AMD Poised Interestingly Ahead Of Multiple Product Ramps

There have been no earnings previews at Beyond The Hype this earnings season. Earnings previews and even earnings reports are of reduced importance in the current wayward and mercurial policy environment.

It has been a volatile few months with tariffs, trade uncertainties, product restrictions, and concerns about worldwide economic malaise. Coupled with tariff-related demand pull-ins and pushouts, these dynamics make it very difficult to estimate companies’ short-term results. And whatever earnings that companies announce or the guidance they give leaves open the possibility that the results are not a good representation of the companies’ prospects and are artificially impacted.

Earnings related bets should generally be limited even at normal times and they become super speculative at times like now. In context, this note is an exception.

The long-term trajectory of Advanced Micro Devices (AMD) has changed very little during the last few months. And it is unlikely to change much in any given quarter if the company continues to execute its strong product roadmap. But a long-term view does not explain why the Company stock has been beaten down as much as it has been.

There are two negative narratives going into Q1 earnings that are driving the narrative and putting pressure on the stock price.

1. Expectation that AMD’s data center GPU revenues will drop between $1B and $2B in 2025 due to the new restrictions on China-market MI308. Supply chain leaks have indicated that AMD CoWoS orders are down as a result. As a result, several brokerage houses have now cut 2025 GPU estimates to the $5B to $6B level from the previous $7B to $8B level.

2. Expectation that Intel (INTC) is waging a price war and driving down ASPs. Despite demand pull-ins, the price war expectation seems to be limiting analysts to forecast only a modest $100M PC driven beat for AMD in Q1. There is a further expectation that since Q1 upside is due to tariff related pull-ins, Q2 will be softer than seasonal.

As a result of these expectations, while AMD’s Q1 revenue and EPS expectations have moved up slightly or stayed flat, Q2 and 2025 numbers have gone down. Consensus Q1 revenues now stand at $7.1B with $0.94 in EPS.

Let us discuss the reality of these two vectors and the disconnect between sentiment and reality.

Is The GPU Business About To Disappoint?

There is little doubt that the focus of investor attention will be on the GPU business. The main questions are twofold:

1. Do the MI308 restrictions mean a $1B to $2B revenue cut to 2025 GPU guidance?

2. More importantly, will MI355 help AMD gain material amount of GPU business or is the GPU business likely to flatline until MI400?

Firstly, the backdrop for GPU demand in Q1 was strong (note that Q1 demand dynamic is different from widely reported hyperscaler lease cancellations which will not have impact on demand until 2026 or 2027). All major hyperscalers have indicated that demand is strong and capacity in Q1 was tight.

Furthermore, while Beyond The Hype predicted a strong late-Q1 GPU uptick in China due to DeepSeek if there were no new restrictions, very few analysts have had that expectation. It appears that DeepSeek demand spike occurred based on the relatively large size of the MI308 write-down guidance that AMD provided. The question for investors is if AMD shipped a meaningful amount of MI308s in Q1 prior to the new restrictions. Given the buzz out of China earlier in the quarter about MI308 and the timing of the new restrictions, it is likely that AMD was able to ship MI308s in Q1.

Note that, for a couple of quarters now, there has been a strong expectation of further tightening of China export controls. Given this backdrop, it seems highly unlikely that AMD CEO Lisa Su would include huge China numbers in guidance. Between the hyperscaler demand strength and China MI308 demand, there is likely a $200M +/- $100 upside to Q1 guidance.

While the “up to $800M” write off for MI308 is likely to hit the Q2 numbers, it seems unlikely that the impact on MI308 restriction would lead to a huge guide down. If this assessment is accurate, the consensus GPU numbers should move up after the earnings call.

The question about MI355 is tougher to answer until we can be sure of new hyperscaler design wins or until the GPU is officially launched in June. But the prospects for MI355 are strong. Based on reading the teal leaves, Beyond The Hype is nearly certain that MI355 will have performance much higher than the current Nvidia (NVDA) Blackwell B200 chips and closer to that of upcoming Blackwell B300.

While MI355 will be is strong product, Blackwell will continue to enjoy a strong scaling advantage. While it is not confirmed yet, what MI355 seems to be missing is the scale-up connectivity (the new UALink fabric not ready yet) and this means that there is no competition to NVLink until the MI400 generation.

The lack of strong scaleup fabric limits MI355 to HGX class systems (i.e. 8 GPU connectivity) and leaves the larger rack scale NVL72 class systems space to Nvidia. As a result, GB200 and GB300 are likely to dominate training as well as reasoning inference applications.

However, a very large percentage of chatbot demand does not need reasoner models due to latency and cost issues. Zuckerberg at Meta, for example, does not seem to be expecting a big infrastructure move to reasoners. For the large demand space that does not need reasoning inference, MI355 would not be at a disadvantage.