1d:I[70616,["/_next/static/chunks/0kcf5dt6yxdox.js","/_next/static/chunks/1su5n1uv1ti4s.js","/_next/static/chunks/0gvke1wv6udfg.js","/_next/static/chunks/1dfrle3jgmlzh.js","/_next/static/chunks/162lgvduyr-fa.js","/_next/static/chunks/2gcnnopam8e9w.js","/_next/static/chunks/2g60iab5a1pkg.js"],"Breadcrumb"] 1e:I[90762,["/_next/static/chunks/0kcf5dt6yxdox.js","/_next/static/chunks/1su5n1uv1ti4s.js","/_next/static/chunks/0gvke1wv6udfg.js","/_next/static/chunks/1dfrle3jgmlzh.js","/_next/static/chunks/162lgvduyr-fa.js","/_next/static/chunks/2gcnnopam8e9w.js","/_next/static/chunks/2g60iab5a1pkg.js"],"ArticleGrid"] 1f:T1898,

Huawei used its Connect conference to move the Ascend 960DT up by three quarters to the first quarter of 2027, pairing it with a 4,096-accelerator Atlas SuperPoD built on near-package optics. Per-chip, the part still trails Nvidia's next generation by a wide margin. That may matter less than it sounds, because the chips it actually competes against inside China are the ones US export rules still permit.

Key takeaways

The Ascend 960DT arrives in Q1 2027 with up to 288GB of memory at 9.6 TB/s and 2 PFLOPS FP8 / 4 PFLOPS FP4 — double the compute and capacity of the 950 series shipped earlier this year.
A companion 960PR follows in Q3 2027 with 8 PFLOPS FP4 but only 192GB at 2.4 TB/s, splitting inference so prefill runs on one chip and decode on the other.
Huawei says near-package optics let it collapse roughly 48,000 800Gbps pluggable transceivers into about 5,500 Hi-ONE modules, cutting 550 kilowatts and halving failure rates.

What the 960 series actually delivers

The 960DT — DT for decode and training — carries up to 288GB of memory running at 9.6 TB/s, roughly a 2.4x bandwidth jump over the 950 generation, plus 2.2 TB/s of chip-to-chip interconnect. Compute lands at 2 petaFLOPS in FP8 and 4 petaFLOPS in FP4. Huawei is also pushing its own HiF8 and HiF4 numeric formats alongside the standard FP and MX types, on the argument that a single format tuned for both precision and dynamic range can replace two conventional ones.

The 960PR, due two quarters later, inverts the trade. It doubles FP4 throughput to 8 petaFLOPS while dropping to 192GB at 2.4 TB/s, which is exactly the shape you want for the prefill stage of inference, where a prompt is chewed through in one compute-heavy pass. The memory-bound decode stage, where tokens actually stream out, stays on the 960DT. Nvidia had aimed its cancelled Rubin CPX at the same split, and Google's TPU pods have long leaned on the same logic.

How far behind Nvidia is it

The Register's read of the specs puts the 960DT in the neighbourhood of Nvidia's B300 family on memory and bandwidth while delivering about half the FP8 and a third the FP4 compute. Against Rubin, which promises 35 to 50 petaFLOPS of FP4 with 288GB of HBM4 and 22 TB/s, the gap is not close.

The catch is the one that keeps recurring in this market: Rubin and AMD's MI455X cannot be sold in China. Measured against the best accelerator Nvidia is currently permitted to ship there, the 960DT holds its own on dense floating-point work while offering considerably more memory, more bandwidth and support for much larger scale-up domains. For a Chinese lab choosing hardware for 2027, that is the comparison that binds.

Why the pod is the real product

Huawei's strongest card is networking, which is the business it has always been in. The Atlas 960E SuperPoD, slated for Q3 2027 and liquid-cooled, ties 4,096 accelerators into one unified memory domain and claims 8 exaFLOPS FP8 and 16 exaFLOPS FP4. It is built on near-package optics rather than pluggable transceivers — Huawei says the switch replaced about 48,000 800Gbps pluggables with roughly 5,500 Hi-ONE engines, saved more than 550 kilowatts and cut failure rates in half, landing at a claimed 99.8% availability.

