Originally written Nov 3rd, posted Nov 11th
Ultimately for AMD, the year is going to be a tale of two halves. The promise of a strong Q3 and Q4 has been sunk by strong macroeconomics and a strong drop in the consumer PC market. While the Data Center market and embedded segments remain strong, it’s the consumer that’s dropped from a $2B quarter to a $1B quarter, or less in future.
So what looked like to be on course for a 360 million unit year in consumers is currently supposedly at a run rate of less than 50m/quarter (or 200m/year), and with new products on the horizon, companies like AMD are actually under-shipping current generation hardware in order to not saturate the market. The choice would be to undership or to drop prices, and AMD would rather do the former and keep margins high, whereas other companies might do it differently.
As a result, we saw AMD showcase a $5.56 billion revenue quarter overall, down from those original estimates from Q2 of $6.7 billion. This featured a one-off inventory correction of $160m, and investments in future supply chain resulting in a very different GAAP (regulatory filing) and non-GAAP (underlying base health of the books) results.
Let’s start with the numbers in GAAP vs Non-GAAP with YoY differences.
1. Revenue $5.56B (up 29% YoY)
2. Gross Margin GAAP 42% (down 610bps) vs Non-GAAP 50% (up 150bps)
3. Op Ex GAAP $2.43B (up 113%) vs Non-GAAP $1.52B (up 47%)
4. Operating Income GAAP -$64m (down 107%) vs Non-GAAP $1.26B (up 20%)
5. Net Income GAAP $66m (down 93%) vs Non-GAAP $1.1B (up 23%)
Those large OpEx and OpIncome differences are hard not to notice.
To showcase the underlying business, here’s the Non-GAAP breakdown of each business unit.
Data Center
a. Revenue $1.6B (up 45% YoY)
b. Operating Margin 31% (up 300bps YoY)
c. Operating Income $505m (up 64% YoY)
This is the 10th straight quarter gain in server CPU sales, with the ramp of 3rd gen EPYC (Milan) now in full swing. Cloud revenue doubled year over year, and the US Cloud market is actually a lot more resilient than most of the other business customers. That being said, while some US Cloud are reducing expected investments in the short term, AMD has sight into a strong growth for this market in the mid-term and is ramping towards that. To pull out a point, AMD stated 70+ new cloud instances built on EPYC were offered in Q3.
For other markets, AMD is seeing China remain weak, however AMD hasn’t penetrated the Chinese DC market as hard as other regions (China is mostly consumer for AMD still). OEM sales in the DC business unit are down, but that’s due to match sets – customers can’t get all the parts they need to finish individual systems, so there are still weaknesses in partner supply chains there. OEMs are also seeing slowing sales in enterprise but that’s part of the macroeconomic conditions as well.
AMD is seeing record sales of CPUs with FPGAs in markets such as Cloud and the financial customers, and combination selling with Pensando solutions are also driven by cloud customers. AMD will be announcing 4th Gen EPYC (Genoa) here on 10th November, with customers ready to go on day one. 2023 will see a ramp of Genoa, with Milan+Genoa being a combined market offering through the full year. In 2023, AMD will also launch Bergamo and Siena. Overall talk is looking at a 20-30% revenue growth for server in 2023, built from increased market share and increased ASPs based on new CPU designs. AMD states that they’re not a cost play, they’re a feature play in cloud and enterprise.
On the GPU DC market, this is down year-on-year primarily due to a big HPC win in last year’s Q3. AMD has already started to talk about chiplet based MI300 GPUs for that market, with more information next year.
Client
a. Revenue $1.0B (down 40% YoY)
b. Operating Margin -3% (down 3200 bps)
c. Operating Income -$26m (down 105%)
The PC Market is going through a rough patch. AMD is reducing downstream inventory by underselling into the market, given that the expected market units for 2022 are almost 80-100m units lower than expected. Both units are down and ASPs are down, although the launch of the new Ryzen 7000 CPUs has seen channel desktop sell-through grow. AMD is expecting some ramping in Q4 with new motherboard designs coming online, however.
AMD is expecting Q4 to be seasonally soft for the consumer market, even in the current climate, and AMD will continue to undersell. If we look at the expected guidance for Q4 and FY (numbers below), with an overall QoQ revenue decrease and an expected QoQ server increase, then consumer CPU is likely to dip below the $1B market. So despite a ramp of new products, AMD would be willing to lose revenue and lose market share in this segment rather than drive ASPs lower with continued volume in the channel.
On the notebook side here, AMD has been focusing on commercial designs. We expect to see the announcement of upcoming mobile processors based on the Zen 4 design in January, perhaps for a late Q1/Q2 launch. If the new mobile CPUs can be as efficient as the desktop processors, it should be a good product offering, though as always the strong Intel incumbent can be hard to fight there.
Gaming
a. Revenue $1.6B (up 14% YoY)
b. Operating Margin 9% (down 700bps)
c. Operating Income $142m (down 39%)
Q3 shows six straight quarters of semi-custom sales growth driven by the high console demand still in play. It has been reported that in Q3, Sony built 6.6m Playstation machines, but sold around half. Given that Q4 is traditionally seasonally weak for consoles and the semi-custom business, there’s some reluctance to say if the growth will extend another quarter.
The graphics business declined during Q3, firstly due to the macroeconomic headwinds, but also a focus for AMD to reduce downstream inventory based on the upcoming launch of new products in this market. On November 3rd we will see the launch of AMD’s new RDNA3 graphics designs, marketed as the RX 7000 series. This is a new chiplet design using 5nm silicon, so will be interesting to see what volume AMD will have, and where the products perform. No date yet on retail or reviews.
Embedded (YoY comparisons impossible based on Xilinx Acquisition)
a. Revenue $1.3B
b. Operating Margin 49%
c. Operating Income $623m
Easily the best performing segment of AMD this quarter. Embedded includes the traditional AMD embedded business (30% margin, $23m op income) and the new Xilinx offerings. There is significantly strong demand and growth from the aerospace, automotive, and telecommunications markets based on the differentiated capabilities. Lots of new North America 5G and wired deployments using AMD’s solutions.
Outlook
In Q2, based on $12.4B revenue year-to-date, AMD’s FY projections were for $26.3B in revenue at the time. Those numbers have decreased sharply, with AMD missing the $6.7B revenue target in Q3 by $1.15B. It means that the full year revenue target is now $23.5B, almost $3B lower. It means that this graph earlier looks quite bad.
However, it might be worth looking at the situation from a different angle. Consider all of these numbers but without the Q1 and Q2 data. It shows the following:
Q3 2020: $2.80 B
Q4 2020: $3.24 B (+ 0.44 B)
Q1 2021: $3.44 B (+ 0.20 B)
Q2 2021: $3.85 B (+ 0.41 B)
Q3 2021: $4.31 B (+ 0.46 B)
Q4 2021: $4.83 B (+ 0.42 B)
Q3 2022: $5.57 B (+ 0.74 B or + 0.24/q)
Q4 2022F: $5.50B (- 0.07 B)
With this perspective, the Q1 and Q2 numbers were big outliers enabled by pandemic ramps and expectations. The gains from Q4 2021 to Q3 2022 are modest at an average +0.24B per quarter, compared to the +0.40 growth over previous quarters, but by this metric AMD is still sizing up its ramp. Q4 is seasonally weak in a number of key segments, especially consumer as mentioned, and although while cloud is still ramping EPYC 3rd Gen and 4th Gen is coming online, the 4th Gen revenue will still take a quarter or two to kick in as they’ll both be sold in the market.
AMD is set to exit the year with 51% Non-GAAP GMs and $23.5B in revenue. If you had asked if Lisa and the team would want that back in early 2020, they’d have broken your arm off. They’re cutting back heavily by undersuppling in the market ahead of a slow consumer market and RDNA3/Zen4 products ramping. The weak point in Q4 is going to be that client market, which will be shipping less than $1.0B in revenue as all the macroeconomic factors come into play.
