Cambricon's 6th-gen AI chip targets training as it rides a 108% revenue surge

Cambricon, China's leading AI chip maker, confirmed its sixth-generation processor is in active development while revealing it has already adapted five of the country's most important open-source large models for inference. The disclosure came at its 2026 half-year results call, alongside a revenue jump that shows domestic demand for AI compute keeps climbing while Washington tightens export controls.

The 6th-gen silicon

CEO Chen Tianshi said the sixth-generation AI processor's microarchitecture and instruction set are still in the research phase, with the new silicon built specifically to optimize both training and inference for large models. That's a notable positioning choice. Cambricon has historically leaned on inference workloads for its Siyuan (思元) and Xuansi (玄思) product lines, so an explicit focus on the training side signals it wants a share of the market where Nvidia's remaining export-allowed chips and Huawei's Ascend line currently dominate.

Cambricon says development will push harder on programming flexibility, ease of use, performance, power control, and die-area efficiency. The company frames this as part of a "chip-model co-evolution" strategy — co-designing hardware against the exact model families its customers run, rather than shipping generic accelerators and hoping software catches up.

Five models done, more in the pipeline

On the inference side, Cambricon confirmed it has completed adaptation work for five of China's flagship open-source large models: Zhipu's GLM, DeepSeek, Alibaba's Qwen, Moonshot's Kimi, and MiniMax. On the training side, it says it is actively optimizing DeepSeek, Qwen, and Tencent's Hunyuan. The specific lineups matter — every major Chinese model lab is now either inference-ready or being trained on Cambricon silicon, a signal that the domestic chip-to-model software stack has matured well beyond the "it works but rewrites your code" stage that has historically held Chinese alternatives back against CUDA.

The progress is being tracked closely because Huawei's Ascend has been the default domestic answer to Nvidia, but Cambricon's relentless earnings growth makes it a credible number two — and its model-adaptation momentum is the kind of ecosystem moat that compounds.

The numbers behind it

Cambricon's 2026 first half delivered 5.996 billion yuan (about $830 million) in revenue, up 108 percent year over year, with net profit attributable to shareholders at 2.311 billion yuan, up 122.6 percent. Those figures would be eye-catching for any chip firm; for one at the center of China's self-sufficiency push, they read as confirmation that export restrictions aren't just an obstacle — they're a tailwind for domestic procurement. The big open question is whether Cambricon's training ambitions can break into a supply chain where Huawei already has deep incumbency and Nvidia keeps evolving its stripped-down China-specific designs.

What to watch

Whether Cambricon slots the sixth-generation processor into the Siyuan family or launches it under a new brand, and whether it can expand its model roster into training at the scale Huawei's Ascend enjoys.

Do export controls do more to help China's domestic chip champions than they do to slow them down? Tell us in the comments.

AI 101 — How do AI video generators work?

An AI video generator is a model that starts with a grid of random noise and repeatedly refines it into a moving picture that matches your prompt — the same trick image generators use, stretched across time. Nothing is filmed, and nothing is stitched together from clips the model has seen. The pixels in the finished video were invented, frame by frame, all at once. Why it matters right now Kling 4.0, Kuaishou's flagship video model, went fully live on October 15 with up to 30 seconds of vid

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings