SigmaStar has been engaged in edge AI for 20 years.

The term "edge-side" has been plastered everywhere this year.

01

It appears on smart glasses, robots and vehicles. To put it simply, it means data is not transmitted to the cloud, and the device completes all calculations locally on the spot.

Sounds very new, right? The first type of device that applied this technology on a large scale was the surveillance camera at the entrance of your residential compound.

The work of a camera is divided into two steps. The first step is to convert light into a clear, recognizable image; the second step is to identify on the spot inside the chip whether the content in the image is a person, a vehicle, or rising smoke.

The first step relies on ISP, the image processing technology, and the second step relies on NPU, the computing unit dedicated to recognition tasks.

Have you noticed? The three core assessment criteria for edge-side AI today — operating normally even when disconnected from the network, low power consumption, and low per-unit cost — were already the challenges the security industry started to tackle 20 years ago.

One of the players that performed best in this field is SigmaStar Technology.

According to the statistical caliber of Frost & Sullivan, in terms of shipment volume in 2024, it is the world's largest visual AI SoC supplier with a 26.7% market share. If we only look at the security segment, its market share reaches 41.2%, ranking first globally.

What is SoC? It is a chip that integrates the entire "brain" of a complete device.

With such a large scale, by the end of 2025, the cumulative shipment of chips with AI computing power has exceeded 550 million units. What does this concept mean? A considerable number of "vision-enabled" machines in the global market are equipped with its visual processing chips.

The company was registered in Xiamen only at the end of 2017, but its core team has a far longer history than the company itself.

Its predecessor was the security chip team of Mstar Semiconductor. MediaTek began to acquire Mstar back in 2012, and completed the full acquisition step by step in early 2019.

The team is led by Lin Yungyu, who served as the general manager of Mstar's Shenzhen office for 15 years starting from 2003, and concurrently took the position of COO of Mstar Semiconductor from 2015.

He has been in charge of the entire surveillance chip business from its very beginning. Before his former employer was fully acquired, he led the team to set up a new independent entity in mainland China to continue focusing on the same track.

Some people say this company is just a "shadow stock" of MediaTek, inheriting core technologies from its "parent company".

This account is worth sorting out clearly.

During the regulatory inquiry in August 2022, among the 51 authorized patents corresponding to its core technologies, 24 were inherited from the MediaTek ecosystem, accounting for 47.06%. Nearly half of the core patents were provided by its former employer.

By its listing in March 2024, it had 203 authorized patents in total, plus 223 patents under application; the company did not disclose the exact ratio of independently developed and inherited patents among the newly added ones under the same statistical caliber, but its patent portfolio has visibly expanded significantly.

The equity structure is another separate account: MediaTek remains its largest shareholder to date, holding a 28.74% stake, and has explicitly stated in official documents that it will not seek control of the company, so the company has no actual controller.

As you can see, the two accounts are recorded separately: on the shareholder register, it is associated with MediaTek, but on the patent certificates, more and more of its intellectual property is independently owned by itself.

There is another unavoidable part of its development history: around 2020, the player with the largest market share in this industry shrank its business due to sanctions, leaving vacant market positions that several remaining players filled, and SigmaStar was one of them.

The company itself defines this as "strategic transformation", shifting its positioning from a visual SoC supplier to an integrated edge-side high-computing-power solution provider.

What exactly does this transformation refer to? It refers to the expansion of its customer list. Twenty years ago, the only customers sitting across the negotiation table were security manufacturers, but now robot manufacturers, automotive companies and smart glasses makers have also joined its customer base.

Its core technology has not changed, but the types of buyers coming to source its products have changed. Edge-side AI did not turn it into a new company; instead, new buyers have found this mature, established manufacturer. However, behind the prosperity of this established manufacturer, there is a long-standing issue: its quotation has never had enough bargaining power for its long-time old buyers over the past 20 years.

02

In the first half of this year, its quotations suddenly became much firmer. How significant is this change?

Its performance forecast released on July 18 shows that the attributable net profit ranges from 820 million yuan to 900 million yuan, a year-on-year increase of 583% to 650%.