With the Q&A from the analysts, one of the best questions was about the volume of systems sold in 2022 in the market. At the start of the year, all the big players were predicting a ‘run rate’ of around 360m units over the year, or effectively one million a day. The current downturn in the macroeconomic environment for consumer and commercial systems is pivoting down to 250m units a year, but that actually belies the issue that the last quarter was only 50m units, or around 200m/year. UBS asked about what the rate is going to be in Q4 given the economic situation, and what AMD is doing about it – stating that August and September must have been well below expectations. AMD responded saying that it drained a lot of consumer inventory in Q3 (eg undersupply the market so OEMs stock of parts reduces), and will continue into Q4, and AMD are planning for a weaker market.
On the topic of substrate supply, AMD has said continually over the last few quarters that it has been investing in its substrate supply – effectively making efforts to ensure that it has more and that its manufacturing partners can scale. The substrate is the green thing that the chips are connected to. With the launch of Ryzen 7000, I can point to one specific change AMD has made in its design to help with this. The substrate design of the Ryzen 7000 is super thin compared to previous generations. AMD has engineered a substrate with fewer layers, allowing its partners to use those extra layers in order to make more substrates. The thinner design correlates to high volume. Doing this is difficult and required R&D, and actually one of the downsides is platform integration (specifically it increases the size of the heatspreader to maintain compatibility with previous generation cooling mechanisms) which increases running temperature, but within qualified parameters. If substrate oversupply occurs, AMD could return to thicker substrates (easier to design, end up with better thermals), however this is now a tool and feature that AMD has in its pocket.
Update on November 11th
When I started writing this piece, AMD’s 4th Gen EPYC (Genoa) had not been launched. It is now out in the market, and the first performance numbers are in – I’ve got a dual socket Genoa system waiting at home for me, but I was able to run some numbers.
It certainly is very fast, and other media reviewers are showcasing that by all expectations, AMD will surpass Intel in performance quite easily, perhaps by a full generation. I found it is also a lot more power efficient than the previous generation EPYC, which had been one of the downsides of adopting the platform. AMD’s pricing is competitive, and the value proposition for general purpose cloud and enterprise is currently, unmatched. It’s going to be difficult for Intel’s Sapphire Rapids to compete here when the product launches in the next few weeks given what we know already. What does this mean for AMD’s server growth? Personally I think it’s going to remain steady. We’re seeing a steady 1-4% growth in market share per quarter, and even with the expected cloud growth next quarter, AMD still has to overcome the incumbent player in the market. I’m expecting to test Intel’s Sapphire Rapids as well as it is made available, so it’s going to be interesting exactly where Intel wins and where AMD wins.
AMD Links
Analyst Q&A (unedited)
Q: Wells Fargo - Client down in Q3, work through inventory, how think about where the bottom is?
A: PC biz been volatile, underperformed in Q3. Q4 guided PC down again, that will be significant. Monitor macro, but exit year in better place.
Q: Server supply for ramp?
A: Been ramping, made progress through year. More supply in Q4 than earlier in year. Some investments coming online Q4. Into 2023 with Genoa not to be supply constrained.
Q: Goldman Sachs - Outlook for 2023 in Server, how think +ve against macro? Grow 20-30% range?
A: Little early to talk. In nearterm, some macro headwinds in all markets. It varies by segment - NA cloud is most resiliant, AMD is strong, good progress. Into 2023, expect growth
A: China has been weak in 2022. Not forecasting recovery in 2023. On Enterprise, most impacted by macro, have seen customers take longer to make decisions. Feel good about product portfolio, even with choppy market, can gain share. DC trend are good.
Q: GMs still 300bps down in Q2, client is down, but server and embedded is up. From a mix, should be better, what's driving margins down?
A: Client margins coming down - Q3 all biz in line except client. In DC, Cloud has lower ASPs. Some new capacity from supply, there some additional costs from the agreements, they are baked in.
