[Table of Contents]
I. Technology and Product Comparison
II. Business Model Comparison
III. Financial Comparison
IV. Risk Analysis
V. Facing the Current Industry Benchmark Infineon, Where is the Breakthrough Point for the Rising Stars?
VI. Conclusion and Outlook
As global digitalization and AI (Artificial Intelligence) technologies evolve rapidly, data centers are experiencing "explosive growth," but they also face unprecedented energy consumption and regulatory pressures. In major global markets, the compliance threshold for Power Usage Effectiveness (PUE) in data centers is being tightened. For instance, China's national target requires the PUE of large data centers to drop below 1.25 by 2025, and Beijing and Shenzhen have set strict PUE red lines for both new and old data centers, with non-compliant enterprises facing punitive tiered electricity prices. Meanwhile, the advent of the AI "megawatt era" has caused the power consumption of computing chips to surge dramatically.
Taking NVIDIA's Blackwell B200 as an example, the power consumption of a single GPU has reached a staggering 1,000W, and the total power requirement for a single AI rack is leaping from the traditional 30-40kW to 100kW or even higher. Under the dual pressure of compliance and the high energy consumption driven by high computing power, the traditional silicon-based power supply architecture has reached its bottleneck under limited physical space and cooling conditions, making energy efficiency improvement the absolute mainline for data center development. Wide bandgap semiconductors such as Gallium Nitride (GaN), which feature higher switching speeds and lower losses, are becoming the core of this energy efficiency revolution. In this arena, Innoscience and Navitas, as two dazzling rising stars, are engaged in fierce competition.
I. Technology and Product Comparison
1. AC-DC Conversion Node (Mains to 48V/12V Rack Power Supply PSU) In the AC-DC front end that converts grid AC power to DC bus voltage, both companies are committed to breaking through the industry benchmark of "Titanium Plus+" (efficiency >96%).
· Navitas: Navitas focuses on extreme integration and hybrid architectures. Its CRPS185 (3.2kW to 4.5kW) power platform for AI data centers innovatively adopts a SiC and GaN hybrid solution—using GeneSiC silicon carbide devices in the PFC stage to handle high voltage and high temperature, and its GaNSafe power ICs in the LLC stage. GaNSafe highly integrates drive, control, sensing, and protection functions into a single 4-pin TOLL package, enabling short-circuit protection within 50 nanoseconds. This platform achieves a power density of up to 137W/in³ and a peak efficiency of over 97%, reducing the volume by 40% compared to traditional silicon solutions.
· Innoscience: Innoscience relies on its "full-chain" GaN-on-Si capabilities to launch the SolidGaN series (such as the ISG612x half-bridge integrated chips), specifically designed for high-power server power supplies from 1kW to 6kW. This series offers TOLL and TOLT packages, supports switching frequencies up to 2MHz, and features extremely low thermal resistance (e.g., the thermal resistance of ISG6124TP is only 0.48℃/W). Compared to traditional silicon solutions, SolidGaN can improve power supply efficiency by 1-2% and increase power density by 50%.
2. DC-DC Conversion Node (48V to 12V Intermediate Bus and Core Power Supply) At the DC conversion node inside the rack, the two companies demonstrate completely different technical routes:
· Innoscience (Deepening the 48V Standard): Innoscience currently holds a strong defensive and counter-offensive position in the industry-standard 48V Intermediate Bus Architecture (IBC). Its low-voltage GaN products (such as the ISG3204LA integrated half-bridge) are specifically designed for 48V to 12V or lower voltage conversion, capable of outputting a peak phase current of 22A. Supported by third-generation technology, it only takes 16 InnoGaN devices to achieve the conduction loss of traditional 32 silicon MOSFETs, doubling the power density while reducing drive losses by 90%.
· Navitas (Directly Disrupting the Architecture with 800V): Navitas, on the other hand, attempts to completely disrupt the existing 48V architecture. At the 2026 NVIDIA GTC conference, Navitas launched a revolutionary 800V to 6V/12V Power Distribution Board (PDB). This solution combines 16 650V GaNFast FETs with 25V silicon MOSFETs, directly bypassing the traditional 48V intermediate conversion stage. At a switching frequency of 1MHz, this single-stage conversion architecture achieves a full-load peak efficiency of 96.5% and a power density of up to 2100W/in³, freeing up valuable server motherboard space for GPUs and memory by significantly reducing the number of conversion stages.
II. Business Model Comparison
Innoscience (IDM Model): As a rare pure GaN Integrated Device Manufacturer (IDM) globally, Innoscience covers the entire industrial chain from epitaxial growth, wafer manufacturing to chip design and packaging & testing. Its core barrier lies in possessing the world's largest 8-inch GaN-on-Si wafer capacity, with planned capacities of up to 70,000 wafers per month at its Suzhou and Zhuhai factories. Although the IDM model requires huge initial capital expenditure, through the economies of scale of 8-inch wafers and a yield rate of over 95%, Innoscience can achieve extremely low per-wafer costs and has a natural advantage in supply chain autonomy and controllability.
Navitas (Fabless Model): Navitas adopts an asset-light Fabless model, investing the vast majority of its funds into R&D and intellectual property construction. In 2024, its R&D expenditure accounted for as high as 91% of total revenue. This model endows Navitas with extremely high agility, enabling it to rapidly develop highly integrated GaNFast and GaNSafe ICs (Integrated Circuits). However, heavy reliance on foundries also exposes it to significant supply chain risks.
