Humanity has spent sixty years cramming computing power into silicon chips the size of a fingernail, only to be halted once again in recent years by one of physics' most fundamental laws: copper has electrical resistance, and electricity generates heat.
When the power per rack breaks through 100kW, the traditional 48V power supply architecture is like a country road suddenly forced to accommodate a heavy-load high-speed train. From Infineon's roadmap to Delta's 800V prototype, everyone knows the direction, but no one can pinpoint exactly when the inflection point will arrive.
It was not until the announcement of this additional purchase of silicon carbide (SiC) epitaxial wafers from Xiamen materialized that the industry consensus of "AI power supply requires SiC" was nailed down for the first time into a procurement order backed by real money. This is not the starting point of the story, but ironclad proof that the inflection point has arrived: the industry has officially entered the stage of "voting with orders," and the position of SiC in the AIDC power supply chain has since shifted from theoretical deduction to an absolute necessity.
On September 24, 2026, the eve of the Mid-Autumn Festival, a voluntary announcement posted by Hantian Tiancheng on the HKEX turned the tranquility that should have belonged to the holiday into ripples of industrial impact spreading outward.
The order, worth nearly USD 95.8 million, covers 6-inch and 8-inch SiC epitaxial wafers. The delivery deadline has been compressed to the end of December that year, leaving a delivery window of only three months for the production line.
I believe that when many people first saw this news, their eyes were firmly fixed on the order amount, marveling at the scale of this raw material procurement, or wondering whether Hantian Tiancheng could withstand the pressure of short-cycle delivery.
But if we treat it merely as an ordinary commercial contract, we will miss the true signal of the times behind this announcement.
This epitaxial procurement contract from a global leading SiC customer is not an isolated one-off transaction. More accurately, it is part of an upstream material-locking wave spawned by the intelligent power supply revolution in AIDCs amid the explosion of AI computing power. The power reconstruction of computing clusters is rewriting the underlying logic of power semiconductor demand. The growth story of SiC has shifted from a single new energy vehicle (NEV) track to a new cycle driven by the dual engines of automotive and AIDC applications.
All computing power races, traced to their roots, are power races; and the limits of power systems are ultimately defined by materials.
This large-scale order is the most vivid market footnote to the continuous surge in SiC demand from the AIDC industry chain.
Over the past few years, whenever SiC is mentioned, the immediate reaction of industry professionals has always been NEVs. Vehicle main inverters, onboard OBCs, and onboard DC-DC power supplies have supported the most core foundation for the global expansion of SiC substrates, epitaxy, and devices. The capacity planning of almost all domestic SiC companies targets automotive electronics as their primary benchmark market.
In brokerage research reports, industry summits, and industrial exchanges, everyone has taken for granted that automotive applications are the eternal growth engine for SiC. This narrative held true in past industrial cycles, but the rapid surge in AI computing infrastructure is quietly tearing open another gap in the application boundaries of SiC.
As the demand for large model training and inference continues to expand, giant AI data centers are springing up one after another. The traditional data center power supply architecture, used for decades, has hit a physical ceiling in the face of high-density computing power. The inherent characteristics of SiC—low loss, high-temperature resistance, and high switching frequency—perfectly meet the technical demands of the new generation of AIDC intelligent power supply systems. AIDC is no longer a distant concept in industry PPTs; it is transforming into real, rigid, and large-scale procurement demand.
Some might ask, is USD 95.8 million a large amount?
For a leading SiC epitaxy company with over 30% global market share, it is not exactly an earth-shattering contract. However, its timing perfectly hits the bottleneck before AIDC power supply transitions from prototypes to mass production. While NVIDIA is pushing the 800V DC architecture, it is scheduling compatible racks for the second half of 2026 and full-row 2MW-level solutions for 2027; more than 80 companies in the industry are participating in related standards; and the product schedules of Delta, Vertiv, and Eaton are all being pushed forward.
At a time like this, top customers using procurement orders instead of technical white papers is more honest than any research report: Don't wait until the computer room is fully built to realize that it's over—the electricity is unaffordable.
Therefore, this order for Hantian Tiancheng is essentially about global leading SiC device manufacturers locking in upstream materials in advance to stockpile ammunition for the explosion of the AIDC track.
Currently, a fundamental revolution in power supply is taking place in AI data centers. The power supply chain of traditional data centers is a low-voltage system that steps down voltage layer by layer. Mains electricity needs to undergo multi-stage power conversion, ultimately outputting 12V low voltage to supply server chips. This architecture, iterated over decades, is stable and reliable, but its shortcomings are infinitely magnified in the face of the nearly crazy power consumption of AI servers.