Q: Bernstein - Extra week in Q4 driving extra revenue?
A: Not really counting on it. Plus and minus in quarter, Q4 guidance is based on gaming and PC being lower. Also holidays. DC and embedded are higher, but not expecting material impact.
Q: Units vs pricing. In Q3, is client margins come down, how did units vs pricing do and into Q4.
A: Look at 3Q. Lots of dynamics - market was weak, consumer was weak, our footprint most in quarter. Units down, ASPs also down.
A: ASP is still up YoY, been discplined in the pricing environment, did see dynamics in Q. Continue to watch. Q4 expected to be competitive, embedded that in the guide.
Q: Bank of America - What is the client revenue for Q4? $800m-$900m? What does client recovery look like - $2B? Competitor suggested client TAM only down 4-5% in 2023. How to derisk the model?
A: In Q4, we're guiding modestly down for client and gaming. Coming off a low Q3 base. Also correct inventory, so we're going to undership consumption in Q4. For 2023, lots of factors - this year PC down almost 20%. Next year, down single digits would be good, to 10%.
Q: On Gross Margins - how much are the right-off or pricing actions in Q3 and in Q4 will they recover in Q1?
A: Look at Q3 solo, came in 50% GM, we had $160m charges, if you adjust for those, the margin in Q3 is 52%. We guided to 54%, came in at 52%, due to weak client PC market and client margins down.
Rest of business went to expectations. Feel good about products, increase margin, guide 51%, go from there.
Q: Cowen - DC segment. Look back to pre-announcement, came in a little bit under. Cloud revenue is still double YoY, can you break down DC between cloud strength, but there were some enterprise headwinds on HPC from CPU/GPU.
A: Focus on Q3, cloud did more than double. Enterprise was down. GPU side, had strong Q in 3Q in 2021, that was down too. NA Cloud is best DC segments, some macro but resiliant. Market share is very good there, expect to grow.
As Biz is more cloud weighted, that tends to have downward pressure on margins. But focus is on footprint on cloud and enterprise. Positioned well for platforms.
Q: Super dynamic in PC market - demand changes a lot. How did Q progress from original guidance? How quick did you see it, how does this impact monitoring inventory or operational changes?
A: Very volatile. Weaker than expected - consumer is where we are weighted and that was weak. Overall macro econ has also weakened, customers have become more cautious. Sell through inventory but not replenish stock to same levels.
Previously wanted to gain inventory - but situation is dynamic. Better relationships in supply chain, market remains volatile. All biased towards reducing inventory levels.
Q: Morgan Stanley - Visibility in embedded/Xilinx?
A: Embedded biz done really well, strong Q3. See growth in Q4. Been focused in the sub-segments. It's very broad - comms was up, aero was up, auto was up. Some weakness in consumer. Some subsets of industrial was weaker.
Still supply constrained on legacy nodes, but progress on supply in Q4. Multiple Q visibility due to long lead times. Also gained share.
Q: Deutche Bank - Q4 down modestly in client and gaming. Visibilty for DC?
A: On absolute, flat to Q4. Modest decline with gaming, expect gaming to be down in Q4 but that's offset with new graphics launch. In DC/Embedded, roughly same, modest flat to plus.
Q: OpEx, desire to stick to Analyst Day biz model, vs gaining market share and new products? Are buffers conflicting?
A: Continue to invest in strategic areas. Roadmaps, supply. Slowed down hiring. Remain disciplined. Into 2023, bring more in balance with revenue.
A: Multiple levels in Opex. Feel very good about DC, embedded, commercial direction. Continue to invest. Be more conservative in consumer-facing through 2023.
Q: BMO - PC biz, disparity between AMD Q3 guide and Intel Q4. Things usually normalize, but is it a case Intel had the medicine earlier and AMD overshipped?
A: As we exited 1H, we were planning for a reasonable 2H PC half. Usually 3Q is seasonally higher, we built with our customers. Market is cautious, behaviour is changing.