III. Financial Comparison
Financial data clearly reflects the different strategic stages the two companies are currently in.
· Innoscience: Relying on capacity release and scaled cost reduction, the company's performance has ushered in an explosion. In the first half of 2025, Innoscience achieved a revenue of 553.4 million RMB (a year-on-year growth of 43.4%). More symbolically, the company successfully reversed its previous gross loss, achieving a first-time positive gross margin of 6.8% in the first half of 2025, and the net loss during the period narrowed to 429 million RMB from the same period last year. Its overseas market is equally rapid, with overseas revenue growing by 118.1% during the period.
· Navitas: For the full year of 2024, Navitas achieved a revenue of $83.3 million (a year-on-year growth of 5%). However, in 2025, the company implemented a major strategic transformation dubbed "Navitas 2.0"—proactively abandoning the lower-margin Chinese mobile phone and consumer electronics markets to fully pivot to high-power, high-margin markets such as AI data centers and NEVs (New Energy Vehicles). Affected by this strategic divestiture, Q3 2025 revenue fell back to $10.1 million (a year-on-year decrease of 53.4%). However, the company maintains a strong balance sheet (holding $150.6 million in cash) and sustained a high non-GAAP gross margin of 38.5%. Its massive $2.4 billion customer pipeline (of which $165 million is from AI data centers) has laid the foundation for a strong rebound in the future.
IV. Risk Analysis
· Patent War Risk (Innoscience): As an industry challenger, Innoscience has been targeted by patent litigation attacks from traditional giants Infineon and EPC. However, Innoscience recently achieved a decisive victory: in early 2026, a US ITC judge preliminarily ruled that Innoscience's redesigned products did not infringe on Infineon's patents; meanwhile, the US Patent and Trademark Office (USPTO) declared EPC's core patents invalid, and the US Customs and Border Protection (CBP) explicitly allowed the free import of Innoscience's new generation products with the "AD" suffix into the US, clearing the biggest legal obstacle for going global.
· Supply Chain Transfer Risk (Navitas): Navitas is facing a severe foundry crisis. Its long-term cooperative foundry giant TSMC (Taiwan Semiconductor Manufacturing Company) announced that it will exit the GaN foundry market before July 2027 to focus on high-margin AI logic chips. This forces Navitas to urgently transfer production to PSMC's (Powerchip Semiconductor Manufacturing Corporation) 180nm process line. This underlying conversion of the supply chain not only brings 12 to 24 months of technical adaptation and certification risks but may also trigger customer concerns about delivery stability during the capacity ramp-up phase.
V. Facing the Current Industry Benchmark Infineon, Where is the Breakthrough Point for the Rising Stars?
Facing the absolute industry giant Infineon—which not only possesses "silicon + SiC + GaN" full-power coverage capabilities but is also advancing towards 300mm (12-inch) GaN-on-Si wafer manufacturing, attempting to crush new entrants through cost advantages and hybrid solutions. Facing this "ultimate boss" and the entrenched traditional architecture standards, the two rising stars have chosen different breakthrough blades:
1. Dimensional Strike at the Architecture Level (Navitas): Navitas is no longer just selling chips, but promoting a brand-new data center power supply architecture. By converting directly from 800V to core voltages (eliminating the 48V intermediate stage), it attempts to render the multi-stage traditional power designs favored by giants like Infineon completely obsolete, seizing the discourse power of AI power supply at the system level.
2. Absolute Scale and Cost Crushing (Innoscience): Innoscience's breakthrough point lies in utilizing the massive capacity of its 8-inch IDM to "commoditize" GaN devices and drive their prices down to levels close to silicon devices. Once the cost dividend of GaN becomes apparent, the economic logic for data centers to maintain traditional silicon solutions or expensive hybrid solutions will cease to exist.
3. Deep Binding with AI Computing Giants: Both companies have bypassed traditional power supply manufacturers to cooperate directly with core computing power setters like NVIDIA. Innoscience won NVIDIA's GaN supplier award for its contributions to the 800VDC rack power supply architecture; Navitas's platform has also been fully integrated into NVIDIA's MGX infrastructure roadmap. Through deep customization, they are leaping from "Tier-2 suppliers" to "Tier-1 technology partners" defining the underlying hardware of AI computing power.
VI. Conclusion and Outlook
In this energy efficiency revolution for AI data centers, Innoscience and Navitas have demonstrated two completely different paths to the future: the former represents unparalleled scale manufacturing and cost control under the IDM model, outputting highly cost-effective "full-chain" GaN products to the global market by breaking through patent blockades; the latter represents ultimate technological agility and architectural disruptive power under the Fabless model, attempting to redefine data center power supply standards through high integration and leapfrog power designs.
By 2030, data centers are expected to consume 7% of global electricity. In the "quantum leap" from kilowatt-level racks to megawatt-level AI factories, whoever can provide the highest conversion efficiency and most reliable supply within the smallest physical space will dominate the energy infrastructure of the new century. Whether it is Innoscience's capacity crushing or Navitas's architectural revolution, this battle of the rising stars will not only reshape the landscape of the power semiconductor market but also become the most solid foundation supporting the endless evolution of global AI computing power.