AMD: Winning The Generative AI Battle Likely Won't Be Easy

Summary

• AMD may have won two of the supercomputer contracts thus far. However, it still lags behind NVDA in terms of PC and discrete GPU market share.
• While it remains to be seen how AMD's MI300 may perform, market reviews for NVDA's H100 have been stellar for training large language models.
• With the battle for generative AI at a fever pitch, it appears that market sentiments surrounding AMD's execution remain uncertain, worsened by its lower-than-expected forward guidance.

Aksonov

The Generative AI Investment Thesis

Despite its excellent GPU offerings, it appears that market sentiments surrounding Advanced Micro Devices' (NASDAQ:AMD) prospects are still mixed, attributed to the stock's less than impressive recovery against its peer, Nvidia (NVDA).

AMD & NVDA 6M Stock Prices

Seeking Alpha

At the time of writing, AMD only recorded a 45.08% recovery from the October 2022 bottom, compared to the S&P 500 Index at 12.9% and NVDA at 109.76%. This was perhaps attributed to the hype surrounding ChatGPT, which had been trained on NVDA's A100 GPUs, especially aided by the latter's tailwinds for financial recovery from the promising early recovery of GPU sales.

Furthermore, AMD's execution seemed hampered by the ongoing PC demand destruction, due to the prudent FQ1'23 revenue guidance at the midpoint of $5.3B (-9.8% YoY) against the consensus estimates of $5.56B. The macroeconomic outlook through 2023 appeared uncertain as well, with the January 2023 CPI still indicating a -6.2% YoY impact in the consumer index for computer products.

Nonetheless, we remained confident about AMD's recovery, attributed to the promising MI250 Accelerated Processing Unit [APU], a hybrid CPU-GPU-accelerator chip for intense cloud computing works. The company had reported great success in sampling the product in hyperscalers thus far, with the new MI300 to be launched by H2'23 as well.

Given the notable improvements in performance compared to MI250 by 8x fold in performance and 5x fold in power efficiency, it is unsurprising that the MI300 has been selected to power the world's fastest supercomputer, El Capitan, in Lawrence Livermore National Laboratories by late 2023.

By then, investors may be greatly encouraged that the first and second of the world's supercomputers (by the end of 2023), will be utilizing AMD's market-leading APUs, suggesting its market dominance against peers, such as NVDA or Intel Corporation (INTC). This naturally builds upon the company's strategic 3D stacked chiplets, which boosts computing efficiency through improved energy performance, data sharing, and memory use.

AMD has also been working closely with many large cloud providers to qualify the MI300 for AI workloads, suggesting potential revenue contributions from H2'23 onwards. This may be the company's tailwind for market share growth, significantly aided by the exciting new chapter in high-performance computing for generative AI.

Particularly, Microsoft's (MSFT) Azure has been using AMD's MI200 for "large-scale AI training" since May 2022, which was reportedly 5x more efficient and 1.2x faster than NVDA's A100 GPUs. Based on the recent interview at the Morgan Stanley Technology, Media & Telecom Conference, MSFT has also been utilizing the company's MI250 since 2022.

These hardware advancements also leverage AMD's existing AI software stack, ROCm, and Xilinx AI engine capability, demonstrating the company's stellar integration thus far. Considering that the Xilinx acquisition was only completed a year ago, we reckon there are further tailwinds for opportunistic portfolio synergy in the company's generative AI capabilities across hardware and software offerings indeed.

Nonetheless, it remains to be seen how NVDA's H100 (the successor to A100) and AMD's MI300 may compare against each other, since the latter has yet to be released.

Early reviews for H100 has been stellar, with the product delivering faster speeds by 9x in AI training and 30x in AI inference on large language models compared to its predecessor. Furthermore, with the MI250 supposedly outperforming NVDA's A100, it raised the question of why OpenAI's ChatGPT was not trained on the former instead, given their similar pricing points, with MSFT already unveiling new virtual machines powered by NVDA's H100.

In addition, AMD still had a long way to go in competing with NVDA in the PC or discrete GPU end market, due to the latter's growing dominance thus far. By FQ4'22, NVDA commanded 17% of the PC GPU market, with AMD reporting 12% and INTC reporting 71%. At the same time, NVDA also maintained its lead at 84% in the discrete GPU market, compared to AMD at 11% and others at 5%.

Furthermore, AMD's latest flagship Radeon RX 7900 XTX graphics card (released in December 2022) could not compete with NVDA's flagship GeForce RTX 4090, only going head-to-head against the latter's slower GeForce RTX 4080 (November 2022).

With most (if not all) comparisons touting NVDA's flagship chip as the outright winner, it appeared that AMD still lagged behind the market leader. Only time will tell how the generative AI battle may turn out for AMD indeed.

In the meantime, anyone interested in extremely technical hardware analyses may refer to these links below, which we found to be rather interesting during our research:

• The Odious Comparison Of GPU Inference Performance And Value
• AMD Teases Details On Future MI300 Hybrid Compute Engines
• Will AMD's MI300 Beat NVIDIA In AI?

So, Is AMD Stock A Buy, Sell, or Hold?

AMD 1Y EV/Revenue and P/E Valuations

S&P Capital IQ

For now, the pessimism has been reflected in AMD's valuations as well, with the stock trading at an EV/NTM Revenue of 5.41x and NTM P/E of 26.62x, against NVDA's expanded numbers of 19.50x and 52.91x, respectively.

Therefore, we reckon AMD is still arguably a value play at these levels, especially given our price target of $116.59, suggesting an excellent 43.6% upside potential from current levels. The number is based on market analysts' projected FY2024 EPS of $4.38 and current P/E valuations, which is notably higher than its 1Y mean of 21.79x though lower than its 3Y pre-pandemic mean of 37.33x.

AMD 1Y Stock Price

Trading View

However, AMD has also bounced off its February support and is trading above its 50-day moving average. Therefore, while we remain confident about the company's forward execution, the optimism surrounding generative AI may be at a fever pitch, suggesting potential volatility in the short term.

As a result, investors who choose to add AMD here must only do so if it consecutively reduces or matches their dollar cost averages. Otherwise, bottom fishing investors may try waiting for another dip to the low $70s at its previous February 2023 support, likely expanding their margin of safety.

Disclosure: I/we have a beneficial long position in the shares of AMD, NVDA, INTC, MSFT either through stock ownership, options, or other derivatives. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.