Registered in Shenzhen in May 2026, by mid-September, Jiuwanli Future Technology completed its seed and angel rounds in just four months, with a cumulative amount exceeding 100 million USD. The investor list practically gathers the bellwethers of domestic semiconductor investment: Sequoia China, Lanchi Ventures, Lenovo Capital, Huaye Tiancheng, as well as professional funds like Huadeng High-Tech, Wuyuefeng Capital, and CAS Star, with Gaoqi Capital serving as the exclusive financial advisor.
The background of founder Chen Peng has been widely reported in the industry: nearly 20 years at HiSilicon, where he was the mastermind behind Kunpeng and Ascend; in 2023, he joined Horizon Robotics as President of the Chip Product Line, leading the rollout of the Journey 6 series, with the J6M achieving shipments of over one million units for two consecutive years. No need to elaborate further here.
What truly deserves readers to pause and reflect upon is the pricing logic behind this funding round. What justifies the valuation for a company with no public products and only four months of operation?
I believe that Chen Peng's own capability in mass production closed-loop is the most direct source of credit for this funding round. Kunpeng, Ascend, Journey 6M—the commonality of these experiences is not just "having made chips," but "pushing chips from tape-out to million-level mass production." In the chip industry, this capability is scarcer than any PPT. Capital first bets on the person, but a person's credit only forms the pricing basis when the industry direction is recognized. The two are joint conditions, not substitutes—the consensus on the direction explains the heat of the track, while the person's credit explains why it is this company and this price.
The funding news simultaneously released another signal: top-tier institutions are willing to pay a premium for the direction of "edge-side high-compute inference." Edge-side inference, which means running large models directly on smartphones, automobiles, robots, and PCs, is evolving from a concept into a competitive track. Capital bets on the person, but the direction chosen by the person happens to step right on the node of value realignment in the industrial chain.
| Old Rule: Every Time the Computing Platform Changes Hands, the Industrial Chain Gets Reshuffled
Let's start with a repeatedly verified historical rule: the migration of each generation of computing platforms rewrites the value distribution in the industrial chain.
In the PC era, profits concentrated in the Wintel alliance; in the smartphone era, value shifted to SoC platforms, camera modules, and OEM brands—the supply chains of Qualcomm and Apple each created a batch of hundred-billion-dollar companies. Behind this rule lies the fact: the "bottleneck segment" of a new platform is the high ground for new value.
In this round, where is the bottleneck? Industry data already states it plainly: the computing power of mobile NPUs almost doubles with each generation, while the growth rate of memory bandwidth falls far behind; under real large model workloads, edge-side chips spend most of their time waiting for data rather than computing. Translated into business language: at the edge AI poker table, the scarcest resource is not the computing power itself, but the entire supporting capability to "feed the computing power full."
Whoever solves the bottleneck captures the value.
The "bottleneck" itself has two sides: for the supply side, it is a value high ground; for the demand side, it may be cost pressure. This is particularly critical in the memory segment.
| Three Upstream Cards: Chips, Memory, and Packaging
The First Card: Inference Chips
The chip itself is the origin point of this map and inevitably the most fiercely contested square. In the mobile main chip camp, Qualcomm and MediaTek are integrating increasingly powerful AI computing power into SoCs; Apple pursues a fully self-developed closed loop; in the automotive and robotics tracks, NVIDIA and Horizon Robotics have both deployed heavy forces.
The survival logic for independent edge-side inference chips is not head-on collision, but ecological niche: providing terminal manufacturers with a computing power option that is "not bound to the main controller and can be upgraded independently." There is no disagreement on the direction of terminal intelligence; only the speed varies. Within the window period, "one more computing power supply" itself aligns with the interests of all OEMs.
However, the length of the window period requires prudent assessment. In early 2026, Gartner downgraded its AI PC penetration rate forecast for the year from 55% to 49%, explicitly pointing out that consumers "lack reasons for paid upgrades"; the agency also projected that the achievement of a 50% penetration rate might be delayed until 2028, and entry-level PCs priced under 500 USD might even disappear by 2028. A confirmed direction does not equal a confirmed pace. For independent chip suppliers, if the terminal replacement wave arrives two years later than expected, the survival window will be much narrower than optimistic estimates.
Chip design talents, IP licensing, EDA support, and verification services will continue to be in short supply as new players enter the market in batches.
