lockdown induced boom. But this doesn’t mean it won’t bounce back, as record opening weekend sales for the latest Call of Duty title suggest.

Other trends may be harder to unwind. Nvidia’s strong third-quarter data centre numbers landed despite the recently passed US Chips and Science Act, which restricts certain chips being exported to China that can be used for super computing and AI. For Nvidia, this means no Shenzhen-bound sales of its A100 and H100-based products.

It’s unclear exactly how much this will impact the business. At its third-quarter results, Nvidia said “sequential growth was impacted by softness in China”, but didn’t specifically link this to the legislation. However, any form of further escalation between the US and China over Taiwan would be worrying for investors. Last year, 26 per cent of revenue came from China, 16 per cent was from the US and 32 per cent from Taiwan.

To get around the sanctions, Nvidia has already designed a less powerful GPU that can be sold to Chinese data centre businesses but cannot be used in supercomputing. However, it is always possible the US could change the law again to prevent any GPUs designed in the US being shipped.

Knowing how much of a discount to apply to this geopolitical uncertainty is tricky. But it is certainly contributing to a multi-year discount in the market valuation for a business with near endless potential.