On September 11, SICC revealed at its semi-annual performance briefing that the company's "Diamond Single Crystal Preparation and Substrate Surface Fine Processing" project has entered the sample stage.
Two details are intriguing:
The total investment for the project is only CNY 7 million, with CNY 609,400 invested in the first half of the year and a cumulative investment of CNY 5.369 million;
The second diamond project disclosed at the same time last year, "Diamond Substrate Material R&D and Application" (total investment of CNY 16.8657 million), has quietly disappeared from this year's semi-annual report.
On one hand, there is the industry's frenzy with Huanghe Whirlwind betting CNY 2 billion, Liliang Diamond redirecting CNY 1.028 billion, and Huifeng Diamond building a factory with CNY 1 billion; on the other hand, there is the global leader in silicon carbide (SiC) substrate market share (27.6%), using a budget of less than CNY 10 million for "small-step trial and error".
01. 2026, the First Year of Commercialization of Diamond Heat Dissipation
The power consumption of AI chips is breaking through the physical ceiling of heat dissipation. The power consumption of a single NVIDIA GPU card has climbed from over 400W for the A100 to over 1500W for the B200. Once the local heat flux density exceeds 500W/cm², traditional copper-based heat sinks have approached their limits—and the upper limit of copper's thermal conductivity is only about 400W/(m·K).
Diamond's answer is: thermal conductivity of 2000-2200W/(m·K), which is 5 times that of copper and 4-5 times that of silicon carbide.
In 2026, this dormant industry narrative was completely ignited:
NVIDIA disclosed at CES 2026 that the next-generation Vera Rubin architecture GPU will adopt a combined solution of "diamond-copper composite heat sink + 45℃ warm water direct cooling"; GTC 2026 further confirmed that the Rubin and Feynman architectures will be fully equipped with "diamond + liquid cooling" as standard.
Akash Systems delivered the world's first batch of H200 commercial AI servers equipped with diamond heat dissipation to India's NxtGen AI, improving GPU energy efficiency by 15% and securing USD 300 million in orders.
Lenovo released the world's first consumer-grade laptop with diamond-copper heat dissipation, extending the technology from data centers to consumer electronics.
TSMC selected single-crystal diamond as the backside heat dissipation solution for high-end AI chips after completing comparative tests between silicon carbide and single-crystal diamond.
Industry estimates show that in the AI server scenario alone, the market space for diamond heat dissipation materials is between CNY 12.6 billion and CNY 25.2 billion. If coupled with advanced packaging upgrades, the ceiling will be further raised.
02. Superhard Material Enterprises "Collectively Change Tracks", While SICC Chooses to "Stand Still"
Superhard Material Camp - Capacity First, All-in on Heat Dissipation:
| Enterprise | Action | Scale |
|---|---|---|
| Huanghe Whirlwind | China's first 8-inch diamond heat sink wafer production line put into operation, planning 300 MPCVD equipment units in 3 years | Investing CNY 2 billion in the next 3 years, targeting an annual capacity of 150,000 wafers |
| Liliang Diamond | Changed the use of raised funds, redirecting the lab-grown diamond expansion project to diamond functional materials | CNY 1.028 billion |
| Sf Diamond (Tianxuan Semiconductor) | Building a diamond heat dissipation wafer production line in Shayar, Xinjiang | CNY 450 million, 25,000 wafers/year, expected to be put into operation in Q1 2027 |
| Huifeng Diamond | Construction started on the Baotou CVD diamond project | CNY 1 billion, 500 MPCVD equipment units in the first phase |
The logic of this camp is very clear: the capacity for synthetic diamond has long been overcapacity (the collapse in lab-grown diamond prices is the best footnote), and extending to downstream heat sink wafers and heat dissipation wafers is the shortest path to monetize existing material capabilities. According to public reports, Liliang Diamond's products have entered NVIDIA's validation process, and the yield rate of Huanghe Whirlwind's 8-inch production line is stable at over 85%.
SICC Camp - R&D Positioning, Refusing to Build Factories:
SICC's diamond layout began two years ago: in April 2025, it centrally announced 5 patents (covering diamond polycrystalline substrates, SiC-diamond composite substrates, etc.); in May 2026, it disclosed the "Diamond Crystal Growth Method" patent, reducing the total dislocation defect density of the crystal to below 10⁵/cm² and the absolute value of stress to less than 20 MPa.
However, over the past two years, the actual cash investment is only CNY 5.369 million—not even enough for Huanghe Whirlwind to buy one MPCVD equipment unit.
Why? Looking at the financials makes it clear.
