Super Micro Computer (SMCI 8.89%), more commonly known as Supermicro, has seen its stock skyrocket by 2,220% over the past three years. That stunning rally was fueled by the rapid expansion of the artificial intelligence (AI) market, which drove data center operators to purchase more of its high-performance AI servers.
A lot of Supermicro's growth can be directly attributed to Nvidia (NVDA 6.18%), which provides the high-end GPUs that process complex machine learning and AI tasks. Nvidia worked closely with Supermicro to design a new line of servers and workstations that fully supported its H100 GPUs, and that tight relationship enabled Supermicro to carve out a high-growth niche with its AI servers in the heavily commoditized market for pre-built servers.
Image source: Getty Images.
Yet Supermicro's dependence on Nvidia is a double-edged sword. It struggled to secure a steady supply of Nvidia's GPUs in early 2023, and its two largest competitors -- Hewlett Packard Enterprise and Dell Technologies -- have also been working with Nvidia to design new AI servers. In its latest 10-K filing, Supermicro admitted that it doesn't hold any long-term agreements with Nvidia or its other suppliers that actually lock them in as exclusive partners.
That's why AMD's (AMD 2.37%) recent expansion into the data center GPU market might be great news for Supermicro.
Why AMD could catch up to Nvidia in the AI race
AMD only controlled about 17% of the discrete GPU market last year, according to Jon Peddie Research, putting it in a distant second place behind Nvidia with its 80% share. AMD's share mainly consists of gaming GPUs for PCs, but it has been expanding its reach into the data center market with its Instinct GPUs for processing AI tasks.
AMD rolled out its first batch of Instinct GPUs (the MI6, MI8, and MI25) in 2017. It launched its newest MI300 Instinct GPUs, which are manufactured using TSMC's 5nm and 6nm process nodes, in late 2023. By several industry benchmarks, AMD's high-end MI300X actually beat Nvidia's H100 in terms of raw processing power and memory usage.
That's a bright red flag for Nvidia, since the H100, which faces ongoing supply chain constraints, still costs about four times as much as the MI300. Nvidia claims the H100 still beats the MI300 when it's running optimized software, but that slight difference probably won't justify its premium price tag for cost-conscious data center operators.
That's why it wasn't surprising when AMD's CEO Lisa Su recently said the MI300 was on track to be the "fastest revenue ramp of any product" in the company's history. Su also estimates that AMD's Epyc CPUs have claimed 25% of the server CPU market at the expense of Intel's (INTC -9.20%) market-leading Xeons. Between the growth of those two businesses and the expansion of its programmable chip business (from Xilinx), AMD has more data center bundling options than Nvidia.
Cozying up to AMD
Supermicro already works closely with AMD to design servers for its Epyc CPUs and Instinct GPUs. In November, Supermicro CEO Charles Liang predicted that AMD's MI300 GPUs, Nvidia's latest GPUs, and Intel's Gaudi AI accelerator chips would all "gain broad adoption and expand our share in the accelerated compute market." In January, Liang predicted that ongoing diversification would "more than double the size" of the company's AI portfolio.
If Supermicro sells more AMD-powered AI servers, it could reduce its long-term dependence on Nvidia and insulate itself from any future supply chain constraints. Competitive pressure from AMD could also drive Nvidia to lower its GPU prices, which would boost Supermicro's gross margins by reducing its component costs.
A well-balanced play on the AI market
Supermicro already generates about half of its revenue from its AI servers, and Bank of America believes it could expand its share of the dedicated AI server market from 10% today to 17% over the next three years.
That's why analysts expect its revenue to grow at a compound annual rate of 42% from its fiscal 2023 (which ended in June) through its fiscal 2026. That's a stunning growth rate for a stock that trades at just 3 times this year's sales.
That low valuation already makes Supermicro an attractive long-term play on the AI market, but its gradual diversification away from Nvidia with AMD-powered servers might make it a more balanced play on the AI market than either chipmaker.
