Large models regardless of the parameters frommillionslevel totrillionsThe evolution of the level, or fromUnimodaltoMultimodalcontinues to drive the demand for AI computing power.
Recently, all concepts linked to computing power have not been low, such asOptical transceiversRelated companies such as Zhongji Innolight, Tianfu Communication, Optical Library Technology, Xinyisheng, etc.;AI serversCompany Inspur Information, Sugon, etc.
According to IDC**, the scale of China's intelligent computing power will reach 1271 by 20264eflops, CAGR 2022-2026 will:Up to 528%In the next 5 years, there will be a massive demand for computing power.
We know that computing power is actually the computing power of chips, and AI chips are directly related to AI computing power, but at presentMore than 90%.market share in foreign manufacturersNvidia, AMDIn the hand, in the context of domestic substitution, domestic AI chip manufacturers are commendable.
Cambrian is seen as "".The first share of AI chipsCompared with NVIDIA, the market value once exceeded 100 billion, and now even if it is almost "cut in half", there is still 60 billionBut where is it now?
First, it has lost 48 in the past 7 years since its establishment7.6 billion, revenue is still declining.
Cambrian's revenue has increased by 93% in five years from 2017 to 202246 times,But the losses continued to widenThe growth rate of revenue is far less than the decline in non-net profit.
In 2022Be included in the Entity ListIt has a great impact on the ** chain, and the revenue growth rate has also decreased significantly, and the revenue in the first three quarters of 2023 will even show negative growth, down 44% year-on-year84%, explainedIn the short term, Cambrian has not benefited from the explosion of demand for AI computing power and domestic substitution
The company's continuous losses, in addition to the lack of revenue growth, are partly due toThe period fee is too high
Cambrian continues to increase R&D investment, and the management expenses of paying equity incentives are also more than 200 million yuan per year on average, and in November 2023, Cambrian will release an equity incentive plan, which will last until 2026 and will be given to the companyIt is not a small burden
Of course, performance losses are not a big problem, and we mainly focus on the products, competitors and future growth of technology companies.
The product update speed is slow, and the performance is far behind NVIDIA
Cambrian's main products include:Cloud smart chips and acceleration cards, edge smart chips and acceleration cards, terminal processor IP and supporting software development platform.
1.Cloud-based smart chips
Cloud smart chips and accelerator cards are mainly used in cloud servers, data centers and other fields to provide computing power for cloud computing and data centers, and are also Cambrian's main income**.
Artificial intelligence is like the "brain" of a person, which is divided into the process of learning Xi and judgment, used for ".Xi"The process of the chip isTraining cardsJudgment"The process of the chip isInference cards
Obviously, the training card is more important than the inference card, whileTraining cards happen to be scarce in the country.
The Siyuan 290 is Cambrian's first training card, but the Siyuan 290 is in line with the Nvidia H100Computing power, memory, and memory bandwidthand many other aspectsThere are considerable gaps, and after its launch in 2020, in recent yearsNo new training chips are coming to market
Siyuan 270 and Siyuan 370 are both inference chips, Siyuan 370 has a great improvement on the basis of 270, compared with Nvidia T4, it has a great advantage in codec capabilities and memory, butNvidia L4 has three times the computing power of T4, so the Siyuan 370 is compared to the L4The gap in computing power is still large
Edge smart chips
Edge intelligent chips provide computing power to devices between terminals and the cloud, which can effectively make up for the lack of computing power of terminal devices.
Siyuan 220 is an edge chip launched by Cambrian in 2019, and now Siyuan 220 has reached the late stage of its life cycleBut the new product has not yet been launched
To develop new products, huge R&D investment is indispensable, and Cambrian's R&D expenses will grow at a compound annual growth rate from 2017 to 2022Up to 11933%，But in the first half of 2023 it wasR&D expenses have been reduced, R&D personnel alsoReduced by more than 200 peopleIt remains to be seen whether technological breakthroughs will be made in a short period of time in the future.
Competitors continue to emerge, and it is difficult to build an ecosystem
AI chips can be roughly divided into four categories: GPU, FPGA, ASIC, and brain-like chips according to the technical architectureAccounting for more than 80%.The Cambrian is biased towards ASICs.
Among enterprise-class GPU chips, NvidiaThe proportion is as high as 914%It is the absolute leader, the share of GPU chips of domestic manufacturers0 only1%and in the entire AI chip marketLess than 10%.
The reason why NVIDIA's position is strong is that the chip performance is leading on the one handThe biggest core competitivenessIt lies in the computing and programming platform (CUDA) accumulated over the years, as well as the software ecosystem.
Many algorithm engineers have become Xi to NVIDIA's programming language, and software and hardware manufacturers have also been deeply bound to NVIDIA, if you want to replace it with other AI chipsSwitching costs are high, some manufacturers are not willing to switch at all.
In this case, Cambrian wants to build its own ecology willFaced with many difficulties, not something that can be done in a short period of time, and there are still in the countryOther chip manufacturersKeeps coming.
The biggest competitor in the country isHuawei HiSilicon's Ascend chipAscend is close to NVIDIA in terms of performance, and it can also leverage Huawei's brand influence and sales channels, so it is relatively easy to build an ecosystem.
In addition to that,Jing JiaweiIt is a leading domestic GPU company, and its products can also be used in data centers and cloud AI computing powerThere is a threat of substitution for Cambrian products
Kunlun core, Haiguang informationAll are strong competitors of Cambrian, and some have only been established in recent yearsBiqian Technology, Suiyuan TechnologyIt should not be underestimated, their AI chips are also continuing to narrow the gap with foreign countries, and their computing power performance is outstanding.
Bili is the chip model of Biqian Technology).
Therefore, in the environment where Nvidia is on the outside and a number of domestic manufacturers are emerging inside, it is still difficult for Cambrian to turn losses into profits in a short period of time.
Finally, to sum up.
Cambrian is the leading AI chip leader in China, but the performance continues to lose, R&D expenses have also declined, the product level and foreign giants have a large gap, and many domestic manufacturers are eyeing the company, although the valuation has dropped by half, but it is still necessary to continue to pay attention to the company's R&D situation.
The above is only for the analysis of listed companies and does not constitute specific investment advice.
Follow Feijing Investment Research (ID: feijingtouyan) and receive the most valuable "Growth 20: 20 Companies with the Most Growth Potential in 2023" investment report now!
**: Flying Whale Investment Research