Tencent’s Canghai V2 Chip Pushes Video Encoding Into Its Cloud Infrastructure Stack

Article summary

Tencent Cloud says its self-developed Canghai V2 video encoding chip has entered mass production after leading MSU hardware encoding benchmarks. The company is positioning the chip as a way to cut bandwidth and compute costs for AI video, live streaming and cloud media workloads. The next test is whether benchmark leadership turns into wider deployment across Tencent Cloud services and external customers.

Tencent’s Canghai V2 Chip Pushes Video Encoding Into Its Cloud Infrastructure Stack

Image source: pandaily.com

Tencent Cloud is moving its Canghai V2 video encoding processor from benchmark proof point to production hardware, giving the Chinese internet group a more direct role in the chips that support cloud media and AI video workloads.

Technology Promise

The company said Canghai V2 has entered mass production after leading the Moscow State University hardware video encoding competition across speed tiers from 30 to 240 frames per second.

Tencent said the processor ranked first on SSIM, PSNR and VMAF measures, ahead of competing products from AMD, Nvidia, Intel and other vendors in that test set.

The central claim is compression efficiency.

Tencent said H.266 performance is about 30% better than the previous generation, while AV1 optimization targets short-video distribution and live-streaming scenarios.

For cloud providers, those gains matter because video traffic can create both bandwidth expense and server load.

Cloud Infrastructure Signal

Canghai V2 is a cloud infrastructure story rather than a consumer-device chip story.

Tencent is using its own silicon to tune video processing for workloads it already runs across games, social platforms, enterprise cloud and media services.

The source also points to an internal deployment path.

Tencent said the chip has been used in major video scenarios including WeChat Channels, Tencent Video and Tencent Cloud Live.

That gives the company a test environment before it sells the capability more widely through cloud services.

Competitive Landscape

The benchmark result gives Tencent a proof point against global chip suppliers, but it does not by itself prove a durable semiconductor advantage.

Video encoding chips are judged by quality, speed, power efficiency, software support and deployment economics.

The strategic value is that Tencent can reduce dependence on general-purpose processors for a workload where it controls both demand and software integration.

If AI-generated video and real-time media keep expanding, specialized encoding hardware could become a larger part of the cloud cost equation.

What To Watch

The next question is scale.

Tencent needs to show that Canghai V2 can deliver lower operating costs in production, not only win benchmark categories.

External customer adoption through Tencent Cloud would be the clearer signal that the chip is becoming a commercial platform rather than an internal optimization project.

#chips semiconductors #Tencent #Canghai V2 #AI video

Chips & Semiconductors

Marvell Teralynx T100 Puts AI Data-Center Switching Into the Chip Race

Marvell announced planned availability of its Teralynx T100 switch chip for AI training and inference infrastructure. The 102.4 Tbps chip is built on a 3nm process, supports up to a 512-port radix and is claimed to use 25 percent lower power than competitive solutions. The practical test is whether data-center customers use lower-power, high-radix switching to ease latency and power constraints in larger AI clusters.

Chips & Semiconductors

Intel Xeon 6+ Launch Puts CPU Supply on the AI Infrastructure Watchlist

Intel launched Xeon 6+ "Clearwater Forest" at Computex 2026 for scale-out data center workloads. The processor tops out at 288 Darkmont E-cores per socket, 576MB of L3 cache and compute tiles built on Intel 18A. The practical test is whether constrained CPU allocation becomes a larger bottleneck for agentic AI data center deployments.

Chips & Semiconductors

Nvidia's RTX Spark Turns AI PCs Into the Next Chip Battleground

Nvidia is entering the AI PC market with RTX Spark, a MediaTek-linked SoC that combines Blackwell GPU technology with a CPU on a single chip. The move shifts Nvidia's AI strategy closer to edge devices, where agentic AI could run locally instead of relying only on cloud infrastructure. Analysts cited in the source said the PC opportunity is still small compared with Nvidia's data center and networking businesses.

Chips & Semiconductors

Ciena's $50 Billion AI Network Target Puts Optical Capacity on the Hyperscaler Clock

Ciena says AI demand could roughly double its addressable market to about $50 billion by 2029 as hyperscalers and service providers invest in optical networking. It cited RLS Hyper Rail, DCOM, coherent modules and 400G/800G pluggable optics as demand areas while planning $250 million to $275 million in capex this year. The practical test is whether AI compute buildouts convert into durable network orders.