As we’re in financial season, I frequently produce graphs like this showcasing how revenue is changing with the companies I track. I post them to twitter, but also I’ll share them here.
Takeaways from the AMD Q2 Financial call, in no particular order:
5nm Zen 4 launch in Q3, Ryzen (desktop) and Genoa (server)
RDNA 3 launch in Q4
AMD is still supply constrained, however more is coming online through 2H. This is largely why AMD can post growth predictions.
AMD is about 40-42% weighted in the datacenter, tending to 50/50
Whereas Intel showcased a 30% YTD revenue drop in notebook last week, AMD is saying that client (including notebook) increased 25% QoQ, driven by Ryzen Mobile sales. I asked for a breakdown of units vs ASP mix, especially as the PC market is trending down around 15% according to AMD. I was told the statement of ‘sales’ was purely about revenue, and they weren’t going to break out any more details on that
Console sales will remain strong through Q3, projecting to a record year for semicustom, which offsets the gaming GPU market being weak. No new semi-custom customers, still mostly Sony/Microsoft with a small amount from Steam Deck and Magic Leap.
Embedded, which now includes Xilinx, is up 2228% YoY. That’s not a typo, that’s just because it now includes Xilinx. That being said, AMD has improved Xilinx’s supply chain to increase margins and opportunity. Xilinx revenue grew 20% sequentially (so far).
AMD’s visibility into EPYC demand is strong, with cloud customers already profiling Genoa and demanding it ASAP. AMD is set to focus on the high volume cloud and high volume enterprise SKUs first. They expect a long overlap between Milan (current gen) and Genoa (next gen), especially as cloud and enterprise offerings develop. Some enterprise/OEM offerings are limited today due to lack of components to complete systems (e.g. a Dell server might be limited by a lack of ethernet cards to sell those systems), hence the build out and growth as those markets get better as well
AMD initiated $920m of stock buybacks in Q2, making $2.8B year-to-date. Another $7.4 billion are still authorized, although no time limit
Some people ask why AMD is doing stock buy backs instead of reinvesting in supply or R&D - AMD is already going full bore on R&D, with lots of hiring, which takes time. It’s the time that’s the killer - you can’t simply start a chip design product and pluck 500 people for it off the street. On a lot of projects, more people won’t help either, it’s a case of going through the motions (such as having a child can’t be accelerated, or playing a piece of music as intended). During that time, you don’t sit on money, you return it to investors through buybacks. This gives the company/main shareholders more leverage and control for sure, but also raises share price which in turn can be used as leverage for cheap debt which is sometimes more favourable.
With Xilinx in the mix, AMD’s market share numbers are a little off kilter. In the Q&A, one financial analyst said it’s now in the mid 20% range, which would be double the low 10% numbers we saw a few Q ago. It’s gone up sure, but it’s still a slow increase comparing apples to apples.
AMD is still insulated from a lot of the industry slow down. The markets it plays in (DC, cloud, enterprise, commercial) are more TCO focused than chip price focused. Even in consumer markets, AMD is focused on the high-end which is more about product than price (although is still price sensitive). The super price sensitive markets, such as government and embedded, AMD is shielded from that as they have very little business there. AMD doesn’t have to engage there at this time as they’re spending capacity on high revenue parts and they’re selling like hot cakes. Hence the record revenues.
It was noted that even though revenue grew, and with the Xilinx acquisition, the free cash flow (FCF) was only up slightly. AMD said that this is due to a number of factors - investment in working capital, the tax rate going up from 3% to 10% with large Q2 payments due to cycles, shipment times also affect FCF with supply, and also AMD’s investments in capacity involve a good amount of up-front pre-payment to suppliers (read, TSMC and OSATs).
Non-GAAP GMs were 54%, but GAAP GMs were 46%. That’s a big difference.
Those were my takeaways. I won’t go through the individual numbers here right now (it’s gone midnight), as they’re easily accessible through ir.amd.com along with slides, infographics, and the detailed docs.
If you didn’t listen into the Financial Analyst Q&A, here is my rough transcription. If you’re going to use any of these as quotes, please note that this was done live and not as a secondary listen through. There may be mistakes.
Goldman Sachs Q: Guidance - revenue and margins for Q3/Q4?
