On September 6, Dawning Information Industry Co.,Ltd. announced the collaborative launch of China's first AI computing open architecture with over 20 upstream and downstream enterprises, introducing an AI super cluster system. The single cabinet can support 96 accelerator cards with computing power reaching 100P level and can be scaled to million-card capacity. Additionally, Dawning launched the "AI Computing Open Architecture Joint Laboratory" based on the National Advanced Computing Industry Innovation Center.
Li Bin, Senior Vice President of Dawning Information Industry, compared this solution to the "Android model for domestic computing power" and disclosed that Dawning has built over 20 large-scale computing clusters and deployed more than 500,000 heterogeneous accelerator cards in the past decade to support systematic promotion.
Beyond demonstrating hardware capabilities and market influence, Dawning Information Industry's move implies multiple explorations of domestic computing power ecosystems, capital markets, and domestic computing power pathways.
**Engineering Declaration or Ecosystem Bet?**
Dawning Information Industry's proposal of single-cabinet 96-card configuration, 100P-level computing power, and promises of "compatibility with multi-brand accelerator cards, compatibility with mainstream software ecosystems (including CUDA)," open storage-computing collaboration, liquid cooling/cold plate specifications, and basic software stack addresses customers' most practical concerns - hardware density, heat dissipation and energy efficiency, software compatibility, and operational thresholds.
For procurement parties, the integrated consideration of whether to adopt is basically premised on clarifying "whether modifications can be minimized, whether stable operation can be achieved, and what long-term operational costs will be."
Dawning Information Industry's simultaneous proposal of engineering specifications and open strategies aims to position usability and accessibility as primary selling points. Li Bin stated, "The domestic intelligent computing industry, like the smartphone sector, needs not only large enterprises like Apple but also Android-style ecosystem innovation models."
Meanwhile, Li Bin emphasized Dawning's cluster operation experience and national platform background, providing natural advantages in promoting "open, standardized, high-efficiency" cluster solutions. This directly positions Dawning not as merely selling machines, but as wanting to position itself as an integrator of "national-level/industry-level" computing power solutions.
Evaluation of such positioning and computing power capabilities should actually focus on the "ability to become productive," rather than making "high-efficiency" judgments based solely on peak values. A common industry mistake is treating computing power "peak values" as everything, but card stacking is often just a gimmick. True value comes from converting "cards into reliable training/inference output" - that is, productivity.
NVIDIA's difficulty in being replaced in the short term is not only due to powerful GPU computing power, but also the existence of CUDA, NCCL, and numerous optimization tools that enable developers and large model teams to migrate models directly to their dedicated clusters with minimal engineering effort.
Dawning Information Industry's emphasis on compatibility with NVIDIA's CUDA ecosystem and exploration of multi-brand card access pathways represents a pragmatic bypass strategy. In fact, Cambricon's ability to break through under NVIDIA pressure also relied on CUDA compatibility, thereby reducing the real costs for domestic cloud service providers when building IDCs and conducting subsequent inference training. Before domestic chip ecosystems become absolutely mainstream, this approach provides a low-friction access layer, promoting customer adoption of local solutions.
In reality, making developers willing to minimize modifications on Dawning's platform requires long-term toolchain, SDK, scheduler, and community support. In other words, Dawning's path is "first circle customers and verify industrial capabilities, then gradually accumulate software and ecosystem" - this is a race of time and execution capability.
Capital markets are typically both enthusiastic and cautious about such companies: if verification nodes are rapidly conquered, valuations will expand; if they only remain in the "showroom" stage of engineering prototypes or demonstration projects, market enthusiasm will quickly cool.
Additionally, Dawning's bundling with joint laboratories and the National Advanced Computing Industry Innovation Center may actually hint at two advantages in China's macroeconomic context: first, initial project landing opportunities are more easily obtained; second, preferential procurement logic for procurement parties (universities, research institutes, state-owned enterprises) toward domestication and security controllability is more easily established.
