The biggest problem facing AI chip startups

It is reported that in recent years, as another major force to fill the gap in machine learning computing power, AI chip startups have launched new products in an effort to fill the hole faster. In the five years of rapid development of AI, it has attracted a lot of investment, but to many people's confusion, except AIoT, these startups have made a bit of waves in the autonomous driving market, other AI because of their emergence, the market landscape seems to have not changed at all, even they themselves are in crisis.

Product and talent issues

One of the biggest issues facing AI chip startups is product availability. Unlike big companies that can roll out new products every year, these startups launch at a much slower pace. Due to expensive design traffic, long cycles, etc., many companies have even moved away from stealth at this stage, with the product yet to be unveiled. The official websites of these companies are often very simple, such as a series of descriptions of their technology, a conceptual rendering of the research chip BQ24007PWP, and so on, with little information available to the outside world.

Even if the product is announced, it's still a long way from mass production and out of the box. That's why industry leaders were speaking before. "Although there are many AI startups building AI chips, there has been no usable product." And it's not just AI startups that want to design their own chips that are plagued by this problem. It's also the chips of AI Internet makers.

Secondly, the "super" situation in the AI hardware industry is really pressing some AI chip companies out of breath. Nvidia's GPU occupies an absolute dominant position. For the hardware and software required by the AI industry, Nvidia has done the extreme. With the strength of AI chip start-ups, it's hard to make a dent.

Not to mention commercial AI, even in AI research, Nvidia GPU is also the absolute king. In ZetaAlpha's 2022 statistics, the number of GPU mentions in the paper by AI Nvidia reached 21,452 times, far more than any other. ASIC, FPGA hardware, its V100 also became the most popular model, and even consumer Gpus like RTX3090 were mentioned 3,663 times.

Another major reason for the lack of product competitiveness is talent. On the surface, the lineup looks good, especially the founders. A veteran of the SoC industry, the great god of machine learning algorithms, is slightly outnumbered in numbers and overall strength. These bosses need the cooperation of design engineers at all stages, and then the architecture is set up and optimized by software engineers to create a perfect one. AI chip. Obviously, talent is definitely a pain point on their list of jobs from these startups.

Financing and acquisition issues

VCS targeted by semiconductor start-ups have seen a significant drop in funding this year, with PitchBook statistics showing a total of just $7.8 billion, down 46% from last year. In the face of this challenge, however, capital markets have not lost each other. AI's preference for chips, even if you don't use the word AI in your product description, decreases the likelihood of successful financing.

There have also been A number of recent rounds of funding for AI chip start-ups, such as the Israeli company that just closed a $35 million Series A round. NeuReality, the Dutch enterprise completed the $27 million Series A financing Axelera, while completed the $67 million Series B1 financing SiMa.ai and so on.

However, the initial investment is only enough to sustain these start-ups for a short time. In the R&D semiconductor industry, if there is no stable investment before the product is sold, it will soon fall into a desperate situation. Mythic, a US analog AI chipmaker, for example, is in trouble for a cash crunch.

In the relatively fragmented AI market, it is common for big players to acquire chips. A few years ago, Intel paid a lot of money for HabanaLabs and so on. That said, big factories have been buying fewer AI startups in the past two years, preferring to invest first rather than start directly. So, at this stage, unless the product improves or the technology is particularly suited to these big factories (especially in the server space), it won't be easy for them to "sell" themselves successfully.

Summary

Could be even harder for AI chip startups than the current economic downturn. Both cars and servers seem to be showing some signs of decline and order cutbacks. If you can't get through this period, the AI hardware market is bound to see another shakeout.