Against the Atlas 950, Huawei claims 2.3x training throughput and 2.5x inference throughput on a 10-trillion-parameter workload, with about 70% lower inference latency. Beyond a single pod, it says RoCE or its Lingqu fabric can chain up to 512,000 accelerators, and that a multi-rail topology puts million-NPU clusters within reach. That last figure is a paper configuration, not a deployment, and worth reading as roadmap rather than result.

What to watch before believing the roadmap

The practical question is not the slide deck but whether the silicon trains anything. DeepSeek reportedly hit trouble training on earlier Ascend parts and reverted to Nvidia hardware — a software and reliability failure, not an arithmetic one, and the kind of problem a faster interconnect does not fix on its own. Roadmap slides from the event already sketch an Ascend 970 for 2028 at 14 petaFLOPS FP4 and 14.4 TB/s, and a 980 for 2029 at 28 petaFLOPS FP4 with 384GB, but three-year roadmaps in this industry are drafts.

Still, the direction is consistent with what is happening one layer up, where Chinese open-weight models have been taking share on developer routing platforms. A domestic training substrate that is merely adequate, available in volume and shipped nine months early is a different strategic object from one that wins benchmarks.

FAQ

When do the Ascend 960 chips ship?

Huawei says the 960DT arrives in the first quarter of 2027, three quarters ahead of its original schedule, with the 960PR following in the third quarter of 2027. The Atlas 960E SuperPoD that houses them is also targeted at Q3 2027. None of these are shipping today.

Is the Ascend 960DT competitive with Nvidia's Rubin?

Not on raw compute. Rubin is quoted at 35 to 50 petaFLOPS of FP4 against the 960DT's 4 petaFLOPS, though the two chips carry similar memory capacity. Huawei's counter-argument is scale — 4,096 accelerators in one coherent domain — rather than per-die performance.

What is near-package optics and why does it matter here?

Near-package optics moves the optical engine next to the chip package instead of using separate pluggable transceiver modules. Huawei says that cut roughly 48,000 pluggables down to about 5,500 Hi-ONE units, saving over 550 kilowatts and halving failure rates — which is what makes a 4,096-chip scale-up domain practical to operate.