The power consumption of a single high-end AI server easily breaks through several kilowatts, and the power consumption of high-density computing racks heads straight for tens of kilowatts. Electrical energy is continuously dissipated during each voltage conversion, ultimately all turning into waste heat. The continuously rising power loss means, on the one hand, that cloud providers have to bear astronomical electricity bills; on the other hand, the massive amount of waste heat forces liquid cooling and air cooling systems to be continuously upgraded, further driving up the construction and O&M costs of data centers.
Power loss and heat dissipation pressure have become invisible shackles restricting the continuous expansion of ultra-large-scale AI clusters. The core idea of the new generation of AIDC intelligent power supply transformation is to raise the bus voltage, streamline the power conversion hierarchy, bid farewell to the outdated 12V and 48V low-voltage solutions, and shift to high-voltage buses of hundreds of volts.
Power conversion in high-voltage scenarios is precisely the home ground for SiC power devices. Traditional silicon-based MOSFETs and IGBTs suffer from persistently high switching losses and severe heating under high-voltage and high-frequency operating conditions, making it difficult to adapt to the compact and high-efficiency design goals of AI racks. SiC devices can operate stably at higher voltages and higher switching frequencies, compressing the volume of power modules, reducing power conversion losses, lowering heat dissipation supporting investments, and increasing the overall power density of the machine.
For a giant computing cluster composed of tens of thousands of servers, a tiny improvement in the efficiency of a single power module, under the superposition effect of hundreds of thousands of hours of uninterrupted operation, will ultimately precipitate into visible cost savings and O&M advantages for cloud providers.
AI computing clusters are essentially digital factories that continuously consume and output massive amounts of electricity. The efficiency of power conversion directly determines the upper limit of computing costs. When the computing power competition extends from model parameters and GPU chip computing power to the underlying competition of power supply system efficiency, SiC naturally becomes the core material in this computing infrastructure race.
Computing chips determine how fast AI can run, while the power supply system determines how long AI can run and how cheaply it can operate. It is precisely this hardcore logic that drives global leading cloud service providers and computing infrastructure builders to batch-introduce SiC power modules in new-generation AIDC projects. For downstream device manufacturers to seize the window of opportunity in the computing market, they must lock in the core intermediate material—epitaxial wafers—in advance to ensure device manufacturing capacity. This is the underlying root cause of the birth of this large-scale epitaxial order.
The transmission chain is already very clear: large models continue to grow larger → clusters continue to expand → racks continue to become denser → 48V hits the ceiling → 800V HVDC and SST transition from concepts to infrastructure → the power supply system shifts from "logistics" to a "prerequisite for computing power" → power semiconductors change from supporting roles to leading roles → SiC epitaxy, substrates, devices, and modules are all revalued across the board.
Hantian Tiancheng is just a drop of water splashed in the middle of the chain, but this drop proves that the water level has risen. AI data centers do not "just happen to use some SiC"; the new architecture cannot do without SiC.
Throughout the entire SiC industry chain, epitaxial wafers are a leading indicator that connects upstream and downstream and provides insight into the industry's temperature. Its prosperity always runs ahead of terminal device orders.
In SiC chip manufacturing, the first step is to prepare the SiC substrate, and then grow an epitaxial layer with specific parameters on the substrate surface. Core indicators that determine device performance, such as chip breakdown voltage and on-resistance loss, are almost entirely determined by the thickness, doping concentration, and intra-wafer uniformity of the epitaxial layer. The substrate is the foundation of the material, but epitaxy is the soul process that defines device performance.
It can be said that the stocking rhythm of epitaxial raw materials is the most sensitive window to peek into the real downstream demand.
Terminal device orders can still be divided into multiple stages such as sample validation, small-batch trial production, and large-scale mass production. However, large-scale procurement of epitaxial wafers, especially orders with such short delivery cycles, is by no means for laboratory R&D samples; it is forward stocking for scaled mass production.
The announcement disclosed that the customer for this order is a globally leading listed company in the SiC field, possessing mature capabilities to supply SiC power devices in bulk to global AIDC customers. The customer's simultaneous procurement of both 6-inch and 8-inch epitaxial wafers hides the profound meaning of layered industry demand.
The 6-inch SiC epitaxy process is mature with controllable costs, making it the main choice for mass production of medium- and high-voltage power devices today; the 8-inch wafer is the industry-recognized medium- to long-term evolution direction. A larger wafer size means more chips can be produced per wafer, diluting the manufacturing cost per device.
Top manufacturers simultaneously stepping up efforts on two generations of sizes means that the implementation of SiC devices in the AIDC field is no longer a single-point pilot in the laboratory, but a new stage of multi-specification parallelism and large-scale commercial implementation. The customer's choice to complete delivery in just three months sends a particularly strong signal: the implementation pace of downstream AIDC projects is accelerating, and device manufacturers dare not bear the risk of raw material supply interruptions. They must lock in upstream capacity in advance to avoid having their delivery schedules bottlenecked by material constraints.
