Digits to Dollars China Electronics
This is a digest of highlights from news we found interesting about developments in China's semiconductor industry and electronics complex.
Interesting News
China’s BIG IC Fund has new management and we are starting to see signs of how they are altering their investment strategy. This is an even better post, but it is in Chinese and Translate apps struggle with a lot of the language. Right now they seem to be selling stakes in public companies, largely on the assumption that their capital is better deployed on earlier stage companies. The key question will be how the fund divides its investment between privately-held start-ups and new initiatives within State-Owned Enterprises (SOEs). And how much of the fund will be diverted to jobs programs that have limited commercial futures, like YMTC. The answer to this is unclear, but we do think it is fairly likely that the Fund will shift even more upstream, away from fabless design towards tools for wafer fabrication.
Rhodium published a white paper on China’s coping mechanisms post the US sanctions. A realistic look at China’s vulnerabilities, but also a lot of good data on where they are strong. Not least an immense amount of foundry capacity operating at the trailing edge.
There is growing focus on how Chinese firms are seeking to circumvent US sanctions. From stories of travelers being arrested at customs with suitcases full of GPUs to elaborate shell company structures. When the sanctions first came down we wrote that this would become a perpetual game of whac-a-mole, enough to supply small projects but not enough to support an entire industry.
Huawei seems to be intent on single-handedly reproducing the entire wafer fabrication supply chain. A few months back they announced they had developed an EUV light source. Now they are rolling out their own EDA tools. If anyone can pull this off, Huawei can.
A wide-ranging look at the recent restructuring of the Ministry of Science and Technology. Parsing official policy documents is very hard for outsiders, making this a useful decoder ring. This contains a big focus on ways to improve manufacturing and increase basic research, and a handy list of priority areas including new energy, new materials, biotech, energy conservation AI and equipment manufacture. Also interesting to see comments on where China is weak - including developing engines, digital machine tools and (of course) chips.
It is important to bear in mind that Chinese companies are not some fully integrated monolith operating under a centralized plan. There is as much rivalry and competition among Chinese companies as anywhere else (if not more). Case in point, car maker AION, the electric vehicle brand of Guangzhou Automobile, has just scaled back its cooperation with Huawei. Reading between the lines, this looks like a car maker worried about falling under the dominance of its software provider and instead choosing to go it alone. The same dilemma US car makers are facing.
The Wall Street Journal ran a piece on China’s weaponization of M&A regulatory approvals. An important topic, that merits a lot more attention, albeit not exactly a new phenomenon.
For those interested in China’s EV industry we found China EV Hub on Substack, which does a great job of covering the industry.
An eye-opening chart showing growth of China’s Public Cloud market. We cannot vouch for the source of this data, but it is in line with what we have heard elsewhere. The telecom operators, especially China Mobile, are outpacing their competitors. This is one of those noticeable divergences from the model elsewhere, and a deliberate policy choice to prioritize the operators.
IP problems continue to plague China’s handset makers. This time Vivo lost a lawsuit with Nokia in Germany.
Tencent opened a new AI Training and high performance computer cluster. This is powered by Nvidia’s H800 GPUs, a version of the H100 tweaked to squeak just below the US restrictions. The size of the cluster is not entirely clear, but it looks to be about 16,000 GPUs, which is big for GPT, but small for many other applications. Tencent is something of an outliner among the hyperscalers - they do not really run a public cloud like AWS or Ali Cloud. Their workloads look a lot more like Google’s or Facebook’s, but they are also working on a whole host of other projects such as autonomous vehicles.
One area where China has a clear technical lead is energy. They have a very sophisticated national electricity grid and now they have an “artificial sun”, a tokamak fusion device.
A look at China’s semis market with a focus on memory (in Chinese). We do not agree with all (a lot) of this, but there are still some categories where China can compete in memory.
Venture Investment Activity
Auto maker BYD invests in Kunlunxin an automotive AI chip designer.
VCs in China are moving away from GPU investments. Not surprising given the fate of Biren.
Baidu invests $147 million in RISC V CPU start-up StarFive. It takes a lot of money to build a CPU. This is both an important milestone for the RISC V project, but also a warning sign to CPU makers elsewhere that there are real competitors in China.
Analog front-end designer SenSilicon raised a $17 million Series A. Maybe this company has more under the hood than its public profile lists (there are hints of some digital designs coming), but that is a lot of money for this market. We know a comparable company in the US that would be ecstatic with 10% of that, and they are already in revenue.
China and GPT, so far
Like much of the world, people in China’s tech industry are trying to understand what GPT means for them. And like much of the world, no one really knows yet.
Our take is that GPT is too new for anyone to really know what it will ultimately be capable of. And so we are seeing a vast wave of experimentation and analysis. It is still very early days, but there are a few themes that seem to be perking up.
Not surprisingly there are now countless projects underway in China’s software community to do something with GPT. The big Chinese Internet companies seem to have been taken off guard by Chat GPT’s sudden success and momentum. But then again, so were Google and Amazon. Now they are wasting no time in launching their own GPT deployments. Alibaba, Baidu and JD have all rolled out their own generative chat apps. On launch, most of the focus in the Western media was how these tools struggled to comply with China’s censors. That has largely faded away as everyone seems to be struggling with similar problems, and seem to have installed a few filters to limit the most egregious missteps.
There are also many efforts underway from smaller entities. These range from individuals like mathematician Terrance Tao incorporating GPT into his academic work to a Meituan founder, Wang Huiwen, investing $50 million to build a Chinese Chat GPT. There is a lot of activity going on here and a lot of interesting ideas.
We did a fair amount of searching in WeChat for GPT commentary. This turned up a lot of the stories we would expect including much of what we detailed above. Given our filter settings there was, not surprisingly, a fair amount of thought dedicated to what GPT means for the semiconductor industry. This included thoughtful looks such as this one, which provides a fairly good explainer for GPT silicon needs, even in translation. However, there were a lot more posts like this one which are looking for GPT related investments. This one is even more direct, it details stock price movements from the launch of GPT through each Chinese semi company’s GPT-related announcements. There is not a lot of nuance in this, and its hard to tell, but definitely strikes us as fairly cynical.
To be honest, at this stage, China Tech’s stance on GPT looks an awful lot like what US Tech is experiencing. Lots of enthusiasm for something new, but if we are honest, a lot of people thinking (hoping) Chat GPT is the Next Big Thing, the new investment frontier.
Of course, China has to engage in this process behind the very real concerns as to what will US sanctions will mean for it. As we detailed a few weeks back, the semis industry there is still very much undergoing the seven stages of grieving, with a lot of Denial on display online. So far, much of the view of GPT seems to be that China’s companies will be able to experiment freely using someone’s cloud. We suspect that over time the very real limits posed by the US sanctions will temper a lot of this enthusiasm. So far, we have not seen anyone online warn that China will fall behind the US in GPT, or AI more broadly, but there is a very real possibility that China’s GPT efforts will eventually get hobbled. The one thing we are comfortable predicting is that there is going to soon be a lot of effort in China working on GPT systems that can be run on end devices and not reliant on the Cloud.