Supermicro bets on next-gen chips to carry it through economic downturn

Customers just love taking risks on pricey new tech, right?

Analysis If one thing was abundantly clear from Supermicro's Q1 earnings call this month, it's that the server maker really needs Intel, AMD, and Nvidia's next-gen datacenter components to be a hit if it's going to weather the economic downturn.

On the surface, the first quarter of Supermicro's 2023 fiscal year was business as usual. The company posted revenues of $1.85 billion, up 79 percent year-over-year, on net income of $184 million. This was helped along by the fact Supermicro didn't have to worry about offsetting plummeting PC demand.

But the company's flat outlook on Q2 growth – in the range of $1.7 billion to $1.8 billion – showed the company isn't ruling out headwinds from a recession that's already battering cloud and hyperscale customers.

Such a large refresh could prove to be more of a curse than a blessing for risk-averse enterprise datacenters used to seeing new features and standards roll out more gradually

Pressed by analysts during Tuesday's call, Supermicro's CEO and CFO hammered their points on Intel, AMD, and Nvidia's next-gen CPUs and GPUs as an insulating force that should shield the company from an economic slowdown and even present opportunities to carve out a larger share of the market.

"The macroeconomic condition is hard to predict at this moment – it looks like there will be some headwind for sure," CEO Charles Liang said in response to questions. "I believe our business in next 12 months should not [experience] too much impact."

A big refresh

Liang isn't wrong. It looks like 2023 is shaping up to be a big year for IT infrastructure – no fewer than four new datacenter CPUs will hit the market and at least as many GPUs are expected to enter volume production.

"We are ready to launch a full line of next-generation products with technologies across our key partner ecosystem," Liang said.

And these won't just be the AMD Epycs and Intel Xeons we've come to expect. Nvidia's Grace CPUs are slated to begin making their way to customers in Q1, while Ampere's Arm-based third-gen CPU should – barring any delays – begin ramping production some time next year. Supermicro currently offers systems from all four vendors.

Beyond offering an even larger selection of CPUs this time around, these chips introduce a bevy of new standards to the datacenter market, the likes of which we haven't seen since the switch to DDR4 circa 2014.

These chips will not just add support for faster and far more expensive DDR5 memory, they'll also double the I/O bandwidth with PCIe 5.0 and introduce the first iteration of the Compute Express Link (CXL) 1.1 standard. The latter should be a welcome relief for anyone disappointed by Intel's decision to can Optane earlier this year.

At the same time, customers interested in adding AI/ML acceleration to the mix will have just as many if not more options. In addition to Nvidia's and AMD's regular fare of of datacenter GPUs, Intel has entered the race with its Ponte Vecchio and Datacenter Flex platforms.

New = an expensive risk

Choice is rarely a bad thing, and the new chips will certainly garner some interest. However, such a large refresh could prove to be more of a curse than a blessing for risk-averse enterprise datacenters used to seeing new features and standards roll out more gradually. This time around, buying a new system could mean taking them all on at once.

Worse, those new features won't come cheap. As of writing, DDR5 memory is roughly twice the price of DDR4, and that's before you factor in what will almost certainly be higher prices for CPUs and GPUs this generation.

Supermicro certainly expects to sell them for a premium above current generation products. You "may have heard that a lot of GPU prices and CPU prices are going up, especially with the new the new refreshes that are coming out. So we anticipate that [average selling prices] will continue to go up," CFO David Weigand told analysts on Tuesday's call.

Depending on the margins – what Supermicro can buy components for, versus the price customers will pay for complete systems — the company may not need to sell anywhere near as many units to achieve its revenue targets.

But as we've seen with Nvidia's RTX and AMD's Ryzen refreshes this fall, new tech doesn't always mean greater demand when it's tied to higher prices – or butts up against a weak economy.

While there remains a healthy backlog of demand for datacenter infrastructure, analysts expect that to begin dissolving toward the middle of next year as supply chains improve. And if economic conditions continue to deteriorate, companies like Supermicro may discover enterprises don't have quite the appetite for expensive new datacenter kit they were counting on.

Supermicro's hyperscale problem

However, according to Supermicro's latest earnings presentation, enterprise customers may not be its biggest priority. Over the past year, the company's mix of revenues has trended heavily toward large datacenter customers, with traditional enterprises falling from 70 percent of revenues to 45 percent.

If enterprises cut back on IT spending in the face of worsening economic conditions – something we're told is is typical of past recessions – Supermicro could find this mix shifting even further toward the hyperscale and cloud segment.

Compared to the typical enterprise customer, hyperscalers and cloud providers don't tend to be as risk averse or price sensitive, adopting whatever is available when they need it. Several cloud providers have already committed to offering instances based on chipmakers' next-gen CPU and GPU platforms.

But hyperscale demand tends to come in waves of massive capex followed by lulls of integration of optimization. And after a period of explosive growth triggered by the COVID-19 pandemic, analysts expect the next cloud and hyperscale digestion phase may not be far off – especially as uncertain economic times drive customers to investigate cost-saving alternatives. ®