The Second Card: Storage and Memory
As mentioned earlier, the real bottleneck in edge-side inference lies in memory: insufficient bandwidth and tight capacity. This means that in future terminal products, the value proportion of the memory subsystem will continue to rise—larger capacity LPDDR, higher bandwidth memory solutions, and new storage architectures close to computing units (collectively referred to in the industry as "near-memory computing/processing-in-memory"). In AI computing, the energy consumption of data movement accounts for 60% to 80% of the total energy consumption—whoever saves on data movement in the storage segment helps terminals save on battery life and thermal dissipation.
The analogy is easy to remember: smart EVs shifted the value center of the whole vehicle from the engine to the battery; edge AI is shifting the value center of terminals from computing power to storage.
But this analogy only tells half the story. Rising battery prices are a boon for battery manufacturers but a cost pressure for OEMs; the same applies to rising storage costs. The direct reason for Gartner downgrading the AI PC penetration rate forecast is precisely that surging storage costs have pushed up the selling price of whole machines. The increase in the value proportion of the storage segment and the suppression of terminal shipments by storage costs are two sides of the same coin.
A deeper issue is: if rising storage prices push the selling price of terminal devices so high that it suppresses replacement demand, the "volume ramp-up" of edge-side inference chips themselves will be delayed. The storage bottleneck is not just a cost issue for terminal brands; it is simultaneously a "timing risk" for the entire edge-side inference chain—the revenue realization of all segments in the chain hinges on the premise of terminal shipments. For storage solution providers and startups in new storage architectures, this is the very center of "bottleneck equals value"; for terminal brands, this is a cost variable that must be locked in advance over the next two years; and for chip suppliers, this is a timing variable that must be incorporated into cash flow models.
Global storage giants have already made their moves, and the entrepreneurial window for new storage architectures has just opened—but within the window period, the demand side's bearing capacity must also be factored into the judgment.
Storage solution providers, memory interface and module makers, and startups in new storage architectures are at the very center of "bottleneck equals value"; terminal brands should incorporate storage costs as a core variable in their 2027-2028 product cycles.
The Third Card: Advanced Packaging
To bring computing power and storage "closer together," it does not rely on drawing circuit boards, but on advanced packaging: packaging multiple chips together at high density to shorten distances and increase bandwidth. The inherent demand of edge-side inference chips for "chip + storage integration" will multiply the demand for this segment several times over.
The key point is that this is the segment with the highest global standing in China's semiconductor industrial chain: JCET, TFME, and Hua Tian Technology have long maintained their positions in the top tier of global OSAT. When the global demand for edge AI explodes, China's capacity in this segment happens to be in a structural tailwind position.
However, a constraint needs to be added: advanced packaging capacity itself is currently in a state of global shortage. Advanced packaging capacities such as CoWoS are the key bottleneck in the current AI supply chain, with equipment, substrates, and raw materials all in tight supply. The premise of a "tailwind position" is having capacity to release—if global advanced packaging capacity itself is insufficient, Chinese packaging and testing enterprises indeed have demand, but whether they can translate demand into revenue within the window period depends on the speed of capacity expansion. The judgment of "highest certainty" needs a qualification: demand certainty is high, but the pace of capacity realization is uncertain.
Packaging and testing capacity, packaging materials, equipment, and interposer support are the "highest certainty" beneficiary segments on the map—but "certainty" refers to the direction of demand, not the timing of revenue realization.
| Three Downstream Cards: Terminal Brands, Software Ecosystem, and Industry Customers
The Fourth Card: Terminal Brands
For smartphone, PC, automotive, and robotics manufacturers, the maturity of edge-side inference means two things.
First, the dimension of product definition has changed. The last replacement wave relied on imaging and screens; the key selling point for the next wave will be "intelligence that works well offline"—an Agent experience that can complete the entire process of perception, planning, and execution without an internet connection. Brands that take the lead in differentiating the use of edge-side computing power will be the first to reap the replacement dividend.
Second, the bargaining structure is improving. An independent computing power supplier option in addition to the main control SoC increases the procurement leverage for OEMs. This is also why industrial capital like Lenovo Capital appears on the investment list—the statement by Song Chunyu, Vice President of Lenovo Group and Chief Investment Officer of Lenovo Capital, is quite representative: smart terminals are evolving into Agents with perception, reasoning, and autonomous execution capabilities, and a new computing paradigm will spawn entirely new chip architectures and industrial opportunities.
