SUNMMIO, which focuses on the R&D and mass production of 3D architecture AI cloud high-computing-power chips, recently announced the completion of its Pre-A2 round (including its additional round) and Series A funding. The raised funds have three clear destinations: expanding top-tier R&D talent, increasing investment in core technology iteration, and accelerating the mass production and deployment of its self-developed 3D TokenPU large model computing power chip. All three point to the same goal—transforming technological advantages into production line shipping capacity as quickly as possible.
| Why "3D TokenPU"
The large model inference market is rapidly scaling up. The hype around training has been discussed for a long time, but what truly drives continuous computing power consumption is inference—every invocation and every token involves substantial computing power expenditure. The essence of inference is a competition in energy efficiency.
The bottleneck lies in data transmission. As model parameters grow larger, the transmission speed between computing chips and data increasingly fails to keep up with the computing speed. Chips spend a significant amount of time "waiting for data," which is commonly referred to in the industry as the "memory wall." The traditional approach is to continue reducing power consumption and increasing frequency on advanced process nodes. However, this path involves high costs and diminishing marginal returns, and is constrained by advanced process node resources, making it particularly unfriendly to domestic manufacturers.
SUNMMIO chose not to compete in the general-purpose GPU market, but to redesign from the architectural level. Its pioneering 3D TokenPU architecture focuses on extreme energy efficiency optimization for large model inference scenarios, concentrating computing resources on token processing—the true bottleneck—rather than distributing efforts evenly.
In June this year, the first-generation product A4E was officially taped out. It adopts a relatively mature domestic process node and supply chain system. Through 3D hybrid bonding technology, it vertically stacks 8 layers of memory wafers on top of the logic wafer, achieving an ultra-large memory access bandwidth of 16TB/s. In simple terms, it uses 3D stacking to directly "open a high-speed channel" for data, bypassing the limitations of the memory wall at the physical structure level. This is the key differentiator from conventional solutions and the foundation for realizing the combination of "high bandwidth, high energy efficiency, and high cost-effectiveness."
It does not rely on the most advanced process nodes, but trades architectural innovation and advanced packaging processes for performance headroom. Under current supply chain conditions, this represents a more realistic and autonomous direction for breakthroughs.
Dr. Wang Fuquan, Founder and CEO, clearly defined the company's positioning: "From the very first day of its establishment, SUNMMIO chose to use the native 3D TokenPU architecture of the large model era, bypassing the traditional path of general-purpose GPUs and focusing on the extreme energy efficiency of token processing." Regarding his judgment on the future, he stated the goal of accelerating mass production and delivery, joining hands with supply chain and ecosystem partners to make AI computing power "a universal infrastructure like water, electricity, and gas." In the current context of soaring computing power costs, this statement points to the same question: Can inference become cheaper?
| 7 Years of 3D IC Accumulation
Architectural design is only the first step. The real challenge for 3D stacked chips lies in engineering—whether the design can be implemented, whether wafers can be bonded, whether the yield can be sustained, and whether mass shipping is achievable. If any link in this chain fails, the preceding technological advantages will be reduced to zero.
SUNMMIO's confidence stems from the team's track record. Core members have been deeply engaged in the 3D IC field for nearly 7 years. During this period, they joined hands with supply chain partners to pioneer the Wafer-on-Wafer hybrid bonding 3D IC supply chain in mainland China, accumulating mass production experience of over 10,000 3D IC wafers and generating a cumulative revenue of over CNY 1.2 billion. This means the team has not only done the design but also completed the entire chain from wafer bonding to mass shipping.
In the chip industry, having complete experience from design to mass production is often scarcer than a single technical metric. Achieving a good architectural metric might rely on a single-point breakthrough, but running through the entire process from design and bonding to mass shipping tests the control over the entire supply chain. Especially for processes like hybrid bonding, the thresholds for yield and consistency are very high, and mass production experience itself is a barrier. It determines whether the company can turn the architecture in the laboratory into a product that can be stably delivered.
| Who is Investing in SUNMMIO
This Pre-A2 round and Series A were led by state-owned capital platforms at the Beijing municipal and district levels, such as Yizhuang State-owned Capital Investment. Over the past year, SUNMMIO has completed multiple rounds of financing. Investors cover state-owned platforms such as CDB Capital, Beijing Industrial Investment, Beijing Shunxi, and Zhongguancun Science City, market-oriented institutions such as Source Code Capital, Xianghe Capital, and Jiayu Capital, as well as industrial funds such as Shexi Capital, Lenovo Capital, and Zhiwei Capital, with some investors continuously increasing their stakes.
This list has three layers of meaning. The entry of state-owned platforms brings both financial and policy support, as well as considerations for regional industrial layout; market-oriented institutions value the commercial prospects of the technological route; and industrial funds focus more on its positioning in the supply chain. The presence of three types of capital simultaneously, along with additional investments from some existing shareholders, indicates that all parties see the value of this company from their respective perspectives.
It is worth mentioning the aspect of local industrial synergy. Chen Zhicheng, Chairman of Yizhuang State-owned Capital Investment, stated that SUNMMIO's 3D TokenPU high-computing-power chip is highly aligned with the direction of the Beijing Economic-Technological Development Area in building an AI and IC industry cluster. He expressed the expectation to use this investment as a link to promote synergy between SUNMMIO and upstream and downstream enterprises in the area, such as manufacturing, advanced packaging, and large model applications, creating a full-chain ecosystem from chip design to scenario deployment.
For a chip company currently in the ramp-up period of mass production, the value of such resource docking from the local industrial cluster is sometimes no less than the financing amount itself—partners in manufacturing, advanced packaging, and applications are right in the area, making synergy costs much lower.
| Mass Production is the Key
There are several points worth continuous observation.
First is the weight of the route. 3D stacking, focusing on inference, and binding to the domestic supply chain represent a differentiated but narrower path. If it succeeds, the barriers will be very high; if it fails, the room for error correction is limited. The financing of nearly CNY 2 billion is a vote of confidence from capital in this route.
Second is the realization of mass production. The A4E has been taped out, but from tape-out to scaled delivery, there are hurdles such as yield, advanced packaging, and supply chain integration. The team's 7 years of experience and over 10,000 wafers are a plus, but the market ultimately looks at delivery results—the speed of customer onboarding, the scale of shipments, and product stability are what truly determine success or failure.
Third is the time window. The inference market is exploding, but competition during the window period is equally intense—overseas, there is NVIDIA's mature product line, and domestically, a group of GPU and ASIC manufacturers are competing in the same arena, all vying for the incremental growth in inference. Whether SUNMMIO can complete scaled mass production and customer onboarding before the window closes is a tougher test than financing.
Money solves possibilities, but mass production and deployment determine success or failure. The answer sheet SUNMMIO needs to submit next is not in the financing announcement, but on the production line.
The content of this article is compiled based on public information, striving for objectivity, but does not guarantee the completeness and accuracy of the information. The views in the article do not constitute investment or decision-making advice, and those who act on them do so at their own risk.