Simply put, joint laboratories bring not only technical communities but also sources of "demonstration projects." Once end-to-end deployment is completed in research or government projects and produces quantifiable effects (such as training time, energy consumption reduction, cost savings), demonstration projects become reference templates for subsequent commercial replication.
From a secondary market perspective, this "national brand" binding commercial pathway means higher certainty for initial orders, which can be considered a trigger for revenue and cash flow in the short term. Of course, secondary market performance logic places more emphasis on whether Dawning can convert these projects into long-term, replicable commercial contracts and service businesses, rather than one-time demonstration projects.
**Reference: Using Inspur as a Sample**
To place Dawning in a secondary market comparison framework, proper benchmarks are needed. For instance, Inspur is one of the domestic listed companies most directly overlapping with Dawning in servers and data center solutions.
In 2024, Inspur achieved operating revenue of 114.767 billion yuan (up 74.24% year-over-year), with server and component revenue of 114 billion yuan (accounting for 99.33%, up 74.74% year-over-year); gross margin fell to 6.85%, net profit was 2.292 billion yuan, up 28.55% year-over-year.
In the first half of 2025, Inspur achieved operating revenue of 80.192 billion yuan (up 90.05% year-over-year), attributable net profit of 799 million yuan (up 34.87%), and non-recurring net profit of 672 million yuan (up 61.1%), with server business accounting for 93.9%; however, gross margin further declined to 4.51%, server business gross margin was 4.32%; working capital increased significantly, with net cash flow outflow reaching 5.576 billion yuan.
Clearly, while Inspur rapidly captured orders in the AI server market, gross margins were under obvious pressure and cash flow deteriorated. This presents Dawning with an industry portrait of "high revenue + low gross margin" while reminding that pure delivery volume cannot translate into profit quality. Capital markets prefer to see improved profitability and cash flow structure alongside growth.
Comparing Dawning Information Industry and Inspur, three points deserve investor attention: product positioning, commercialization pathways and revenue structure, and valuation and market expectation sensitivity.
Inspur has long focused on servers and cloud-integrated solutions, with customers primarily being cloud service providers and enterprise data centers, offering diverse product lines pursuing shipment scale; Dawning Information Industry emphasizes "national-level/research-level" clusters and open architecture, targeting universities, research institutes, and enterprises with domestication needs. Therefore, in product positioning, these two companies have little competition and even collaboration in some scenarios.
Over recent quarters, Inspur drove rapid revenue growth through large customer orders, with financial reports directly reflecting order-to-revenue conversion; for Dawning Information Industry to demonstrate similar performance improvement in capital markets, it needs to complete the "demonstration project → commercial contract → repeat procurement/service" chain and reflect this in financial reports.
Therefore, when Dawning Information Industry's first domestic AI computing open architecture can convert into contracts with amounts and confirmable terms will inevitably be a capital market focus. Currently, Inspur's stock price fluctuations have largely reflected industry order rhythms and server demand, while Dawning Information Industry, following this announcement, if subsequently disclosing specific cooperation lists and contract amounts, could very likely become a trigger for market revaluation of its growth potential.
In other words, Dawning Information Industry's valuation elasticity highly depends on the progress rhythm of verifying this architecture's commercial value and actual landing scale.
Dawning Information Industry's AI computing open architecture and AI super cluster represent a directional strategic bet: using "open standards + national-level endorsement + engineering capabilities" as entry points, attempting to gain first-mover scale and ecosystem stickiness during the window of rising domestic computing power demand.
However, capital markets will see not just the engineering parameters of "96 cards/cabinet," but whether they can evolve demonstration projects into replicable business models and demonstrate long-term service capabilities.
For secondary markets, short-term positives lie in the possibility of certain orders from policies and demonstration projects; medium to long-term value depends on whether one-time delivery can be transformed into continuous service-based revenue and platform premiums. If Li Bin's so-called "Android model" can be market-validated, then Dawning Information Industry will no longer be merely a server or supercomputing manufacturer, but could become an important carrier in the domestic computing power ecosystem; conversely, if the verification chain cannot be gradually completed, this will be an engineering demonstration, and capital market enthusiasm will quickly return to rationality.
Comments