However, the arrival time of the "replacement dividend" depends on an unresolved contradiction: whether edge-side intelligence can provide a strong enough reason for payment to offset the increase in whole machine costs caused by rising storage prices. Gartner's cautious judgment indicates that this contradiction has not yet been resolved.
For terminal product and supply chain managers, it is recommended to include "edge-side computing power suppliers" in the selection scope for the 2027-2028 product cycle, while simultaneously incorporating storage costs as a variable of the same level into planning.
The Fifth Card: Software and Ecosystem
Chips are only semi-finished products; the other half of the value lies in the software stack: compilers, operator libraries, deployment frameworks, and Agent solutions connecting thousands of industries. The fragmentation of edge-side operating systems (Android, Linux, various RTOS) means that integration and adaptation services that "help chips run smoothly on various terminals" are a business in themselves.
Consider a fact: as of the end of August 2026, Journey 6M has been deployed in over 20 automakers and more than 70 mass-produced car models—from "successful tape-out" to "one million units installed in vehicles," every kilometer in between must be paved by software and services. China has the world's largest developer community and application scenarios; the soil for ecological positioning is not lacking.
Edge-side model deployment toolchains, industry Agent solution providers, and system integrators will reap the second wave of dividends as hardware volumes ramp up.
The Sixth Card: Industry Customers
The last card is on the demand side. Data-sensitive industries such as finance, healthcare, government affairs, and manufacturing have a genuine willingness to pay for intelligence where "data does not leave the domain"—edge-side inference happens to be the only solution: the model runs directly on local devices, and sensitive data never has to leave. For these customers, edge-side solutions are not "cheaper cloud," but "the only option that is compliant." This is an incremental market that barely existed before and is now being opened up.
Peers in To-B services can start building benchmark cases for "edge-side delivery capabilities"; first movers will gain pricing power.
| Three Judgments, and One Reservation
First, the industrial significance of this funding round outweighs the company significance, but the "company significance" should not be underestimated. Four months, over 100 million USD, and the collective presence of top-tier institutions verify that "edge-side high-compute inference" has evolved from a concept into a track. However, the direct trigger for the funding is Chen Peng's scarce credit in the closed-loop mass production of chips. Direction consensus and personal credit are joint conditions: the former explains why the track is heating up at this moment, and the latter explains why it is this company and this price. Separating the two both overestimates the short-term pricing power of the map and underestimates the weight of personal credit in early-stage financing.
Second, value flows to the "bottleneck segments"—but the bottleneck means different things to the supply and demand sides. Chips are the entry point; the largest space on the map belongs to segments like storage and packaging that "feed the computing power," as well as the downstream terminal brands and software ecosystems that turn chips into experiences. However, the rise in value in the storage segment simultaneously means increased cost pressure for terminal brands. Moreover, this cost pressure will transmit upwards along the chain: if terminal shipments are suppressed, chip volume ramp-up will be delayed, and revenue realization across all segments in the chain will be postponed. When evaluating one's own business, it is necessary to first confirm which side of the coin one stands on, and whether one's cash flow can withstand the risk of delayed timing.
Third, the starting hand of China's industrial chain is better than in any previous platform migration. The global standing in the packaging and testing segment, the global share of terminal brands, the largest application market, coupled with the chip and storage shortcomings being made up—during the PC era we were bystanders, during the mobile era we were catching up, and this time, most squares on the map have seats for Chinese companies.
One reservation: the direction on the map is clear, but the time scale on the map is blurry.
Gartner's continuous downward revision of AI PC penetration expectations indicates that there may be a longer road between "confirmed direction" and "scaled implementation" than optimists expect. The suppression of terminal shipments by rising storage prices and the global shortage of advanced packaging capacity are two variables that may further stretch the time scale. The poker game has just begun, and the map goes first—but the "window period" marked on the map needs to be calibrated with one's own cash flow and product cycle, rather than inferred from the speed of funding news.
References:
1. Funding information: Bandao ZongHeng, ZAKER (reported in September 2026)
2. Industry data: PatSnap Eureka (citing ETH Zurich research), Giznova, EEFOCUS
3. Technology roadmap: Science Popularization China, Shenzhen Longgang Government Online; Gartner AI PC penetration rate forecast reprinted from public reports
Disclaimer:
This article is compiled based on public reports. The views expressed represent only the author's personal opinions and do not constitute any investment advice; for relevant information, please refer to official disclosures. In case of copyright infringement or errors/omissions, please contact the official account backend, and we will correct or delete it promptly.