For the whole of last year, this figure was only 308 million yuan. In just the first half of this year, it earned more than twice its total profit of the whole last year. The 6x growth rate is real, but this number is the superposition of four different factors.

Let's use a simple analogy:

If your friend's salary doubled this month, don't rush to envy him, you have to check his payslip first: last month he took a long leave and had half a month's salary deducted; this month he included his part-time income into the same salary card; the company also issued a one-off project subsidy; and finally, his basic salary only increased by 300 yuan.

All of these are income, but their natures are completely different: the leave deduction was an event of last month, the part-time income is a newly added consolidated item, the subsidy will stop once the project ends, and the only part that can be sustained into next year is the basic salary. SigmaStar's 6x profit growth is exactly such a payslip.

The first factor is the low base effect. I checked the financial data of the same period last year.

In the second quarter of 2025, the attributable net profit was 69 million yuan, which even dropped by 12.62% year on year. That was a period when security demand was sluggish and new business lines had not yet taken off. This year's 6x growth is calculated based on this low trough, and the lower the base, the more impressive the growth multiple looks.

This low trough did not appear for no reason. Looking further back, in 2021, the company reached its peak performance, riding the wave of chip shortage and price hike in the security industry. Then its performance receded for two consecutive years, and the attributable net profit dropped to 205 million yuan in 2023, a year-on-year decline of more than 60%.

In other words, the company has already experienced a full cycle of price surges and declines. That is why this 6x growth needs to be analyzed carefully instead of being taken at face value.

The second factor is that the financial statements now include the operating data of an additional company.

At the end of October 2025, SigmaStar acquired a 53.3087% stake in Furun Microelectronics, a Bluetooth chip developer, and consolidated its financial results from November onwards. Furun's performance is not included in the base figure of the first half of last year, but is fully included in the first half of this year's statements.

How significant is the contribution of this consolidated part?

There is a reference data in the acquisition announcement: Furun's revenue in the first half of 2025 was about 58.75 million yuan, and its performance commitment states that its cumulative net profit from 2026 to 2028 will not be less than 100 million yuan.

Its annual revenue of over 50 million yuan is just a tiny fraction of SigmaStar's main business, so this factor exists but contributes very little to the total growth.

The third factor is the successful pass-through of price hikes. Due to the shortage and price rise of upstream storage chips, SigmaStar passed the increased cost to downstream customers completely without absorbing any part of it itself.

What is the direct trace of this in the financial statements?

Gross profit margin. The full-year gross profit margin in 2025 was 34.16%, while the single-quarter figure for the first quarter of this year reached 46.05%. Note that the former is the full-year average, and the latter is a single-quarter figure, which rose by nearly 12 percentage points within one quarter.

There is another more intuitive comparison in the same performance forecast:

The revenue in the first half of the year ranges from 2.62 billion yuan to 2.72 billion yuan, a year-on-year increase of 87% to 94%. Revenue has not doubled, but net profit has increased sixfold, and the huge gap between the two is driven by the sharp rise in gross profit margin.

The fourth factor is the actual increase in sales volume. The exact wording in the first quarterly report is: "The shipment volume of all product lines maintained steady year-on-year growth."

This factor is the most tangible, but also the most ambiguous. It is tangible because shipment volume does not lie. It is ambiguous because there is no detailed breakdown showing which product line sold what and how much.

Interestingly, there are already two completely opposite opinions in the market.

One side argues that this round of growth is entirely driven by price hikes, and the company's true performance will be exposed once the industry tide recedes. The other side calculates conversely: the storage price was already at a high level in the first quarter, and there was no new round of sharp price rise in the second quarter, but net profit still maintained strong sequential growth, which indicates that sales volume and business structure optimization are the main drivers.

Both sides have their own rational points, but neither can provide segmented supporting data. The standard answer to this question is held by the company itself.

Okay, after breaking down the four layers, the only conclusion we can draw is: the low base is a temporary factor, the consolidated revenue is a one-time event, and the price hike is a benefit borrowed from the industry cycle. The only factor that can be sustained in the long run is the actual sales volume.

Can the sales volume continue to grow? It depends on those new customers who have just started sourcing from it: who are they? What products do they buy? When will their orders reach full scale?