For a long time, the value story of the SiC industry has been firmly bound to the electrification transformation of automobiles. The rise of the AIDC scenario proves that the value of SiC extends beyond the wheels. In the crisscrossing power arteries of the computing world, wide-bandgap semiconductors also hold an irreplaceable position. Therefore, the wave of AIDC computing power supply brings the SiC industry a historic opportunity to completely open up its application boundaries.
NEVs belong to the consumer-end market, where order fluctuations follow the ups and downs of vehicle price wars, consumer confidence, and automakers' product cycles, exhibiting phased alternations of heat and cold. In contrast, AIDC computing infrastructure belongs to a rigid track driven by capital expenditures. The demand for large model training and inference from global tech giants continues to expand. Computing power expansion is a long-term strategic choice and will not easily hit the pause button due to short-term market fluctuations.
It can be said that once cloud providers complete the technical validation of the high-voltage power supply architecture and initiate scaled deployment, the orders will have extremely strong continuity. Project cycles span several years, and the demand for power devices corresponding to a single giant computing cluster is astonishing. At the same time, data center power supplies operate uninterrupted year-round, imposing nearly harsh requirements on device reliability and batch consistency. The certification cycle for the entire supply chain is lengthy. Once supply chain barriers are established, customer stickiness is extremely strong, making it very difficult to be easily replaced.
Therefore, as long as leading SiC device manufacturers successfully integrate into the AIDC supply chains of overseas cloud providers, they will form a stable long-term order pool, which in turn will continuously drive the procurement demand for epitaxial and substrate raw materials, avoiding the drastic ups and downs seen in consumer electronics.
Accompanied by the concentrated launch of a batch of large-scale computing projects overseas, global leading SiC device enterprises are taking preventive measures to lock in upstream epitaxial capacity, preventing capacity shortfalls from dragging down device delivery progress. This is precisely the core logic behind overseas leaders directing their orders to high-quality domestic epitaxial manufacturers.
In the coming years, global AIDC construction will maintain high prosperity, and the penetration rate of high-voltage power supply architectures will steadily rise. What follows is an endless stream of SiC power device demand, which will transmit upwards along the industry chain, continuously driving the demand for upstream raw materials such as epitaxy and substrates.
Therefore, the automotive market is the ballast of the SiC industry, stabilizing its core foundation; the AIDC market is the new variable that breaks the growth ceiling and broadens the industry's growth boundaries. The two are not a zero-sum game; rather, they mutually reinforce each other, jointly amplifying the growth space of the SiC industry.
This order is also strong evidence of the continuous breakthrough in the global competitiveness of the domestic SiC epitaxy segment.
*Hantian's capacity layout
For a long time, the core lifeline of the global SiC industry chain was in the hands of overseas enterprises. Upstream material segments such as substrates and epitaxy were once the most prominent shortcomings of the domestic industry. Restricted by gaps in equipment, formulations, and process accumulation, domestic epitaxial manufacturers in the early years could only undertake small-batch sample orders and found it difficult to secure mass production orders from international first-tier device manufacturers.
After nearly a decade of continuous R&D investment and process iteration, leading domestic epitaxial manufacturers have gradually overcome the challenges of large-size epitaxy processes. In core indicators such as epitaxial uniformity and defect density control, they have secured their place in the international first tier. Relying on stable delivery capabilities and cost advantages, they have entered the supply chain systems of global leaders.
In the past, when overseas SiC giants procured epitaxy, they prioritized local suppliers; today, overseas leaders are proactively placing mass production orders worth nearly USD 100 million with domestic manufacturers. This indicates that domestic wide-bandgap semiconductor upstream materials have grown from past followers into an undeniable force in the global supply chain, and it also represents the stamped approval of domestic process capabilities by the world's most stringent supply chain systems.
Looking at it from the energy dimension, future comparisons of AI capabilities between countries will not just be about who has models, data, and H-series cards, but also about who has clean, stable, and scalable megawatt-level electricity.
Ireland, Virginia, Frankfurt, Tokyo, the Yangtze River Delta, and the Greater Bay Area are all hitting the same wall: grid access is tighter than graphics cards, and substations are scarcer than GPUs. Cooling water and green power contracts have become strategic resources. In this context, the value of SiC will be elevated to the infrastructure layer: it does not generate power, but it determines how much of a kilowatt-hour from the grid actually turns into training steps and how much turns into waste heat in the computer room.
In a 100MW intelligent computing center, reducing system loss from 12% to 7% saves several megawatts of power that can be stuffed back into racks; using the common algorithm in the industry, a highly efficient SiC solution under a 100MW grid capacity can yield tens of GWh of available energy per year.
CFOs may not necessarily understand bandgap width, but they certainly understand this math.
*NVIDIA launches the next-generation 800V HVDC architecture, directly converting 13.8kV AC to 800V high-voltage DC outside the data center.
