Chip Startups Take on Nvidia... All-In on Truth Social... New IPO Value Way Below 2021...VCs Share YC Demo Day Favorites
Plus, new deals from Aerospike, Alterome Therapeutics, Diagonal Therapeutics, Obsidian Therapeutics & Pigment
The Main Item
Chip Startups Go Niche to Beat Nvidia’s Incumbency Bias
Hi, it’s Madeline.
The AI boom once promised to be a boon for chip startups, but that’s not how it played out last year. US chip startups just raised $881 million in the first nine months of 2023, down from $1.79 billion over the same period the previous year.
But a flurry of funding rounds for VC-backed chip companies in recent weeks suggests fresh investor enthusiasm. An Israeli startup, Hailo, announced an additional $120 million in funding to challenge NVIDIA from the “edge”— that is, putting chips in devices that don’t have continual access to the cloud, such as robots, cars, and smart cameras. Hailo says its chips can run AI workloads on such devices while using less memory and much less power—and for much less money—than Nvidia AI processors running in data centers.
SiMa.ai, which announced $70 million in new funding led by Maverick Capital, is similarly looking to the edge with chips that can speed up AI performance in devices including cameras and cars.
MatX is focused solely on LLM functions—as opposed to the many tasks performed by Nvidia GPUs—and hopes to compete directly for Nvidia’s AI customers. The team raised $25 million at the end of March from Nat Friedman and Daniel Gross, Bloomberg reported. Recogni, which raised $102 million in February, is also in the mix for AI accelerator chips.
And chipmaker Cerebras Systems is planning to go public in the second half of this year at the earliest, Bloomberg reported.
US Venture funding for semiconductor companies skyrocketed in the boom times like everything else, reaching $2.4 billion in 2021, up from just $700 million in 2019, Crunchbase News reported, before falling back to earth in 2023.
Chips can be an especially rough business in the best of times, with long and costly development timelines and notoriously cyclical demand. When it comes to AI chips, startups also have to confront the specter of Nvidia, which even beyond the chips themselves has built a software and component ecosystem that gives it a huge advantage in building successive generations of market-leading products.
“Nvidia has done a great job in execution, not just on the chip, but also the software and model adoption,’ said Rajiv Khemani, the CEO of Auradine and a prolific AI startup investor who put money into MatX. “I think the only way new chip startups get funded and can deliver 5-to-10x value is by focusing on a specific application and segment, rather than competing broadly against a 10,000 pound gorilla like Nvidia.”
David MacQueen, the editor-in-chief of the Microprocessor Report, said edge computing technologies could eventually be an even bigger market than GPUs and CPUs for data centers.
“Nvidia hardware might support those devices by providing some cloud compute capabilities, but inferencing at the edge is critical for many practical uses of AI,” MacQueen said. That sums up both Hailo and SiMa.ai’s strategies right now.
According to MacQueen, there are also plenty of sectors where a hardware startup could build a unique software component for a specific industry—say, automotive, or healthcare, or smart city infrastructure.
“You don’t just plug in a rack from Nvidia and off you go; that needs a supporting cast, from advanced dedicated AI accelerator chips to network switching and memory, which today are becoming higher and higher performance devices to handle the huge volumes of data needed for generative AI models.”
This “supporting cast” on the hardware side can be big business, as Astera Labs’ IPO last month showed. The company makes connectivity hardware for data centers—tools that AI companies rely on to move around their massive amounts of data. Its shares have fallen modestly in recent days but are still trading at nearly twice the March offering price, giving the company a market cap north of $17 billion.
Another connectivity startup, Celestial AI, which specializes in optical components, recently closed a $175 million Series C.
Not everyone has given up on taking on Nvidia straight-up. We wrote about Sam Altman’s announcement that he was seeking “trillions of dollars” for a new chip venture back in February, and Microsoft, Meta, and Amazon are all developing their own chips. AMD is also making a run. And Groq’s “LPUs,” or “language processing units,” are designed specifically to deal with sequences of data, such as code or natural language, to deliver LLM output faster than GPUs, which were initially designed for graphics processing.
But for scrappy chip founders in the age of big AI, the best opportunity may be to go small.
Eric’s Take
All-In Podcast Defends Trump’s Truth Social
I was embarrassed for Silicon Valley, listening to Chamath Palihapitiya and David Sacks bend over backwards to defend Donald Trump’s Truth Social on their All-In Podcast.
I had to double-check that it wasn’t an April Fools episode.
The All-In crew have become experts at contorted intellectual exercises that might sound clever but break apart on any serious analysis.
“I think this is a really important watershed moment and it reminds me of 1997,” Palihapitiya said about Donald Trump’s social media company, which lost $58.2 million on $4.1 million in revenue last year. Palihapitiya compared Truth Social to David Bowie’s successful experiment where he sold bonds, backed by his royalties, to investors.
“What’s happening is a more sophisticated version of that movement, but at much larger scale,” Palihapitiya said. He compared the company to a “trading card” and said the stock was trading on Donald Trump’s enterprise value.
Of course, as Palihapitiya finally acknowledges toward the end of the segment, Truth Social is not backed by Trump’s personal royalties. It’s a tiny social media company, powered by Trump’s posts, and bears no resemblance at all to a legendary music catalog beloved by millions.
The crew gave the friendliest characterization of a meme-stock you could imagine, invoking GameStop several times. Forget that GameStop’s stock has been on a steady march downward since the mania. Yes, the market can be driven by hype and frenzy in the short-term, but eventually stocks tend to reflect a company’s actual financial performance.
Unless, that is, the stock isn’t about the company at all. Silly you if you had any hope that the All-In crew would be troubled by the fact that they’re celebrating the likely Republican presidential nominee’s end-run around campaign finance rules.
The top commenter in the All-In subreddit sums the segment up aptly:
The Trump meme stock segment perfectly sums up this pod now.
A billionaire TikTok investor visits Trump and then helps him prop up the valuation of Truth Social something like 200x what it would be valued if it were any other company and not one of these guys can call it out.
They all just justify it like it’s a normal thing and not some giant grift and corruption happening out in public.
Host Jason Calacanis was, at least, direct in stating the obvious, that Truth Social was extremely overvalued. “You’ll lose a large amount of money, I predict, if you buy this.”