Worked with customers on inventory. There were some temporal dynamics, we are aware. Some of those are aggressive pricing that we didn't follow. Nothing fundamental, portfolio is strong, very good platforms in place. Work through it!
Q: Will investments in capacity taper off over the next few Q?
A: Cap agreements have some benefit for server space. Long term agreements to support that growth. Given the market backdrop, work with suppliers with timing for ramp. There is some flexibility overall, work on next Q
Q: Instances in Milan - ramp of Genoa. How long does Milan momentum continue into 2023, and how are customers with Genoa?
A: Happy with Milan ramp. Continuing to ramp Milan into 2023. Genoa initial shipments to strategic customers to Q3, continuing in Q4.
A lot of the qualifications and first instances, Q4 to Q1. Milan and Genoa are going to co-exist for quite some time, both are very competitive.
Q: Harlan - Cloud customers growing 30-40%. Building more compute, is this building strong mid-term outlook?
A: In near term, some optimization each cloud vendor is doing. In med term, customers telling us they need more compute. Additional workloads building out. Also upgrade older compute. New product, strong TCO, more efficient.
Q: Citi - share expectations, desktop, notebook, server, next 12 months, taking macro out of it?
A: In DC, we expect to gain share in cloud and enterprise. More underrepresented in enterprise. NA Cloud see line of sight to significant new deployments.
In desktop/notebook, focus is on certain segments, premium, gaming, commercial. In those areas, there are opportunities to continue to gain share. Desktop DIY/channel, have strong lineup, additional products next year.
Q: If downturn in PC continues, ASP declines, would is spread to DC, and would you be willing to give up share?
A: AMD has strong value prop. Our strategy is not to lead on price. Also price in DC is one factor, it's not the leading factor for most customsers. Work to do for us in workload optimization in DC.
As to PC biz, PC biz has become more price sensitive, and will remain competitive in focused profitable markets.
Q: UBS - US/China. China is 25% of AMD revenue, but that's sell-in. How do you think about US/China trade?
A: Watching closely. New regs have come into effect, minimal impact to AMD revenue near term. Working closely with DoC and US Govt. On Med term basis, our view is to follow US regs and need for national security. Majority of China biz is non-DC, it's mostly PC, consumer-facing.
Q: Run rate of PC in Q4? Inventory leans out a bit? Market only absorbed 50m units in Q3, 200m per year, which would be bad for full year. August/September must have been bad.
A: We did drain a good amount of inventory in Q3, drain more in Q4, will make progress. Also depends on macro in next couple of months. Good visibility into market factors. Planning for weaker PC.
Q: Barclays - client business, units vs pricing? Mix away from high-end vs competition?
A: Certainly more units. ASPs down seq, but up YoY. Decline was driven by units, but small ASP factor QoQ. As market weakened, more pressure in high-end, not just velocity. Mix towards lower ASPs, towards more incentives to clear inventory.
Longer term, dont think there's a significant mix change.
Q: GMs down 100bps down Q3 to Q4, mix in favor for Cloud/Embedded. What's the reason for the core margins to be down?
A: Excluding the $160m charges in Q3, we were 52%, we're guiding to 51%, continue with client being weak, clearing inventory. In DC, we see mix is high for NA Cloud, but that's lower ASP, but as we increase capacity, that has a bit of cost impact.
Q: DC / Cloud.
A: DC not necessarily up, macro has impact on spending. Our weighting is towards NA Cloud. As it relates to the market, enterprise is weak (biz being more cautious), China is weak. There's no in general with cloud, some are optimizing footprint, some are building.
Depends on specific customer. Working with customers for 2023, coming from supply constrained, we see which workloads are moving - gives us visibility.
Q: Gaming market in general - is this the peak gaming year in 2022? Still difficult to get a PS5.
A: Still some pent up demand into the holiday season. Q3 is the peak, Q4 down seasonally. For 2023, puts and takes, best to model gaming flattish, see how the holiday season goes.