AMD Analyst Conf: 4 things to know.

Buy more.

• AMD is revising up its AI TAM, again. In Dec ’23 AMD said $400bn by 2027. In Oct ’24 $500bn by 2028. Now it’s $1tn by 2030. That’s 40% CAGR for the TAM and 60% CAGR for AMD total Data Center revenues incl Server CPU, NIC, NPU, etc.

• AMD is usually shy on guidance but not today. AMD provided a 3-5 years outlook that points to ~40% Net Income CAGRr to 2029-30. If we assume some derating (the stock is now trading at an unsustainable 50x), Market Cap will grow at ~20% CAGR. This implies that your share price will 2.1x to ’29 and 2.5x to ’30.

• If AMD is correct on its market share gains, Intel is in trouble. AMD says it will reach 50% Server CPU value share (3-5 years out), AMD is now at 40%. AMD targets 40% PC CPU share, AMD is now at 25%. Remember that PC is the cash cow for Intel (30% OPM).

• How to compete against Nvidia in Data Center Accelerators? IMO, AMD gets it – whereas Intel still doesn’t get it. It is not do everything yourself. It is an open source development platform, tons of software alliances, different levels of board-to-rack integration.

All the slides are available here: https://ir.amd.com/news-events/financial-analyst-day

Total Addressable Market TAM revised up, once again

AMD annual event, either “Advancing AI” or today’s Analyst Day.

• Dec ‘23 AMD estimated its Data Center Total Addressable Market to $400bn by 2027

• Oct ’24 revised up to $500bn by 2028

• Nov ’25 revised up to $1 trillion by 2030

This was Oct ’24

And today:

AMD 3-5 years forecasts: ~40% Net Income growth

AMD released very big numbers on all metrics:

This slide below refers to total Data Center revenues, including Server CPU: >60% CAGR.

There is another slide that probably refers to Accelerators only (GPU, DPU) and growth is higher: >80% CAGR. So we have a range of numbers:

• Traditional revenues (PC, Gaming) >10% CAGR

• Total Data Center revenues, including Server CPU >60% CAGR

• Accelerators only (GPU, DPU) >80% CAGR

• Total revenue >35% CAGR

Little bug: if you plug in 60% CAGR for Data Center and 10% CAGR for “all the rest”, you do not derive 35% CAGR for total revenue. You derive a higher number ~40% CAGR. Never mind. AMD is not going to explain how they calculated these numbers and if some stuff is excluded / reclassified.

If we take AMD’s numbers literally, and let’s say apply to 4 years out (2029) or 5 years out (2030), we get this: Net Income Cagr over 40% in 2029 and 2030.

Next, we need a Net Income multiple (like a PEx that assumes some future dilution like OpenAI shares) and of course:

1) The multiple will normalize. AMD is now trading at 50x 2025 net income and that’s very high, i.e. it assumes high future growth. The stock will derate.

2) What’s a normalized multiple? I think 20x to 25x is the range, the “boring” range for a steady growth Semi company

What we get is:

• Net Income CAGR ~40%

• Market Cap (share price) CAGR ~20%

I had some fun with the PE multiple in this table to get to the magic $1 trillion market cap.

If you want to derive EPS, based on today’s number of shares, you get to US$ 21-28 per share, which is indeed consistent with AMD’s guidance of >$20.

AMD continuous market share gains in the x86 markets, both PC and Server

Here is a graphic illustration of what I would call the brilliant Lisa Su management versus the Gelsinger’s failure. Seriously, losing 35 percentage points of Server value share in 5 years take some determination. In PC, AMD share gains have been more modest, partially because Intel has good products, partially because resource-constrained AMD has put its R&D money on Data Center.

AMD understands the ecosystem approach to Data Center

This spans multiple dimensions:

• Acquisitions for what you don’t have: software, networking, some accelerators, integration

• Huge effort on software compatibility, portability and open sourcing the development platform. If you are the market leader (Nvidia), you can force compatibility to your closed platform (CUDA). If you’re a late comer, you can’t do that: you have nothing attractive to offer.

• You don’t need 1 chip, you need the range of CPU, GPU and other accelerators (FPGA, DPU), the interconnect, the rack integration, the data center level integration… it takes a lot more than claiming good TOPS on a single chip.