The Reason Venture Firms See Startup Opportunity In Chips For Generative AI

In the rich Bay Area tradition of raking in profits by selling shovels to gold miners, venture capital firms are seeing opportunity not just in generative AI itself, but in startups that specialize in creating the hardware and software that will enable generative AI and extend its impact. JL 

Isabelle Bousquette reports in the Wall Street Journal:

With the nascent generative AI market propelling demand for compatible hardware and software, it’s a good time to be a startup. “In stable markets that aren’t changing much—very hard to beat Goliath.” But smaller upstarts could also benefit from an overflow of demand, especially as supply-chain and manufacturing difficulties limit the amount of chips making it to market. “There’s new opportunity for those players because the types of chips that are going to most efficiently run these algorithms are different from a lot of what’s already out there. The number trying to apply AI is ballooning and a massive opportunity we can play into.”

As major chip players— Nvidia Corp. , Intel Corp. , Advanced Micro Devices Inc. among them—rush to capitalize on the popularity of generative artificial intelligence, startups are seeing their chance to grab a bigger piece of that pie as well.

“There’s new openings for attack and opportunity for those players because the types of chips that are going to most efficiently run these algorithms are different from a lot of what’s already out there,” said Brian Schechter, a partner at venture-capital firm Primary Venture Partners. 

Historically, Nvidia has been the market leader in specialist AI hardware, analysts said. Generative AI and large language models like OpenAI’s ChatGPT require massive amounts of computing power to run, and typically rely on chips like Nvidia’s graphics-processing units, or GPUs, that are specialized for these types of calculations.

 

Last week, Nvidia Chief Executive Jensen Huang said on a call with analysts that excitement around these new AI developments could supercharge the market for its chips. 

Dylan Patel, chief analyst at chip research firm SemiAnalysis, said the big companies are in a prime position to benefit from the onrush of demand. But smaller upstarts could also benefit from an overflow of demand, especially as supply-chain and manufacturing difficulties still limit the amount of chips that are making it to market, he said.

Cerebras Systems Inc., a Sunnyvale, Calif.-based chip company founded in 2016, has been able to capitalize on some of that interest, said Chief Executive and Co-founder Andrew Feldman. As demand surges, he said, it is creating space for startups to break through. Cerebras is valued at $4.1 billion.

U.K.-based Graphcore provides specialized hardware and software designed for AI, including its Bow processor.

PHOTO: GRAPHCORE LTD.

With the nascent generative AI market propelling demand for compatible hardware and software to new heights, it’s a good time to be a startup, he said. “In stable markets that aren’t changing much—very hard to beat Goliath,” he said. 

“The number of people trying to apply AI is just ballooning and that is really a massive opportunity that we can play into,” said Nigel Toon, chief and co-founder of Bristol, U.K.-based Graphcore Ltd.

Graphcore provides specialized hardware and software designed for AI that can do several things, among them lowering compute costs by eliminating unnecessary parameters, Mr. Toon said. Graphcore sells primarily to AI startups looking to build and train models at lower cost, he said, and the company is benefiting from the proliferation of those startups.

Anshumali Shrivastava, the founder and chief executive of ThirdAI Corp., said that since the release of ChatGPT, his company has also seen an increase in demand. Houston-based ThirdAI provides technology that helps complex AI algorithms run efficiently on cheaper CPUs, or central processing units, rather than on specialized GPUs.

Dr. Shrivastava said because of ThirdAI’s focus on CPUs, it can also feasibly help enterprises unlock complex AI models on premises and not in the cloud—alleviating privacy and data security concerns for industries that require on-premise solutions.   

Shane Rau, who leads International Data Corp.’s semiconductor research, said chip startups are increasingly pivoting to focus their products on supporting large language models. Still, he added, “you’re going to see a combination of real adaptation and marketing.”  

“There will be the pressure to say: ‘Hey, we’re already relevant, our AI chip technology’s already relevant to generative AI’,” said Mr. Rau. “Many of these AI chip companies—we’re tracking hundreds of them—are going to run out of money before they can make that adaptation.” 

Kavitha Prasad, vice president and general manager at Intel for data center, AI and cloud and enterprise strategy, said incumbents like Intel might also have an edge over startups because of the software they provide clients to program and optimize the chips.  

“There are a lot of startups, but without a focus on the software ecosystem, adoption is going to be very limited,” she said. 

Some chip makers say they expect yet another surge in demand once businesses more widely adopt generative AI.

“We think this demand is both overwhelming—and just the start,” said Cerebras’s Mr. Feldman.