Tertius News is an AI-native newsroom: an AI model reads the source articles linked below — on this story, all of them from one outlet — and extracts what that outlet reported, claim by claim. How this works →

What is a supernode, and why does it matter for the China-US tech rivalry?

As AI models surpass a trillion parameters, Chinese chipmakers are turning to 'supernode' architectures that link thousands of chips. At WAIC, firms like Biren Technology unveiled optical interconnect solutions to overcome copper-wiring limits, shifting competition from raw chip specs to system reliability and memory sharing.

By Tertius News AI Desk 1 outlet · 2 articles Version 1 Coverage Jul 18–25, 2026 Published Aug 14

Supernodes: China’s AI Chipmakers Shift Focus to System-Level Scaling

As artificial intelligence models expand beyond 1 trillion parameters, a new concept is dominating the computing landscape: the "supernode." At this year's World Artificial Intelligence Conference (WAIC) in Shanghai, Chinese chipmakers from Huawei to Biren Technology showcased new hardware designed to knit hundreds or thousands of chips together to act as one giant supercomputer, according to a report by the South China Morning Post (SCMP).

The focus on supernodes at WAIC reflects a shift in competition across China's AI computing ecosystem, the SCMP reported, citing Morgan Stanley analysts. The analysts said the emphasis is moving away from stand-alone chip specifications and towards "interconnects, memory sharing, and system reliability."

Chen Daliang, CEO of Shanghai-based AI computing resources provider Suanova, a subsidiary of Hong Kong-listed Yeebo Technology, was quoted by the SCMP as saying that "people think putting enough GPUs together makes a supernode, which is definitely not the case."

Biren Technology, a Shanghai-based semiconductor design firm, used the conference to unveil its next-generation "supernode" solutions. According to a separate SCMP report, these systems are designed to link thousands of AI chips across a single cluster by using optical data transmission to bypass current hardware limits.

Ding Yunfan, vice-president of AI framework architecture at Biren, said single-GPU performance was no longer sufficient to handle required workloads. The core challenge, he stressed, was transforming massive GPU clusters into integrated, seamless systems.

Speaking at WAIC on Friday, Ding noted that traditional copper-wired electrical connections were hitting a physical wall, effectively restricting current architectures to 128 GPUs per server. To break through that ceiling, Biren is betting on a distributed, decoupled supernode architecture and near-packaged optics (NPO), which integrates optical fibres closer to chips to boost data speeds. The layout could allow Biren to scale up to 1,024 cards across a cluster.

"Optical interconnect has become an inevitable choice to break through this bottleneck," Ding said.

The launch underscores how highly connected server systems have become one of the latest battlegrounds for AI infrastructure companies, the SCMP report noted. The industry is currently racing to scale up raw computing power as AI models advance to trillions of parameters and AI agent applications gain widespread adoption.

Biren is only the latest domestic player to announce progress in addressing infrastructure limitations, the SCMP added.

How this outlet told it

South China Morning Post as of Jul 25, 2026, 1:30 PM UTC

Framing: Introduction of supernode concept — Informative

Facts Included:

Artificial intelligence models are expanding beyond 1 trillion parameters
The concept of 'supernode' is dominating the computing landscape
Domestic chipmakers showcased new hardware at WAIC
The focus is shifting from stand-alone chip specifications to interconnects, memory sharing, and system reliability

Read the source article →

South China Morning Post as of Jul 18, 2026, 8:30 AM UTC

Framing: Biren Technology's supernode solutions — Technological

Facts Included:

Biren Technology's unveiling of next-generation 'supernode' solutions
The use of optical data transmission to bypass current hardware limits
The limitation of traditional copper-wired electrical connections to 128 GPUs per server
Biren's distributed, decoupled supernode architecture and near-packaged optics (NPO)

Read the source article →

What was claimed, and who backed it

Each row is one claim, attributed to the outlet whose wording states it most clearly. Confidence rates how directly the source text states the claim — explicit and unhedged rates high; hedged, pieced-together, or internally inconsistent statements rate lower. It does not measure whether the claim is true. Status counts the distinct outlets we found asserting it — so a single-source claim can still show high confidence, and a multi-source claim can show medium. Every one of those outlets is named beside the status, so you can check the count against the list. For claims extracted before we began storing that list, the row says so: it names the outlet the claim is quoted from and states that we have not recorded which outlets backed it. Outlets wrote at different times, so a figure that evolves — a casualty count, for example — can legitimately differ between rows; check the "as of" time next to each claim's source.

Claim	Confidence	Status
ClaimArtificial intelligence models are expanding beyond 1 trillion parameters. This expansion is driving the development of new computing technologies.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 25, 2026, 1:30 PM UTC
ClaimThe concept of 'supernode' is dominating the computing landscape. Supernodes are systems designed to link thousands of AI chips across a single cluster.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 25, 2026, 1:30 PM UTC
ClaimDomestic chipmakers showcased new hardware at WAIC, designed to knit hundreds or thousands of chips together to act as one giant supercomputer.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 25, 2026, 1:30 PM UTC
ClaimThe focus is shifting from stand-alone chip specifications to interconnects, memory sharing, and system reliability.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 25, 2026, 1:30 PM UTC
ClaimBiren Technology has unveiled its next-generation 'supernode' solutions, using optical data transmission to bypass current hardware limits.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 18, 2026, 8:30 AM UTC
ClaimTraditional copper-wired electrical connections are limited to 128 GPUs per server.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 18, 2026, 8:30 AM UTC
ClaimBiren is betting on a distributed, decoupled supernode architecture and near-packaged optics (NPO) to boost data speeds.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 18, 2026, 8:30 AM UTC
ClaimDing Yunfan, vice-president of AI framework architecture at Biren, said single-graphics processing unit (GPU) performance was no longer sufficient to handle required workloads.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 18, 2026, 8:30 AM UTC
ClaimDing Yunfan noted that traditional copper-wired electrical connections were hitting a physical wall, effectively restricting current architectures to 128 GPUs per server.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 18, 2026, 8:30 AM UTC
ClaimDing Yunfan said 'Optical interconnect has become an inevitable choice to break through this bottleneck.'	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Jul 18, 2026, 8:30 AM UTC

Transparency & Audit Trail

1 outlet · 2 articles consulted: South China Morning Post 10 claims extracted Version 1 Written 2026-08-14