a:["$","div",null,{"className":"mx-auto max-w-7xl px-4 py-8 sm:px-6","children":[["$","script",null,{"type":"application/ld+json","dangerouslySetInnerHTML":{"__html":"{\"@context\":\"https://schema.org\",\"@type\":\"BreadcrumbList\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https://ainewsway.com\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"#huawei\",\"item\":\"https://ainewsway.com/tag/huawei\"}]}"}}],["$","script",null,{"type":"application/ld+json","dangerouslySetInnerHTML":{"__html":"{\"@context\":\"https://schema.org\",\"@type\":\"ItemList\",\"@id\":\"https://ainewsway.com/tag/huawei#itemlist\",\"name\":\"#huawei Articles\",\"numberOfItems\":1,\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"url\":\"https://ainewsway.com/articles/huawei-pulled-the-ascend-960-forward-nine-months-scale-is-the-argument-not-the-die-mu6qm0jn\",\"name\":\"Huawei Pulled the Ascend 960 Forward Nine Months. Scale Is the Argument, Not the Die\"}]}"}}],["$","$L1d",null,{"items":[{"label":"# huawei"}]}],["$","h1",null,{"className":"mb-8 text-3xl font-bold text-zinc-900 dark:text-zinc-100","children":["#","huawei"]}],[["$","$L1e",null,{"articles":[{"id":"a60d7b91-2dc5-40ba-acc1-2d78c48b2255","slug":"huawei-pulled-the-ascend-960-forward-nine-months-scale-is-the-argument-not-the-die-mu6qm0jn","title":"Huawei Pulled the Ascend 960 Forward Nine Months. Scale Is the Argument, Not the Die","subtitle":"The 960DT lands in Q1 2027 with 288GB and 4 PFLOPS of FP4, still far behind Rubin per chip — but Huawei is selling 4,096-accelerator pods, and Nvidia cannot ship Rubin into China anyway","content":"$1f","excerpt":"Huawei used its Connect conference to move the Ascend 960DT up by three quarters to the first quarter of 2027, pairing it with a 4,096-accelerator Atlas SuperPo...","coverImage":"https://cdn.ainewsway.com/uploads/2026/09/65321d23-3032-4012-9e3f-fc4826c21c9c.webp","coverImageCaption":"Huawei's Ascend 960 roadmap leans on rack-scale systems, tying 4,096 accelerators into a single liquid-cooled Atlas SuperPoD rather than chasing Nvidia on per-chip compute","authorId":"6773f67d-d7fa-4b19-8f38-384f10ac761b","status":"PUBLISHED","publishedAt":"2026-09-18T09:10:03.267Z","viewCount":0,"likeCount":0,"readingTime":5,"isFeatured":false,"isBreakingNews":false,"metaTitle":"Huawei pulls Ascend 960 to Q1 2027, bets on pod scale","metaDescription":"Huawei's Ascend 960DT arrives nine months early with 288GB and 4 PFLOPS FP4, paired with a 4,096-chip Atlas SuperPoD on near-package optics.","metaKeywords":["Huawei Ascend 960","Ascend 960DT","Atlas 960E SuperPoD","near-package optics","China AI chips","Nvidia Rubin","AI accelerator roadmap","Huawei Connect 2026"],"ogImage":null,"aiSummary":"Huawei advanced its Ascend 960DT accelerator by three quarters to Q1 2027, with 288GB of memory at 9.6 TB/s and 4 petaFLOPS of FP4 compute, plus a 960PR inference variant in Q3 2027. Per-chip it trails Nvidia's Rubin badly, but Rubin cannot be sold in China. Huawei's real pitch is the 4,096-accelerator Atlas 960E SuperPoD, built on near-package optics that replace roughly 48,000 pluggable transceivers.","aiTags":[],"sentiment":null,"articleType":"ANALYSIS","videoEmbed":null,"videoType":null,"seoScore":85,"isAutoGenerated":true,"isBackfill":false,"pipelineStage":null,"copyrightScore":null,"similarityScore":null,"matchedSourceId":null,"createdAt":"2026-09-18T09:09:34.405Z","updatedAt":"2026-09-18T12:09:44.735Z","contentUpdatedAt":"2026-09-18T12:09:44.734Z","deletedAt":null,"author":{"id":"6773f67d-d7fa-4b19-8f38-384f10ac761b","name":"Seung