국내 GPU 설계 기업 바이런 테크놀로지는 칩 출하 확대에 따라 수익이 20배 증가하며 흑자 전환을 눈앞에 두고 있다.
On August 28, “Hong Kong’s first GPU stock,” Biren Technology, released its first interim report since its IPO: H1 revenue surged nearly 20 times year-on-year to 1.236 billion RMB; the net loss narrowed significantly by 76% to 380 million RMB, with adjusted net loss contracting by nearly 40% to 337 million RMB; gross margin reached 42.7%, up 10.8 percentage points year-on-year.
While domestic GPU companies previously pitched “Nvidia-equivalent specs” and sold on expectations and faith, Biren’s latest results prove that customers are now paying real money.
The capital markets had already priced this in. On its first trading day on January 2, Biren’s shares jumped 76%, pushing its market capitalization past 80 billion HKD. In July, it completed a placement at 46.20 HKD per share, raising approximately 7 billion HKD—a price more than double the IPO level. As of the August 28 close, shares traded at 39.24 HKD, valuing the company at roughly 101.7 billion HKD, doubling its IPO valuation.
However, when placed alongside the “Four Little Dragons” of domestic GPUs, Biren’s market cap appears remarkably “modest.”
Among Moore Threads, MetaX, Biren Technology, and Enflame Technology, Biren was the first to list on the Hong Kong Stock Exchange. Moore Threads and MetaX had already debuted on the STAR Market in December 2025, while Enflame Technology has also passed STAR Market registration.
Despite sharing similar losses and standing at the forefront of the AI computing boom, as of the August 28 close, the market caps of Moore Threads and MetaX were 2.7 times and 2.9 times that of Biren’s Hong Kong listing, respectively.
Is this gap rooted in differing valuation logics between the HK and A-share markets, or does Biren’s “20-fold surge” hide underlying concerns the market has yet to fully digest?
Biren operates as the “engine” for AI computing infrastructure, with products spanning four tiers:
At the base lie the Biri™ series GPGPU chips, with mass-produced mainstays including the BR106 and BR166. Beyond chips, the portfolio includes boards and modules. Moving up, the company delivers multi-card servers and intelligent computing clusters as integrated solutions. Its newest focus is “super nodes”—essentially linking dozens to thousands of cards into a single “supercomputer” designed to train trillion-parameter large models. To ensure clients can “plug and play,” Biren developed its proprietary BIRENSUPA™ software platform, offering Day-0 support for dozens of domestic large models, including DeepSeek, Tongyi Qianwen, and Zhipu GLM.
With the business model clear, let’s examine the data. For H1 2026, Biren reported revenue of 1.236 billion RMB, a 1,997.6% year-on-year increase—nearly 20 times. How does this stack up in the domestic GPU sector? In full-year 2025, Biren’s revenue stood at 1.035 billion RMB. This half alone surpassed last year’s total.
Of this 1.236 billion RMB, the overwhelming majority comes from intelligent computing solutions—the bundled delivery of “chips + boards + servers + software + cluster integration.” This segment generated 1.168 billion RMB in H1, accounting for 95% of total revenue, compared to just 58.15 million RMB in the same period last year—a 20-fold jump in one year. Software revenue, meanwhile, grew from near zero to 65.17 million RMB. Though still modest in scale, it signals that the BIRENSUPA software stack is beginning to generate independent commercial value.
Additionally, rental income from intelligent computing clusters dropped to zero from 707,000 RMB year-ago, indicating a strategic pivot away from renting out compute capacity toward directly selling hardware and solutions.
From the financial report, two key milestones emerge. First, Biren has secured its “ticket” into major internet tech giants. These firms represent the highest technical barriers and largest demand pools in the computing industry, making them notoriously difficult to penetrate. The interim report confirms that Biren has completed supplier certification for these internet leaders and begun bulk deliveries. While domestic GPU firms have long touted “Nvidia benchmarks,” Biren is among the first to successfully convert those claims into actual enterprise spending.
Second, there is the scaled deployment of thousand-card clusters. Partnering with collaborators, Biren delivered a benchmark domestic computing cluster in the Beijing-Tianjin-Hebei region, scaling to several thousand cards. This infrastructure supports sectors like industrial manufacturing, aerospace, and electronics. Such national-level intelligent computing center projects carry high individual contract values and strong demonstration effects, providing a direct boost to revenue.
Underpinning these moves is a product pipeline reaching its harvest phase. The BR106 entered mass production in 2023 but struggled with early adoption. The BR166 followed in August 2025. Throughout 2025, Biren achieved “full-form mass production and scaled delivery” for both chips, laying the groundwork to capture large orders from internet giants and computing centers.
Admittedly, the 20-fold growth rate benefits from a low base. In H1 2025, Biren’s revenue was merely 59 million RMB, whereas MetaX and Moore Threads reported over 900 million and 700 million RMB, respectively, in the same period. Combining this low base with production ramp-up, a broader industry demand surge in H1, and shifting global supply dynamics redirecting orders to domestic vendors, these factors collectively produced Biren’s “breakout” performance.
Market research firm TrendForce has sharply revised its 2026 forecast for domestic AI chips’ share in China’s high-end market upward from approximately 50% to nearly 90%. Meanwhile, investment bank Bernstein predicts Nvidia’s China market share could plummet from around 40% in 2025 to roughly 8%.