The semiconductor industry is not a sprint of single-point breakthroughs; it is more like a marathon of collaborative assault across the entire industry chain. Securing international orders in the epitaxy segment is a step in the upward climb of domestic SiC, far from the finish line.
In the face of opportunities, we also need to remain rational. We cannot fall into blind optimism just because of a large order, let alone use it for speculation.
AIDC power supply solutions are still in the stage of technological iteration. Different cloud providers still have分歧 (differences) in their technology route choices, and the comprehensive popularization of high-voltage architectures requires time for validation; the continuous expansion of the entire SiC industry chain may also lead to phased overcapacity in the future, triggering price competition; the uncertainty of the global trade environment will also bring disturbances to cross-border supply chain cooperation.
Opportunities and challenges always go hand in hand. The AIDC track has opened up the imagination space for the SiC industry, but the long-distance race of the industry will ultimately return to the competition of hardcore capabilities such as product yield, reliability, and cost control.
Let's look at the time dimension again.
The incremental market brought by the AIDC track has given domestic manufacturers a precious window of time to open up a new application battlefield beyond the automotive market. Only by continuously refining processes, accumulating mass production data, and optimizing cost structures can they achieve dual accumulation in technology and scale.
In the geopolitical dimension, Chinese companies are by no means bystanders in this race, but whether they can integrate into the material plans of international giants tests their hard skills. The reconstruction of the AIDC power supply chain will not simply replicate the old order of "European and American power factories taking it all," nor will it ever play out the warm script of "automatic realization of domestic substitution." This is a cruel screening of engineering capabilities: whoever first solidifies the combination of yield, reliability, dual certifications for automotive and data centers, 8-inch scale, and 12-inch reserves will be the one to move from "usable" to "preferred."
As the saying goes, dividends are always reserved for the prepared supply chain; concept speculation cannot help enterprises cross the industrial cycle.
OK, let's get back to the order itself. Hantian's total revenue for the full year of 2025 was CNY 765 million, with the largest customer accounting for 35%, equivalent to about USD 37.07 million. This landed order of USD 95.8 million is equivalent to about CNY 692 million. The order scale is about 258% of this customer's total procurement for the full year of 2025. If fully landed, it will reach 90% of Hantian's total revenue for the full year of 2025.
In my view, the most fascinating part of this announcement is not "I got the order," but rather "the global SiC leader is反过来 (in turn) adding orders to me." This is like a subtle Industry confession: The fire of AIDC has burned from the computer room to power semiconductors, from power semiconductors to epitaxial wafers, and from epitaxial wafers to the entire third-generation semiconductor narrative.
The reason why this USD 95.8 million additional order from Hantian Tiancheng deserves to be放大 (amplified) and scrutinized is precisely because it bridges this gap. It marks the transition of SiC from a "future solution" in PPTs to a BOM list in the procurement plans of global leaders.
In the future, when we recall the Mid-Autumn Festival of 2026, we may not remember the news of some model topping the charts, but we will remember that an epitaxial wafer company in Xiamen, with an order of less than 100 million US dollars, once again powerfully exposed the power shortage in AI data centers.
The distance from electricity to chips is more physical and more残酷 (cruel) than the distance from chips to intelligence.
SiC merely pushes the wall of physical laws outward by a few centimeters: slightly higher breakdown voltage, slightly lower loss, slightly higher junction temperature, and slightly slimmer volume. But it is precisely these few centimeters that allow humanity to cram 100,000 cards into the same building, allow a conversation to run trillions of floating-point operations in a few hundred milliseconds, and make "intelligence" no longer just an algorithm story, but a reality supported by the power grid, materials, processes, cooling, standards, engineers, and capital.
Actually, this epitaxial order of nearly 100 million US dollars, placed in the grand coordinates of industrial time, is just a conspicuous wave in the tide of AIDC driving SiC demand. But it tells the market that the growth narrative of SiC has turned a new page. The power transformation brought by AI computing power is reshaping the industrial landscape of power semiconductors, and the curtain of the era of wide-bandgap materials has just been pulled open.
*Using 800V HVDC compared to 415 VAC can increase wire transmission power by 157%
In the future, along the main line of AIDC, from power modules and power devices all the way upstream to epitaxy and substrates, the entire SiC industry chain will usher in continuous demand pull. The long-term dividends of the computing power era will ultimately only flow to players who master the capabilities of underlying materials.
Only enterprises that can seize the opportunities of computing power supply transformation and continuously refine their technology and supply chain capabilities will be able to gain a firm foothold in this long industrial wave and share the long-term dividends of the digital computing power era.
The endgame of the computing power war is the power war, and the winning hand of the power war is currently lying in the epitaxial furnaces of the Xiamen factory.
Reference Materials:
[1] Hantian Tiancheng
[2] NVIDIA
[3] Infineon
[4] CASA