Jung","username":"Seung Jung","avatar":null},"categories":[{"id":"8126f4b3-ce4a-464f-bc70-7c3c39d5540b","name":"Tech & Business","slug":"tech-business"}],"tags":[{"id":"51075b6e-7339-4ca5-a00d-1b068653c310","name":"NVIDIA","slug":"nvidia"},{"id":"ae47e924-f108-4955-a5ce-e6b15487fe5e","name":"AI Chips","slug":"ai-chips"},{"id":"e9d9ed20-6211-4e4c-a2af-eb3c5eced7c7","name":"AI Infrastructure","slug":"ai-infrastructure"},{"id":"a73db6d4-a509-47ed-85bc-99a56b8f5db7","name":"China AI","slug":"china-ai"},{"id":"bb8d3605-627b-4977-96e1-49ba5d3460e5","name":"Data Centers","slug":"data-centers"},{"id":"72233652-f0b9-46c4-81f2-7b74a45a84b7","name":"Huawei","slug":"huawei"}],"translations":[{"id":"e05c43fe-e1bb-4981-a4f6-b691cd752169","language":"es","title":"Huawei adelanta nueve meses el Ascend 960: el argumento es la escala, no el chip","subtitle":"El 960DT llega en el primer trimestre de 2027 con 288GB y 4 PFLOPS en FP4, todavía muy por detrás de Rubin por chip, pero Huawei vende pods de 4.096 aceleradores y Nvidia no puede enviar Rubin a China de todos modos","excerpt":"Huawei aprovechó su conferencia Connect para adelantar tres trimestres el Ascend 960DT, hasta el primer trimestre de 2027, junto a un Atlas SuperPoD de 4.096 aceleradores.","metaTitle":"Huawei adelanta el Ascend 960 a 2027 y apuesta por la escala","metaDescription":"El Ascend 960DT de Huawei llega nueve meses antes con 288GB y 4 PFLOPS en FP4, junto a un Atlas SuperPoD de 4.096 chips con óptica de proximidad."},{"id":"304c882a-82a4-4ee5-aa5d-b8fe7218c4ef","language":"ja","title":"HuaweiがAscend 960を9か月前倒し。勝負どころはダイではなく規模だ","subtitle":"960DTは2027年第1四半期に288GB、FP4で4 PFLOPSを引っ提げて登場する。チップ単体ではRubinに遠く及ばないが、Huaweiが売るのは加速器4096基のポッドであり、そもそもNvidiaはRubinを中国へ出荷できない","excerpt":"HuaweiがConnectカンファレンスでAscend 960DTの投入を3四半期前倒しし2027年第1四半期に設定、加速器4096基のAtlas SuperPoDと組み合わせた。","metaTitle":"HuaweiがAscend 960を2027年Q1へ前倒し、規模で勝負","metaDescription":"HuaweiがAscend 960DTを9か月前倒しし2027年第1四半期に投入。288GBとFP4 4 PFLOPSを備え、近接パッケージ光学で結んだ4096チップのAtlas SuperPoDと組み合わせる。"},{"id":"3523f815-9d06-42db-b2de-9d43f36f42fb","language":"ko","title":"Huawei, Ascend 960 출시를 9개월 앞당겼다 — 승부수는 칩이 아니라 규모다","subtitle":"960DT는 2027년 1분기에 288GB 메모리와 FP4 4 PFLOPS로 나온다. 칩 하나만 보면 Rubin에 한참 못 미치지만, Huawei가 파는 것은 가속기 4096개짜리 포드이고 Nvidia는 애초에 중국에 Rubin을 팔 수 없다","excerpt":"Huawei가 Connect 컨퍼런스에서 Ascend 960DT 출시를 세 분기 앞당겨 2027년 1분기로 옮기고, 가속기 4096개를 묶은 Atlas SuperPoD와 함께 내놨다.","metaTitle":"Huawei Ascend 960, 2027년 1분기로 앞당겨 — 포드 규모 승부","metaDescription":"Huawei가 Ascend 960DT를 9개월 앞당겨 2027년 1분기에 내놓는다. 288GB 메모리에 FP4 4 PFLOPS, 근접 패키지 광학으로 묶은 4096칩 Atlas SuperPoD가 핵심이다."},{"id":"78aed690-914d-40fe-8c83-9ea340e3fb6d","language":"zh","title":"華為把昇騰 960 提前九個月推出，賣點是規模而不是單顆晶片","subtitle":"960DT 將在 2027 年第一季登場，配備 288GB 記憶體與 4 PFLOPS 的 FP4，單顆效能仍遠不及 Rubin，但華為賣的是 4096 顆加速器的機櫃叢集，而輝達本來就無法把 Rubin 賣進中國","excerpt":"華為在 Connect 大會上把昇騰 960DT 提前三個季度至 2027 年第一季，並搭配可容納 4096 顆加速器的 Atlas SuperPoD 一同推出。","metaTitle":"華為昇騰 960 提前至 2027 年首季，押注機櫃規模","metaDescription":"華為昇騰 960DT 提前九個月登場，具備 288GB 記憶體與 4 PFLOPS FP4，搭配以近封裝光學連接的 4096 顆晶片 Atlas SuperPoD。"}],"celebrities":[{"id":"b521687f-26e5-4133-941b-a0036102e08e","nameKo":"화웨이","nameEn":"Huawei","slug":"huawei","profileImage":null},{"id":"95f8f574-a1e4-4a77-8783-1419c1421b78","nameKo":"엔비디아","nameEn":"NVIDIA","slug":"nvidia","profileImage":null},{"id":"1829636e-4136-46e9-b226-e1f9e580f7f9","nameKo":"딥시크","nameEn":"DeepSeek","slug":"deepseek","profileImage":null}]}]}],"$L20"]]}]

#huawei

華為把昇騰 960 提前九個月推出，賣點是規模而不是單顆晶片

訂閱新聞簡報

#huawei

華為把昇騰 960 提前九個月推出，賣點是規模而不是單顆晶片