Domestic substitution is advancing rapidly, with a clear tier structure emerging: Huawei Ascend leads the pack, Cambricon was the first to achieve profitability, and the “Four Little Dragons”—Moore Threads, MetaX, Biren Technology, and Enflame Technology—form the second tier, all currently scaling up. Beneath the surface lie self-developed ASICs from tech giants like Alibaba and Baidu. It is a track with a massive ceiling and abundant players.
Biren’s 20-fold revenue growth emerged within this window. Yet, bridging the gap from “significantly narrowing losses” to “actual profitability” involves several unavoidable hurdles.
The first is accelerating R&D burn. H1 R&D expenses totaled 804 million RMB, up 40.7% year-on-year. This exceeds the 527 million RMB gross profit for the same period, meaning for every 1 RMB of gross profit earned, 1.5 RMB was spent on R&D.
Looking longer-term makes this clearer. Between 2022 and 2025, Biren’s cumulative R&D investment reached 4.2 billion RMB. Full-year 2025 R&D spending hit 1.476 billion RMB, a 78.5% YoY increase. Against 2025 revenue of 1.035 billion RMB, R&D outpaced top-line growth by over 400 million RMB.
This “R&D inversion” is common in semiconductors; such intensity is standard for chipmakers. The interim report notes that next-generation product series are progressing as planned. Cutting-edge initiatives like advanced packaging, 3DIC, and optical interconnect super nodes require sustained capital injection. Short-term R&D spending will not decline—it will continue to rise. Current high-intensity investment is a bet on the competitiveness of the upcoming BR20X mass production.
The second challenge lies in gross margins. At 42.7% in H1, up roughly 10.8 percentage points year-on-year, the metric shows improvement. However, compared to the 53.8% recorded for full-year 2025, it actually contracted by 11 percentage points.
The likely driver is structural shift. Last year’s higher full-year margin stemmed largely from standardized single-card and board sales, which carried high margins but lower volume. This year’s revenue surge relied heavily on large-scale intelligent computing cluster deliveries. Cluster projects require integrating third-party hardware like servers, storage, and networking equipment. The increased hardware mix diluted overall margins.
Notably, the interim report did not break down gross margins by product line; this assessment is based on revenue composition shifts. Whether margins stabilize as software revenue grows and the BR20X launches will be the critical metric to watch in H2.
The third issue is that revenue has not fully converted to cash. The 20-fold revenue surge remains largely on paper.
While revenue hit 1.236 billion RMB, trade receivables doubled from 532 million RMB at the start of the year to 1.117 billion RMB. Revenue and receivables are nearly at a 1:1 ratio; goods are shipped, but payment is pending. As of the report’s publication date, the cash collection ratio stood at just 23.3%, meaning over 850 million RMB in payments remain outstanding. Simultaneously, inventory rose to 1.215 billion RMB (+28.1% from year-start), and inventory plus service-related prepayments totaled 1.534 billion RMB, a 230% surge from the beginning of the year. Significant cash is tied up in receivables and stockpiling.
There is a constructive angle, however. Biren’s client base consists primarily of major tech firms and government/enterprise entities, which naturally entail longer acceptance and payment cycles. Furthermore, advance inventory positioning secures capacity for next-generation product mass production.
Placing Biren within its peer group reveals that none of the “Four Little Dragons” have achieved true profitability yet: Moore Threads reported a net loss attributable to shareholders of 11.56 million RMB in H1 (151 million RMB after deducting non-recurring items); MetaX posted a quarterly loss of 98.84 million RMB; Enflame Technology reported H1 2026 revenue of 1.1 billion RMB but a net loss of 600 million RMB. All remain in the phase of burning cash to gain scale.
Extending the view to their post-IPO trajectories, as of the August 28 close, Moore Threads traded at 533.88 RMB, up 367% since listing but having retraced over 40% from its intraday peak of 954.99 RMB on debut. MetaX stood at 674.90 RMB, up 545% since listing but pulling back roughly 30% from its mid-July historical high of 997.52 RMB. Biren closed at 39.24 HKD, up 100% since listing, marking the most moderate gains among the three.
The three stocks followed identical momentum patterns: sharp opening highs, subsequent pullbacks, renewed strength in Q2, collective peaks around mid-July, and August retracements. Sentiment across the domestic computing sector essentially traces the same curve.
As of August 28, Moore Threads’ A-share market cap stood at approximately 250.9 billion RMB, and MetaX’s at roughly 270 billion RMB. Biren’s Hong Kong market cap of 101.7 billion HKD translates to about 92.5 billion RMB. Its A-share peers are valued at 2.7 and 2.9 times higher, respectively.
Where does the gap originate? An obvious factor is the listing venue. A-shares attach a “faith premium” to domestic computing stocks, while HK institutional investors prioritize revenue quality and profitability pathways. Additionally, liquidity and share structure differ; scarce hard-tech targets in the A-share market attract capital willing to pay a premium for exclusivity.
The four companies also maintain distinct strategic focuses: Moore Threads emphasizes fully functional general-purpose GPUs, MetaX concentrates on high-performance general-purpose GPUs, Enflame is deeply tied to a major tech giant focusing on cloud-based AI computing for data centers, and Biren prioritizes super nodes and intelligent computing cluster systems—the very battleground defining the next phase of computing competition. Ultimately, how far Biren advances in super nodes and systems engineering hinges on the mass production execution of the BR20X.
Biren has successfully completed the first step from “zero to one.” But the journey from “one to ten” has only just begun.