It’s the best and worst time to be in semiconductors right now. Silicon Valley investors are once again owning up to their namesakes and taking a deep interest in next-generation silicon, with leading lights like Graphcore in the United Kingdom hitting unicorn status while weirdly named and stealthy startups like Groq in the Bay Area grow up.
Growth in chips capable of processing artificial intelligence workflows is expected to swell phenomenally over the coming years. As Asa Fitch at the Wall Street Journal noted yesterday, “Demand for chips specialized for AI is growing at such a pace the industry can barely keep up. Sales of such chips are expected to double this year to around $8 billion and reach more than $34 billion by 2023, according to Gartner projections.”
Yet, all those rosy projections don’t suddenly make the financial results of companies like Nvidia any easier to swallow. The company reported its quarterly earnings last week, and the results were weak — pretty much across the board.
We talked a lot about Nvidia’s stock meltdown last year on Extra Crunch. Crypto miners, who relied on Nvidia’s GPUs to mine cryptocurrencies like Bitcoin, bought huge numbers of chips and just as quickly disappeared into the night, taking Nvidia’s sales momentum and stock along with them. During 2018, Nvidia’s stock fell from about $200 to about $130, a decline of 35% according to Yahoo Finance.
Outside the crypto crash, Nvidia was also buffeted by solid competition from new entrants and large players like Apple and Google as well as geopolitical tensions due to China and the U.S., which we also covered last year on EC. However, last week’s quarterly numbers showed the company’s travails aren’t just being driven by crypto, but by lackluster sales in pretty much every single one of its business lines.
Nvidia’s overall strategy has been to expand into areas outside of its core graphics and gaming customers to more strategic segments like data centers, AI, automotive, and more. That transition has been inchoate, to say the least.
Take cloud computing, which is where most AI chips are likely destined for at least the short-to-medium term (longer term, expect them to show up in more of our devices and gaming consoles as prices and power consumption become reasonable). Nvidia has pushed heavily into data centers as a key pillar of the company’s economic future, but as it wrote in its 10-Q:
Data Center revenue was $634 million, down 10% from a year ago and down 7% sequentially, primarily reflecting a slowdown in purchases by certain hyperscale and enterprise customers, partially offset by growth in inference sales. We believe this slowdown in purchases will likely persist into the second quarter of fiscal year 2020.
(Side note: I love the use of the term hyperscale to describe a monopsony customer, presumably AWS.)
The likeliest explanation is that for all the excitement about AI and chips in the cloud, most companies just haven’t found the applications that require Nvidia’s chips, and therefore there is less demand for them than many expected. Despite Gartner’s estimates, a huge challenge with AI chips is that you still need software to send their workloads to these chips — and without better software that takes full advantage of this new power, the demand just isn’t there.
Hardware investors remain deeply excited about the potential for areas like AI chips, even as customer demand may be weaker today than predicted. (Photo by Kimberly White/Getty Images for TechCrunch)
The weak results weren’t just in the data center. Next, let’s move to Nvidia’s relatively nascent Tegra Processor business, which “includes automotive, SOC modules for gaming platforms, and embedded edge AI platforms.” Results here were similarly lackluster. Revenue “was $198 million, down 55% from a year ago and down 12% sequentially. The year-on-year decrease primarily reflects a decline in shipments of SOC [system on a chip] modules for gaming platforms.”
The decline of Nvidia’s GPU business due to crypto (“down 27% from a year earlier and up 2% sequentially”), and its weakness in both its core gaming market, as well as its expansion markets, shows that the race to AI chips and next-generation silicon is still in its earliest phases, and customer demand still needs to be encouraged before serious dollars pour into this category. Meanwhile, Nvidia’s gross margins declined 6.1% year over year due to price pressures and a change in the mix of its sales.
That drumbeat of negative news has hit Nvidia’s stock again, which had a run up this year to $192 on April 10th, but has since declined to $151 today, a drop of 20% or so in just about a month.
Few tech companies seem to sustain the kinds of wide price gyrations like Nvidia. Operating expenses were “up 21% from a year earlier and up 3% sequentially, reflecting primarily employee additions and increases in employee compensation.” If the stock is that variable, it should surprise no one that compensation has to, well, compensate.
Part of the reason for that huge variation is that AI chips handling AI workflows seems to be an agreed-upon inevitability, and the biggest challenge for investors is identifying exactly when the transition takes place. Nvidia is well positioned to capitalize on this excitement — whenever it shows up. The question is, how long can the public markets wait for this customer segment to truly materialize? It doesn’t have the flexibility of silicon startups, which can presumably exit whenever the excitement is there in the next couple of years.
And so we all wait for the glittering riches of silicon to come. Investors, founders, and the industry writ large can’t stop talking about our new AI (chip) future. Ultimately though, customers need to get on board, and timing that is any one’s guess.