TheChek News Flow

AMD has the most downside because its shares rallied on the deal, making it the most expensive of the three.

The deals keep coming for semiconductor firms and OpenAI. In the past month, Nvidia, Advanced Micro Devices and Broadcom have each reached very different agreements to supply chips to OpenAI.

The arrangements open up all three to the risk that OpenAI, which expects to burn $115 billion through 2029, can’t afford to keep its commitments. The deals pose different risks to each company’s business, but AMD faces the biggest downside because its shares have rallied more than 40% since the announcement of its deal last Monday.

Before the deal news came out, all three chipmakers were trading at around 32 times next year’s earnings before interest, taxes, depreciation and amortization. After the rally, AMD is now at 43 times next year’s Ebitda compared to Broadcom’s 33 times and Nvidia’s 27 times.

Broadcom’s announcement, which involves a partnership to develop custom chips with OpenAI, is the most clear-cut. Nvidia, in its deal, pledged to invest in OpenAI as its build-out of artificial intelligence data centers progresses. The structure funds OpenAI but increases Nvidia’s exposure to the risky startup.

AMD last week pledged to give up to 10% of its own stock to OpenAI, in the form of warrants, in exchange for OpenAI’s commitment to buy its AI chips. Both Nvidia’s plan to invest in OpenAI and OpenAI’s plan to invest in AMD will play out over time and are contingent on OpenAI actually using the chips. In AMD’s case, the chipmaker isn’t planning on putting up any cash up front, and it will only give OpenAI the final chunk of its warrants if its own stock price roughly triples first.

Investors in AMD worry about what will happen if OpenAI doesn’t grow as fast as expected. “The question is, if OpenAI overestimated demand for compute, who will they support?” said Jamie Meyers, who helps manage the holdings of the three stocks at $1.6 billion asset manager Laffer Tengler Investments. “We suspect the pecking order is Nvidia, Broadcom, AMD.”

AMD CEO Lisa Su told investors in a call on the day of the deal announcement that the deal’s structure would help align AMD’s incentives with OpenAI’s.

“OpenAI actually has to do a lot of work to make sure that our deployments are successful.…The more OpenAI deploys, the more revenue we get, and they get to share part of the upside,” Su said.

Nvidia’s deal to invest up to $100 billion in OpenAI should also boost its top line meaningfully. OpenAI might end up leasing the chips from Nvidia rather than buying them outright, which could reduce risk for Nvidia. Still, its stock is down slightly since the announcement. That’s likely because it has the most to lose if the broader AI buildout falters. In the most recent fiscal quarter, chips used in data centers, primarily for AI, made up almost 90% of its revenue.

AMD and Broadcom are better diversified. Just over 40% of AMD’s revenue in the most recent quarter came from data centers, up from around 20% in 2021. The rest of its sales come from selling chips for a wide variety of uses, including personal computers and industrial processes.

Broadcom gets just a third of its revenue from chips and other equipment for AI data centers. Its software unit accounts for just over 40% of sales, boosting its margins above those of other chipmakers.

Analysts and investors expect that AMD’s data center segment will continue to grow faster than its other businesses on account of its AI chip deals, making AI a key reason for AMD’s relatively high valuation.

In the last few years, “there have been periods where AMD has traded at a higher multiple than Nvidia because I think there was some anticipation that they’re a new entrant to the market, and there was a compelling view of their foray into this AI GPU market,” said John Vinh, a KeyBanc Capital Markets analyst.

The problem is, AMD’s high valuation means that some of its success in the AI GPU market is already priced into the stock. Vinh says he downgraded the chipmaker’s stock a few quarters ago from overweight to neutral. There are also signs of slowing sales momentum for AMD’s AI chips, said Vinh, who is nonetheless optimistic about the company’s new chip, coming next year.

The bullish case for AMD, Vinh explained, is that the company could someday carve out a 20% market share in AI chips by taking business from Nvidia, which analysts estimate has between 80% and 90% of the market. Still, Nvidia has the technological edge, which could limit AMD’s gains.

Broadcom, whose shares are up almost 10% since the deal, has another advantage over Nvidia and AMD that could protect it if OpenAI struggles. Broadcom designs its chips from scratch for its customers, who as a result are less likely to go back on their purchasing commitments. Nvidia and AMD sell the chips as finished products.

Nvidia’s grip on the AI chip industry might be tough to break in part because of the company’s software, used to program its chips. Foundation Capital investor Ashu Garg, who has invested in companies such as Databricks and Skyflow, says most of the engineers at companies he’s backed are familiar with Nvidia’s software and don’t want to switch to AMD’s equivalent. “They don’t want to learn something new,” Garg said. “They’re like, OK, [AMD systems are] 5% or 10% cheaper. So what?”

Nvidia’s technological advantages in the AI chip business are reflected in its pricing power. That’s surely one reason why AMD has a much lower gross margin—below 50%—than Nvidia and Broadcom, whose margins are around 70%.

At least for now, Nvidia’s technological edge suggests it should be trading at the highest valuation multiple of the three chipmakers rather than the lowest. In the same vein, AMD’s deal with OpenAI suggests it should be at the bottom of the list.

Still, all three stocks reflect, to varying degrees, the same underlying risk.

“We think over the next year or two, demand is going to materialize that OpenAI is going to take advantage of to continue building out [capacity],” said Laffer Tengler’s Meyers. “But in four or five years, could we be having a different conversation? Absolutely.”