Biren’s 2,000% Revenue Surge Shows China’s AI Chip Opening—and Its Limits

Biren Technology’s first-half 2026 results are a clear sign that China’s market for domestic AI accelerators is expanding rapidly. Revenue reached RMB 1.236 billion ($183.9 million), a 1,997.6% increase from a very small first-half 2025 base, driven by scaled deliveries of Bili general-purpose GPU solutions and AI-computing clusters. 1718

That is meaningful commercial traction—not yet proof of a durable market leadership position. The next test is whether Biren can turn an initial shipment ramp into repeat deployments while building enough supply, software support and customer confidence to compete with much larger domestic rivals.

What the first-half results say

Biren’s reported revenue fell within its prior RMB 1.15 billion to RMB 1.30 billion guidance range. That matters because it suggests the outcome reflected an expected delivery ramp rather than an unexpected one-off beat. 18

Intelligent-computing solutions generated RMB 1.168 billion, about 94.5% of total first-half revenue. The product mix was concentrated: Biren’s near-term growth depends heavily on continued demand for AI GPU systems and clusters. 17

The financial profile also shows both progress and continuing investment needs:

• Gross profit was RMB 527.1 million, and gross margin rose to 42.7%, helped by greater cloud-training-product revenue. 1718
• The company nevertheless recorded a RMB 377.2 million net loss, although that loss narrowed 76.4% year over year. 18
• R&D spending totaled RMB 804 million in the first half, exceeding gross profit, according to reported results summaries.

In other words, Biren has demonstrated that its products can be sold at an improved gross margin, but it is still funding a costly effort to advance its chips, systems and software platform.

Why export controls created an opening

Tighter U.S. restrictions on advanced AI processors have reduced the legal availability of certain Nvidia and AMD products in China. Domestic suppliers have moved quickly into the resulting gap: Chinese vendors collectively shipped 1.65 million cards and accounted for nearly 41% of China’s AI-accelerator server market in 2025, according to IDC data reviewed by Reuters. 1

That market shift is broader than Biren. Huawei led domestic suppliers, followed by Alibaba’s T-Head, Baidu’s Kunlunxin and Cambricon. 1 Biren therefore benefits from the same strategic demand for local alternatives, but it does not operate in an uncontested market.

Biren’s nearly 2,000% growth rate should also be read in context. The percentage comparison begins from first-half 2025 revenue of only RMB 58.9 million. Rapid growth from that base establishes demand momentum; it does not by itself indicate that Biren has achieved the production scale or ecosystem reach of the market leaders. 17

Hardware alone is not the product

For AI infrastructure buyers, an accelerator is valuable only if it works reliably with models, frameworks, libraries, deployment tools and existing engineering workflows. Biren has positioned its BIRENSUPA software platform as a way to support domestic large models and make deployments easier.

That makes software execution central to the company’s outlook. A buyer may consider a domestic accelerator attractive because of availability, procurement preferences or supply-security concerns, but the decision can change if migration costs, tooling gaps or performance tuning are too demanding.

Biren’s durable opportunity therefore depends on more than new chip generations. It must convert cluster deliveries into repeatable, well-supported customer deployments and reduce the practical advantage of Nvidia’s established CUDA-based ecosystem.

Export controls have not eliminated access to Nvidia hardware

The domestic market is not entirely sealed off from restricted GPUs. C4ADS reported that Chinese universities and research institutions procured at least 56 restricted Nvidia chips worth $1.7 million between July 2025 and January 2026. According to the organization, the chips were bundled into larger contracts and routed through companies that were not authorized Nvidia partners.

C4ADS also identified 50 shipments of restricted Nvidia GPUs, worth about $13.4 million, that it said were diverted through trade networks involving Vietnam, India, Malaysia and Hong Kong.

Those findings point to possible diversion and resale routes, not evidence that every intermediary or end user violated export-control law. But they matter commercially: access to familiar Nvidia hardware through unofficial channels can reduce the extent to which Chinese buyers are forced to migrate to domestic platforms.

What is known—and not established—about Megaspeed

Singapore-based Megaspeed International has drawn scrutiny because of the scale of its Nvidia purchases. Bloomberg reported that the company imported at least $4.6 billion in Nvidia hardware from its 2023 founding through November 2025, based on Malaysian and Indonesian customs data.

Bloomberg also reported that U.S. authorities were examining apparent inconsistencies involving Megaspeed’s chip inventory, data-center footprint, ownership structure and the possibility of diversion to China. Reporting has raised questions about potential PRC-linked ownership, but these are investigative concerns, not a public finding of wrongdoing.

Megaspeed has said it complies with applicable laws. Nvidia said it had repeatedly visited Megaspeed facilities, found no evidence of diversion and observed a cloud service permitted under export rules.

The appropriate conclusion is limited: Megaspeed is a significant subject of scrutiny, but the available reporting does not establish that it smuggled chips or that it has been found liable for export-control violations.

The scale challenge: foundry capacity and domestic rivals

Biren’s most difficult task may be producing enough chips and systems at predictable cost. China’s domestic AI-chip supply chain still faces constraints around advanced foundry capacity, packaging and high-bandwidth memory. An assessment from Semiconductor Engineering concluded that China’s domestic suppliers will have difficulty matching Nvidia while those foundry, packaging and HBM constraints persist. 13

SMIC is central to that equation. TrendForce reported that new plants serving Huawei could eventually free more SMIC capacity for smaller chipmakers including Biren, but it also noted that the shift would intensify competition in China’s fast-growing AI-chip market. 12

Biren is competing not just for customer demand but for the manufacturing resources required to fulfill it. Huawei’s leading shipment position, plus competition from Cambricon, Kunlunxin and other local suppliers, makes share gains difficult even in a market expanding because of export controls. 1

The verdict: traction is real; durability is unproven

Biren’s first-half results show that domestic AI-chip demand can translate into material revenue. Its 42.7% gross margin indicates improved unit economics, and its delivery ramp confirms that the company is participating in China’s shift toward local AI infrastructure. 1718

But three conditions will determine whether that momentum becomes sustained market share:

1. Reliable production: Biren needs sufficient foundry, packaging and memory supply to deliver at volume.
2. Software adoption: Its platform must make deployment and migration practical for AI customers.
3. Repeat demand: The company must continue winning deployments after the low-base growth phase, even where Nvidia hardware remains accessible through unofficial resale networks.

The revenue surge is therefore best understood as evidence of a newly opened market—not a final verdict on who will lead it. Biren has established commercial momentum. It has not yet established a manufacturing or software moat strong enough to make that momentum durable.