As the economic downturn brings the astronomical level of investor hype over semiconductor startups back to Earth, some venture-backed companies think now is the right time to build for the next boom period, while others stumble and crash.
Global venture capital funding for semiconductor startups reached new heights in 2020 and 2021 after several years of modest growth, but this year’s economic realities have translated into substantially less capital for privately held silicon firms that rely on investor cash to survive or grow.
According to data provided by PitchBook, global VC funding for semiconductor startups in 2022 reached $7.8 billion, as of December 5. That’s a 46 percent decline from the record $14.5 billion investors pumped into silicon firms last year and a 24 percent decline from the $10.3 billion raised in 2020.
At the same time, while the number of worldwide semiconductor funding deals fell this year to 618, it only declined by nearly 20 percent from the 771 deals recorded last year, and it was actually nearly 21 percent higher than the mere 511 funding rounds in 2020.
Investor scrutiny increases, creating greater risk for some startups
Ruta Belwalkar, private investor and chip designer, told The Register that this year’s slowdown in economic activity has increased investor scrutiny over the viability of semiconductor startups.
While the bar has always been higher for silicon firms than software startups because of how capital-intensive they are, investor money was easier to come by previously.
“But now specifically, what they’re asking is, ‘Do you have a customer? Have you taped out your first chip? And do you have a roadmap to the future generations?'” Belwalkar said.
Belwalkar is referring to fabless chip design firms that need to raise enough money to hire people, design integrated circuits, and then pay tens of millions for a tape-out, the final step of the design process where a photomask is sent to a contract chipmaker like Taiwan’s TSMC for manufacturing.
One chip design startup that apparently lost investor interest this year is Mythic.
The Texas-based firm, which had raised a $70 million round last year, tried to stand out by designing analog chips for edge AI use cases, but it ran out of venture capital funding before it was able to generate revenue, according to a November post by a top executive in November. (Company officials have declined to comment further.)
Belwalkar said she wouldn’t be surprised if other chip design startups met a similar demise soon because they didn’t transition fast enough from research and development to commercialization.
“Now if a startup doesn’t have the runway to survive the next year, and they can’t raise money by mid next year, then maybe they will run out of money. I’m not saying that their IP is bad or anything. It’s just there are high chances. Sustaining that many people on the team, it’s tough,” she said.
Alternatively, a startup could end up acquired if there is an interested buyer.
However, such interest could be diminished by increased regulatory scrutiny over merger and acquisition deals as well as capabilities developed by incumbent semiconductor firms, according to a recent note by PitchBook analyst, Brendan Burke on the state of AI chip startups.
Burke is talking about the objections by Western regulators that killed Nvidia’s $66 billion bid to acquire Arm earlier this year, plus acquisitions made by Intel and AMD over the past few years – Habana Labs and Xilinx, respectively – that boosted each of their AI chip capabilities.
There is an area where chip startups could see more M&A interest – automotive – due to larger players like AMD and Nvidia lacking some capabilities compared to auto-focused chip companies.
“The size of the market encourages large bets to capture market share from the range of automotive chipmakers led by Infineon, NXP, and Renesas,” Burke said.
In commercialization, some stumble while others puff their chest
Even if a semiconductor startup starts selling to customers, it doesn’t guarantee future success, a reality that can become more pronounced when the economy goes down.
Graphcore, a well-funded, Bristol, UK-based AI chip startup that has aimed to compete with Nvidia, reportedly had its private valuation slashed by $1 billion this year after losing a key deal with Microsoft, amid other financial woes. The Times reported in October that while Graphcore’s revenue grew slightly to $5 million last year, so did the company’s losses, to $185 million. The struggles prompted Graphcore to lay off roughly 170 employees this year, the newspaper added.
“Graphcore has significant cash reserves and is well positioned … however the macroeconomic backdrop is extremely challenging. This means making some hard but necessary decisions around our priorities to put us in the best possible position for sustainable growth in 2023,” Graphcore said in a statement to The Times.
The 2022 State of AI Report, authored by two AI-focused venture capitalists, underlined the difficulty of smaller AI chip companies competing with Nvidia, showing that GPUs are referenced in AI research papers 90 times more than chips from Graphcore, Intel’s Habana Labs units, and three other well-funded startups: Cerebras Systems, SambaNova, and Cambricon. Out of the smaller Nvidia rivals, Graphcore had the most citations in research papers in 2021 and 2022.
“We’ve got years of runway, and we’ve got many paying customers, and I don’t think the others have,” Andrew Feldman, CEO of the Silicon Valley wafer-scale AI chip company, Cerebras Systems told The Register.
Feldman declined to discuss financial figures of his own company, but he said customers are buying more of its systems “year-over-year.” He added that the startup won’t have to raise another funding round for the next six to nine months.
“How do you know when you’re doing well in a difficult market? When your customers buy more, when you have more customers, and when you’re solving really hard problems for them,” he said.
For some, now’s the best time to build
Cerebras isn’t the only venture-backed silicon firm that’s feeling confident about the future in the midst of an economic downturn, and for two startups, one major reason is how they plan to take advantage of two growing trends in the semiconductor space.
One of them is Eliyan. The Silicon Valley startup announced in November that it had raised a $40 million funding round to commercialize its die-to-die interconnect technology, which the company claims will make chiplet design – increasingly embraced by the industry as the superior way to design chips – more economical and efficient than advanced packaging solutions.
Ramin Farjadrad, Eliyan’s CEO, told The Register that with the economy weighing down on margins for semiconductor firms, the need for its solution has increased because it could help them save on chiplet manufacturing costs in the future.
“One of the key things that we offer as part of our technology by eliminating advanced packaging is improving the overall cost for these types of products,” he said.
Farjadrad said the current downturn has come with other benefits. With demand cooling in the industry, Farjadrad said he has been able to obtain certain materials for less money and quicker. And with semiconductor stocks down from earlier this year, the labor market has become less competitive and made it easier for Eliyan to hire technical talent from larger companies.
“Even up to six months ago, we had maybe trouble snatching really good guys from the big companies, because their [restricted stock units] were high and so on and so forth,” he said. “But right now, it’s much easier. People even approached us, so we don’t have to spend as much money for recruitment.”
Astera Labs is also in the middle of a hiring spree, though the semiconductor startup wasn’t originally planning to do so as a privately held company for much longer. That’s because it had hoped to go public before the year’s end and then decided against it.
Instead, Astera Labs raised another funding round this year from investors, worth $150 million, that more than tripled its valuation to $3.2 billion, going against 2022’s trend of tech companies seeing their valuations dramatically reduced.
- AMD was right about chiplets, Intel’s Gelsinger all but says
- This startup reckons its chiplet interconnect tech can best Intel, TSMC
- Intel is doing so well at AI acceleration, it’s dropped $2bn on another neural-net chip upstart (third time’s a charm)
- Arm’s $66bn sale to Nvidia is off: Deal collapses after world’s competition regulators raise concerns
Jitendra Mohan, CEO and founder of Astera Labs, told The Register that the Silicon Valley firm didn’t need to raise more capital because it has already been generating a “tremendous amount of revenue” and has a decent balance sheet.
However, Astera Labs and its investors believe the company is on the cusp of a major opportunity because it is developing chip architectures that will allow hyperscale and cloud customers to take advantage of Compute Express Link. CXL is a standard being introduced in new Intel- and AMD-based servers that will, among other things, enable cheaper, more flexible, and larger DRAM configurations as well as memory pooling.
“We find ourselves in a leadership position on CXL. Our investors, along with us, looked at the business and said, ‘Look, it’s time to put the pedal on the gas’,” he said. ®
