NVIDIA's May 2022 quarter is August 2018 again: inventory called normal, guidance light. Last time, data center took three more quarters to turn negative.
We caught the technology megatrend in 2016 and then got knocked over by the electronics inventory correction in 2018. By the middle of 2018, Taiwan's supply chain was already hearing about order cuts, while the US companies riding the same trend were still pointing at the long term and guiding optimistically. That is when we learned that even inside a long megatrend, inventory swings create a cycle, and that cycle carries real risk.
Key takeaways
- NVIDIA's quarter reported in May 2022 matches August 2018 on four counts: inventory declared back to normal for the first time, no new gaming card shipped, inventory days climbing back toward the 2018 peak with inventory growing faster than revenue, and the first guidance miss in many quarters while management stayed positive on demand.
- In 2018 and 2019 the damage came out one quarter at a time. Gaming cracked in November 2018 and the stock fell 18% after hours. Data center still grew 58% year over year in that same quarter, and only turned negative, down 10% year over year, in May 2019. Consumer first, enterprise last.
- A 50% drawdown from the high is not the same as the correction being priced. At 33 times the next four quarters of earnings, NVIDIA is back at the average multiple that prevailed before the 2018 correction, not below it. A real correction took the multiple to 20 times by the end of 2018, and EPS estimates would still have to come down.
- The group tied to cloud, high performance computing (HPC), advanced process nodes and 3D packaging stays a core long-term holding. Buyers who scale in before the turn is confirmed can start now; in 2018 and 2019, anyone who waited for confirmation did not get it until the third quarter of 2019, when inventory days at Taiwan electronics companies finally fell below their year-earlier level.
NVIDIA's latest quarter, reported in May 2022, brought all of that back. So we went into our old notes to lay out how a company's language and its reported numbers change when a long-term trend runs into a short-term correction.
One thing to say up front: the group that ties together cloud, HPC, advanced process nodes and 3D packaging is still a core long-term holding for us. Inside a long technology revolution, an inventory correction is another chance to add at a better price.
This post covers NVIDIA in 2018 and 2019. Part 2 covers what happened to capital spending (capex) at the large US cloud companies (the hyperscalers) over the same stretch.
Familiar language, familiar numbers
Going back through what the company said in 2018 and 2019, the pattern is this: once inventory builds to an extreme level and consumer demand rolls over, an inventory correction can start. The opening shot is two statements in the same quarter: guidance below consensus, and inventory declared back to normal. That is what happened in August 2018, and several quarters of correction followed. What NVIDIA said this time sounds a lot like it.
So what does the quarter reported in May 2022, covering the first quarter of calendar 2022, have in common with the quarter reported in August 2018, covering the second quarter of calendar 2018?
- After a long stretch of shortages, price increases and customers ordering ahead, the company says for the first time that inventory is back to normal levels.
- No new gaming card had shipped yet, which says gaming demand was not urgent. In 2018 the market originally expected a mid-year launch; after the correction, it did not arrive until January 2019.
- Inventory days keep climbing, past 12 weeks, and inventory is growing as fast as or faster than revenue. At the same time, revenue growth estimates are being cut and the absolute revenue number fell from the prior quarter.
- For the first time in many quarters, guidance for the next quarter came in below consensus, while the company stayed positive on demand.
The correction starts in consumer demand and ends up in enterprise demand
We pulled the NVIDIA memos we kept quarter by quarter back then and summarized the earnings calls below. A lot of the language and a lot of the numbers track what we are seeing now, which is why we are laying them out here.
August 2018 (second quarter of 2018): inventory up, revenue flat, management still optimistic
- The stock sold off after hours because guidance for the next quarter came in below consensus, up only slightly from the prior quarter and 2.69% below what the Street expected. The stock later recovered and made a new high.
- Gaming grew 52% year over year. No new gaming card had shipped yet, but the company said gaming demand was still strong.
- Data center grew 83%, a record 24% of revenue, and AI stayed at the center of company strategy.
- The problem was right there in the inventory line: inventory grew 30% quarter over quarter, and inventory days hit the top of its range at 79 days.
The clearest signal at the time was that jump in inventory. Revenue was flat with the prior quarter while inventory grew 30%, and analysts asked about it on the call.
Management's answer was that the inventory was being built for future sales, specifically the back-to-school third quarter, and that with the Turing, Volta V100 and Pascal architectures all in the lineup, channel inventory and its own inventory were in balance.
November 2018 (third quarter of 2018): gaming inventory correction acknowledged, quarter and guidance both below consensus
- The third quarter of 2018 was the first revenue miss in 13 quarters. Guidance for the next quarter was also below consensus, the stock fell 18% after hours, and the inventory correction was formally under way.
- Gaming revenue dropped hard. The company held back shipments to normalize channel inventory, guided revenue lower again for the next quarter, and expected channel inventory back to normal by the end of the fourth quarter.