A: Q3 revenue growth driven by datacenter and semi-custom, traditional peak in Q3. More conservative in PC compared to last quarter. Current view of PC is down mid-teens this year. In Q4, sequential growth is DC and embedded, with same view of the PC. But also have new Q4 product ramps in 5nm, inc Ryzen and EPYC. For Margins - we're guiding full year at 54%, it's a mix of the businesses.
Goldman Sachs Q: Enterprise/cloud customer visibility? How dependent are you on this? Moving from 1/7th Intel to 1/3rd of Intel.
A: DC biz has grown nicely. Pleased with segment growth QoQ and YoY. In Q2, we saw cloud continue to be strong, AMD still ramping new Milan instances and workloads, also into 2H. In enterprise, making progress in OEM, the trends there are a bit more mixed, more influenced by macro, deals taking longer to close, some match set (DC) that the OEMs are working through. Continue to see growth in 2H and 2023, given strong product positioning. Milan continuing to ramp in 2H, Genoa coming in 2H into 2023. We've gained share, but still under-represented. Visibility with cloud customers is very good, we're planning for next 4Q-6Q.
Bank of America Q: Inventories - is 2H derisked?
A: When we guide, we recognize dynamics. For 2H and FY, we're continuing to see strong DC demand, in embedded, as well as console. We are being more conservative in PC outlook. Our outlook for PC is now 290-300m units, down from before. As it relates to inventory, given some of the 2Q lockdowns, there was some buildup of inventory, but we've taken that into account. AMD portion is modest, will re-balance in 2H. Our portfolio, was still supply constrained in embedded in 2Q, even on server we were tight in 2Q. Still have additional supply coming online to meed demand.
Bank of America Q: Cloud - spending environment with partners with Milan and Genoa vs SPR
A: We work with customers, especially cloud. Each are different, different dynamics, optimization points. A slow down in China, but North America cloud has been strong. Forecasts are robust for next year. AMD is in a share gain position, the product is positioned well, Genoa enhances that in 2023. We'll always be with the customer set, but current view is strong opportunity to grow DC biz.
Stacy Rasgon Bernstein Q: Intel publicly admitted push out on SPR. Does that change Genoa outlook? Do you have the supply next year to upside?
A: We're very focused on our own product ramp. The key for us to continue to work with customer ramps. Genoa looks very good, strong feedback from customers. Lots of interest to ramp quickly in cloud and high-end enterprise. On supply, we have spent last 12 months to build capacity to support the growth the product can handle. Big step up in capacity over 4Q-5Q.
Stacy Rasgon Bernstein Q: Q4 magnitude increases insight?
A: Bulk of increase is led by DC and embedded. Our embedded segment has performed well, pleased with growth. For margin, it's a business mix. DC and embedded are up, but the DC mix is more to cloud than enterprise, embedded weighted to comms. The 7% increase in GM is related to new product mix. It's also a 14 week Q for us.
Ross [?] Q: Gaming and consoles, different dynamic. Client GPU is down YoY in 3Q, what's happening? Is this PC or share gains/product issues? But semi-custom is strong, is that just 3rd year of ramp, or seasonal?
A: On semi-custom, console has been very strong, strong cycle, overall we were more supply constrained 2021 into 1H, we've got additional supply for that, belief it's a strong cycle. Expect console to peak in Q3 and then normal decline in Q4. For consumer graphics, we came off a strong 2021 in graphics where demand was high, we've seen a slowdown in Q2, somewhat due to the supply coming online vs demand and macro issues on consumer spending. In Q4, we're expecting sequential growth due to new products. Gaming as a segment is a long term secular driver with short time dynamics to deal with.
Ross [?] Q: Long time GMs at 57%, how does that change with more capacity coming online?
A: Biz is getting to scale. Increase in margins is a result of mix. DC and embedded can get to over 50% of the company, right now is low 40s. DC to grow faster than rest of the company and drive margin expansion. All working on costs, there are some inflationary costs and we're trying to keep them to a min. Semi cost reductions over time helps. Primary expansion is DC/embedded growing faster than others.
Matt from Cowen Q: Server growth is strong. A lot of that is driven with cloud - roadmap is set to diversify with bergamo/siena. How are the relationships developing with this new diversification?
