If Cerebras can grow beyond its initial benefactor, G42, this IPO will be a rocket ship. If not, …
It may be hard to believe, but there hasn’t been a single IPO for an AI hardware company. Period. Not One. Why? Because none of the startups have yet to achieve the escape velocity needed to stand alone again as Nvidia. That may be about to change. Cerebras has filed with the SEC to offer stock to the public; their IPO could take place later this year. Let’s dive in.
OK, just who is Cerebras, again?
If you follow me, you know that I have long been a fan of Cerebras Systems, their CEO Andrew Feldman who I have known for 12 years, and the Wafer Scale Engine (WSE) approach to AI computing. The idea is so straightforward and so compelling. Instead of cutting up a wafer into chips, packaging those chips with High Bandwidth Memory, installing them onto a board into a computer, and then recombining hundreds of the chips to talk to each other over an expensive network, why not just interconnect the chips resident on the wafer? Thats Wafer-Scale computing which I think will become a huge trend over the next 5 years. Oh, and they don’t need HBM because Cerebras using on-chip SRAM supplemented by a streaming weight server. Yeah, it’s tough to power and cool an entire wafer, and you have to deal with inevitable defects on the wafer, but the resulting math is just too compelling not to successful. That is, if you get the software right.
Why an IPO, and why now?
The timing could be good. Cerebras’ revenue grew more than 14-fold to $136.4 million in the first half of 2024, and the company’s net loss declined from $77.8 million to $66.6 million. That makes a fine backdrop to try to raise more money to accelerate growth. But instead of going back to (expensive) venture capital, co-founder and CEO Andrew Feldman thinks the public markets are ready for a generative AI offering, and he is going for it.
Customer concentration is a problem
While Nvidia has a similar issue, with four customers accounting for over 40% of revenue in the latest quarter, Cerebras’ customer concentration may be a matter of where they stand in their lifecycle. It takes a lot of selling and convincing to get a potential client to the table, especially if you are yet another startup with a deck of slides. This IPO could allow Cerebras to find more G42’s flying under the hyper-scaler’s radars.
Cerebras’s prospectus said G42 represented 87 percent of its revenue in the first half of the year. G42 has committed to buy $335 million worth of Cerebras’s stock by April of next year, which will give it a stake larger than 5 percent. G42 and its subsidiaries are already buying and building a 36-exaflop monster distributed supercomputer with Cerebras.
As an analyst, investors often ask me how Cerebras can grow, given that their biggest potential customer are already building their own AI chips, and of course are buying hundreds of thousands of Nvidia GPUs. I ask them if they ever even heard of G42 before they started buying Cerebras WSEs, and I assert that there are a dozen more G42’s being funded to build local AI factories as sovereign AI farms and 2nd tier public clouds. Perhaps they just need permission to buy.
Conclusions
While we will have to wait to see the market reception to Cerebras’ planned IPO, we will be willing to bet that it will be strong. Stock market investors are anxious to invest in an alternative to Nvidia and many consider AMD to be fully valued. The AI market is also anxious to have an alternative they can deploy for AI deployment, and Cerebras’ value proposition is extremely compelling for both training and inference.
If they can pull off a successful IPO, it could help them grow from a small base; many enterprises won’t buy from startups. And nothing breeds success like success.
Disclosures: This article expresses the author’s opinions and
should not be taken as advice to purchase from or invest in the companies mentioned. Cambrian-AI Research is fortunate to have many, if not most, semiconductor firms as our clients, including Blaize, BrainChip, Cadence Design, Cerebras, D-Matrix, Eliyan, Esperanto, GML, Groq, IBM, Intel, Micron, NVIDIA, Qualcomm Technologies, Si-Five, SiMa.ai, Synopsys, Ventana Microsystems, Tenstorrent and scores of investment clients. We have no investment positions in any of the companies mentioned in this article and do not plan to initiate any in the near future. For more information, please visit our website at https://cambrian-AI.com.