- Data center grew a strong 58% year over year, with the company saying it saw very strong customer demand for its products. Analysts raised concerns about a slowdown in cloud spending; the company said data center would keep performing well.
One analyst asked the company directly: why was inventory not a problem for the past several quarters, and now the revision is this large?
Management said they were surprised too, that the factors were complicated and much worse than expected, and that there were 12 weeks of channel inventory to work through in total. Mining demand had disappeared and that pushed prices down, but once prices fell, unit sales should recover. Gaming inventory was a short-term issue and would be cleaned up within a quarter or two.
February 2019 (fourth quarter of 2018): data center comes in below plan for the first time
- On January 28, 2019 the company cut its fourth quarter guidance. The reported quarter and the guidance both matched the lowered consensus at down 24% year over year, driven mainly by a 45% decline in gaming. Gross margin and operating margin fell sharply, and price cuts were needed to move product.
- Gaming channel inventory was expected to normalize in the first quarter, but a deteriorating macro backdrop, China above all, hurt consumer demand for GPUs, and high-end GPUs on the new Turing architecture sold below plan. Even with gaming that far off, the company still expected a recovery in 2019.
- Data center came in below expectations. Customers turned cautious on investment because of economic uncertainty, the decline in units was broad across vertical end markets and regions, and some large customers stopped buying in the closing weeks of the year. The company called the capex pause temporary, expected a meaningful recovery only later in the year, and argued that over the long run NVIDIA's accelerated platform still had the advantage.
Looking back on it, the company's read was that the gaming inventory problem came from shipping above end demand. To clear the channel, it kept adjusting shipments through the fourth quarter of 2018 and the first quarter of 2019, with inventory expected back to normal by the end of the first quarter of 2019.
As for data center, this was the first miss since the company began breaking out that revenue line in 2016, and management admitted the slowdown in demand was something it had never experienced before.
May 2019 (first quarter of 2019): data center growth turns negative for the first time since the line was broken out
- Full-year guidance was withdrawn. With hyperscale data center capital spending on pause, the company said it was still working through uncertainty that would carry into the second quarter, but believed second half revenue would be better than the first half.
- Gaming fell 39% year over year, with the adjustment running into the second quarter of 2019.
- Data center fell 10% year over year, reflecting the pause in hyperscale cloud spending. Demand at some hyperscale customers bounced back while others paused or cut. Next quarter would improve but still come in below what the company had expected earlier in the quarter, when it was looking for 2019 revenue flat to slightly down against 2018. Now the global pause in data center spending looked likely to run into the second quarter.
NVIDIA's bad news came out one quarter at a time: inventory back to normal, then consumer demand hit, then the data center business everyone assumed was strong feeling the slowdown too. Over the same stretch, the hyperscalers were far more back and forth about their own capex, and we think the swing factor for cloud capex is corporate profits. That is the subject of Part 2.
What is different this time: NVIDIA is already down 50% from the high
Put our May 2022 sell signal on Taiwan's supply chain next to what NVIDIA said in 2018, and the conclusion is that the cuts to fundamentals are only getting started. This time it is not only corporate language and softer consumer demand: the macro data looks like mid-2018 too.
- Manufacturing PMI and durable goods orders growth are rolling over from the highs.
- Taiwan's official leading indicator index is falling clearly.
- Company outlooks are splitting apart, depending on whether the business sells to consumers or to enterprises.
- Inventory balances are up a lot, and every company has its own explanation: stocking ahead of raw material price increases, building for the second half peak season, product transitions, and so on.
The difference is the price. Before the 2018 correction, the stock was still sitting near its highs, and it only broke hard in the month the company admitted the correction. This year NVIDIA is already 50% off the high, so the obvious question is whether that is priced in.
Rather than the size of the drawdown, look at the P/E as the measure of cheap or expensive. At 33 times the next four quarters of earnings, NVIDIA is not obviously undervalued. It is simply back at the average multiple that prevailed before the 2018 inventory correction.
Our own view is that the drop in the first half of 2022 is a P/E correction against the flood of central bank liquidity after 2020, and that the multiple has only come back to a reasonable level. If a real inventory correction runs, it would not be strange to see the multiple compress further toward the 20 times it reached at the end of 2018. And with consumer product sales cooling, the EPS line has room to come down as well.
In the 2018 electronics inventory correction, the low in the market showed up once the correction reached the best companies. Apple cut its guidance on January 3, 2019, and the S&P 500 did not make a new low after that. If you are a left-side investor, meaning you buy before the turn is confirmed, dollar cost averaging into these megatrend names from here should produce a good long-term return. Just keep a safe cash position, buy slowly, and size up as the correction signals get clearer.
Right-side traders, who wait until the turn is confirmed and care about capital efficiency, had to wait until the third quarter of 2019 in that cycle, when inventory days at Taiwan electronics companies finally fell below their year-earlier level and gave a buy signal. You give up the exact bottom, but you get the most efficient stretch of the trend.