A: Love progress in cloud, and earn all the share we can. On enterprise, it takes a bit longer due to sales cycles. Made progress with all the top OEMs, portfolios are expanding. That includes Genoa, Genoa-X at the high end of perf, Siena which broadens telco portfolio. Grow share for enterprise in 23 and beyond. Broader opportunity to sell the portfolio with EPYC+Radeon+Xilinx+Pensando.
Matt from Cowen Q: Revenue inc Xilinx up 70%, but Free Cash Flow (FCF) was only up slightly. What are the variables?
A: Investment in working capital with revenue growth. Up significantly YoY. 3% tax rate going up to 10%. Q2 tax, we had some significant payments with govt. Also timing of shipments affects FCF. Also supply for server and overall, making investments in capacity from a pre-payment standpoint for our suppliers. Investing for growth.
Wells Fargo Q: DC trends over quarters, especially unit decline on Intel. Trajectory of blended overall pricing for server CPU? Bergamo/Siena etc.
A: Genoa has more content than Milan - Rome and Milan are 64-core, Genoa and Bergamo are 96-core then 128-core. On a per-unit basis, the ASP would expect to go up. Milan is expecting to co-exist for quite some time given the different infrastructure (DDR4 vs DDR5, sockets)
Wells Fargo Q: DC GPUs, traction beyond Frontier?
A: Working with large hyperscalers, already partnered with MI250X. It's not a big contributor this year, but larger next year as some of those initial engagements turn into production engagements.
Morgan Stanley Q: Is Xilinx still supply constrained into 2023 1H?
A: Xilinx has performed very well. Demand on all segments. We made significant improvements in supply chain, so step up (on pro forma basis), Xilinx portfolio grew 20% sequentially. Still a bit constrained in a bit areas of the portfolio, but making progress. More supply to come online late in the year, into 2023. Quality of design wins, diverse markets are very strong. Able to continue to relieve some of those constraints for trajectory.
Morgan Stanley Q: Processor market ASPs given supply?
A: In DC, focus on TCO, not initial cost of processor. In PC, we've focused on premium segments, ultrapremium and commercial. Those are more about the product. Some parts of the PC market are price sensitive like the low end, but we've reduced our exposure there. Key thing is a strong roadmap
Jeffries Q: DC biz aligns to Intels. AMD gained 6.6% share from Intel, pro-forma from Xilinx is about 6%. That would be highest share gain ever reported, even back to 2005. AMD market share in the mid 20% now? Any outside contribution from Xilinx or Pensando that might mask
A: This is why we went to new reporting segments. Your math is in our zip code. We're pleased we're gaining share, our expectation was as the portfolio expands, as we ramp, as we increase supply across the customer set, we would see share gains. To your question, there was no outside contribution - the other pieces in the segment are small. Mostly driven by EPYC.
Jefferies Q: DC plans in advance. After a lease, do they approach you for chips in advance, how consistent has the variance been around what they're telling you?
A: Planning that we're doing with customers has been helpful. Most of our conversation is about 2023 and capacity for buildouts. Visibility is very good, things could change, but overall the zip codes on growth and content are very active conversations, been active for a few months. Positive is that the supply chain, there's a recognition for better planning and investment.
Harlan Sur, JPMorgan Q: Cloud customers are awaiting platforms, 20 planned product platforms. Are the high volume cloud SKUs first, what does the qualification look like?
A: We go through a full interlock with customers, our focus is on high volume cloud SKUs and high volume enterprise SKUs first. There's a long qualification cycle. Today we see a strong customer pull on Genoa. We expect to ramp production in Q4 into the first half of next year. Diff customers will have different lead times, but strong engagement.
Tim Arcuri UBS Q: PC TAM, lost units * ASP cost AMD $750m, but DC in line as expected. Is there a Xilinx portion?
A: When we put together guidance FY, we understand not everything goes as expected. As we look into 2H, DC is strong, will grow to 2H, embedded will also grow, consoles will also grow. As we look at those components offsetting the conservative the PC outlook, it's quite nice. Product ramps in Q4 in 5nm, that will drive the sequential growth in Q4.
Tim Arcuri UBS Q: Purchase commitments?
A: In 2021, we had $1B committed and paid. This year steps up for supply requirements. In 1H it wasn't significant, but 2H will impact from there.
Here are some of the key slides.
I understand the reasoning behind buybacks, but it still "strange" that AMD is doing buybacks when they are still somewhat supply constrained. They've could've used that money to buy more supply, no?