03

Starting from the largest new customer by purchase volume, I checked the shipment data in the 2025 annual report: the annual shipment of robot chips exceeded 10 million units. The shipment of the entire smart IoT segment exceeded 41 million units. The shipment of automotive chips exceeded 13 million units.

Compare these figures with its traditional main business: the annual shipment of security chips exceeds 120 million units, and the total volume from all new customers is still less than half of that from its old security clients.

The most promising segment in the new customer list is the robotics industry.

Still according to Frost & Sullivan's statistical caliber, based on the shipment volume in the first half of 2025, SigmaStar ranks second globally in the robot vision chip market with a 23.0% market share. Cleaning robots, lawn mowing robots, pool cleaning robots, and companion chat robots all need a pair of "eyes" to operate normally.

Its products for new customers cover a full range of different computing power levels.

Chips with 1-2 TOPS of computing power are installed in cameras and small household robots. Its flagship SSR670 chip with 8 TOPS of computing power is deployed in edge hosts, which can run large language models completely locally.

The SAC8905 chip with 32 TOPS of computing power is designed for automotive applications, which has obtained designated orders from overseas automakers and is scheduled to enter mass production in 2027. Going further, the planned chip for robot "brains" will reach 128 TOPS of computing power.

What is TOPS? It is the horsepower meter of computing power. The larger the number, the more complex tasks the machine can process locally on the spot.

The development timeline is also moving forward: the SS901 main chip for automotive LiDAR was released at the end of 2025, and has entered small-scale mass production for vehicles in the second quarter of this year. The first generation of AI glasses chips has been installed in mass-produced smart glasses, and the second generation, as officially disclosed, will be taped out this year.

The company has also set a clear goal for its LiDAR business: to become the world's top automotive LiDAR chip supplier within three years. Note that this is a statement released by the company itself, not a third-party statistical conclusion.

The aforementioned acquisition of Furun complements its weakness in wireless connectivity, enabling its vision-enabled chips to also support stable wireless communication.

The product line is very rich, but we also need to look at the full picture of the data: the 13 million annual shipment of automotive chips is still a tiny fraction of that of security chips, and the shipment volume of smart glasses chips is even smaller. The cycle from designated order to mass production for automotive-grade products is measured in years.

Moreover, new customers have more than one supplier on their negotiation table: Ambarella occupies the high-end automotive market, and Rockchip is also a strong competitor in the robot vision segment. The mature technology of SigmaStar is only an admission ticket to the game.

By the way, let's look at the situation of its old clients. There is a figure in the annual report: the top five customers account for 88.16% of its total revenue. Nearly 90% of its products are first taken over by distributors.

In other words, its old clients are very concentrated, and SigmaStar is the weaker party during price negotiations.

The root cause why its quotations had no bargaining power for 20 years lies in this. There is also a highly debated issue outside the company: inventory.

The company's book inventory at the end of the first quarter increased by 30% compared with the beginning of the year, which is an exact public figure. What is uncertain is the next part: after the products leave the factory and enter the distributors' warehouses, how much inventory is held by distributors? This part is completely opaque to external observers.

The reason why this debate has no conclusion for so long is exactly that: 90% of its products first enter third-party warehouses.

Apart from its business performance, the capital market related to its shareholders was also very active in the first half of the year.

I checked its market capitalization data: on July 3, its market capitalization was 23.878 billion yuan. After the performance forecast was released, on July 20, its market capitalization reached 44.976 billion yuan. In less than half a month, its market value nearly doubled.

Before the market capitalization doubled, the company itself took an action earlier: in February this year, it completed a share repurchase of 1.7448 million shares, spending 119 million yuan, with the purchase price range from 63.78 yuan to 72 yuan per share.

This price line set by the company itself was far exceeded by the market five months later. The same company had two completely different valuations at the beginning and middle of this year.

There were several other related moves earlier.

In May 2025, two old shareholders holding more than 5% of the shares transferred part of their shares, and the exact number of shares sold is disclosed in the public announcement. In September 2025, the company submitted its listing application to the Hong Kong Stock Exchange, and whether and when it will be listed will be announced in subsequent