In the first half of 2026, SICC's revenue was CNY 914 million, a year-on-year increase of 15.10%, but the net profit attributable to the parent company was -CNY 58.6166 million, and the operating cash flow was -CNY 203 million. The main silicon carbide business is in a critical period of 8-inch capacity ramp-up, with heavy depreciation pressure. Adding a new heavy-asset business requiring billions of yuan for SICC when the main business is not yet profitable would not be accepted by the capital market.
More importantly: SICC is not a superhard material enterprise; it has no idle diamond capacity to transform. Building a production line from scratch is equivalent to using the cash flow of silicon carbide to gamble on a track where it has absolutely no cost advantage.
Therefore, its choice is: CNY 7 million, R&D samples, tracking technology, and reserving patents—spending a small amount of money to secure a position and wait for the cards to be dealt.
03. Silicon Carbide + Diamond: Natural Strategic Synergy
SICC's true trump card is not diamond itself, but the intersection of silicon carbide and diamond.
First Intersection: Complete Overlap of Customers and Markets.
Silicon carbide and diamond face the same batch of customers—manufacturers of high-power electronic devices, optical modules, lasers, and RF devices. The customer resources, validation experience, and substrate processing capabilities accumulated by SICC can be directly transferred to the diamond business. This is an asset that Huanghe Whirlwind and others do not possess: they know jewelers and abrasive factories, while SICC knows Infineon, Bosch, and China's power device factories.
Second Intersection: SiC-Diamond Composite Substrate, the Natural Connection Point of the Two Business Lines.
In SICC's patent portfolio, the most noteworthy direction is the silicon carbide-diamond composite substrate. This technical route directly addresses the core pain points of composite substrate industrialization—thermal expansion coefficient mismatch and interfacial thermal resistance. And to make SiC-diamond composite substrates, the prerequisite is to first push SiC substrates to the extreme—which happens to be where SICC is the global leader (with a 27.6% market share in conductive SiC substrates in 2025, of which the 8-inch market share is 51.3%).
In other words: when others make diamond heat sinks, they are doing a "materials business"; when SICC makes diamond, it is building a moat for its main silicon carbide business.
A very realistic deduction: if AI chips and advanced packaging fully shift to diamond heat dissipation in the future, the world's strongest SiC substrate supplier, equipped with diamond substrate capabilities simultaneously, will have double the bargaining power and customer stickiness.
04. The Other Side of the Industry: Cold Data Amidst the Frenzy
Of course, a few buckets of cold water must be poured on this frenzy.
Cold Water One: Device-level applications are still far away. Commercialization in 2026 is concentrated on "passive heat dissipation"—heat sinks, composite cold plates, and heat spreaders. Diamond as a semiconductor substrate (power devices, quantum devices) is still in its early stages: Japan's Orbray monopolizes the 2-inch single-crystal substrate market, and China still has a gap of about 5 years in core technologies such as large-size high-quality single-crystal substrates, twin-free (111) plane growth, and n-type doping control. The pattern of "Japan selling wafers, various countries making devices" has not yet been broken.
Cold Water Two: Material routes are still competing in layers. The main force for mass production in the short term (2026-2028) is diamond-copper composite materials with mature processes and lower costs (600-800W/(m·K)); polycrystalline CVD diamond targets the medium-to-large area market after 2027; single-crystal diamond (above 2000W/(m·K)) is so expensive that only extreme local heat conduction scenarios can afford it. The single-crystal route that SICC is betting on is precisely the one farthest from making money.
Cold Water Three: SICC itself is also contracting. A CNY 16.86 million project disappearing from the semi-annual report is more honest than any strategic interpretation—the company admits that resources are limited, and it is more realistic to thoroughly penetrate the single line of single-crystal preparation + substrate processing than to spread out.
Conclusion: Two Paths, the Same Direction
Panning the camera out, this is a race with different starting points but the same finish line:
Huanghe Whirlwind, Liliang Diamond, and others start from the capacity of synthetic diamond, climbing towards the heat dissipation needs of upstream chip customers, betting on the "monetization of material capabilities";
SICC starts from the industrialization and customer capabilities of silicon carbide, diving deeper into more extreme thermal management materials, betting on the "synergistic evolution of the main business".
One is sprinting with heavy assets, the other is positioning with light R&D. It is too early to draw conclusions on who is right and who is wrong—the mass production hurdles such as thermal expansion mismatch of composite substrates, interfacial thermal resistance, laser cutting yield rate, and costs still lie ahead of everyone.
But one thing is certain: when AI computing power pushes chip power consumption into the kilowatt era, heat dissipation is no longer a supporting role in engineering, but the key to determining the upper limit of computing power. For the first time, diamond stands at the center of the semiconductor industry stage. This time, it plays not a jewel, but the "calmest computing power base".
Disclaimer: Compiled by Flink Future Industry Chain. For reprint, please contact: 18158256081 (same as WeChat)