Observed Signal · Oct 26, 2026 · Technical Release · Source: Hello China Tech · Impact: 4/5 · Sentiment: Neutral

iFlytek Shows Where Huawei's Chips Fall Behind

Executive Signal Summary

iFlytek, a Chinese speech-recognition company turned LLM developer, released Spark X2.6 on October 22, claiming it was trained entirely on Chinese chips, though not specifying which. According to its 2025 annual report, iFlytek had planned to release a flagship model on Huawei's Ascend 950, but the chip was not mentioned at launch. Instead, iFlytek's earnings briefings reveal detailed comparisons of Huawei's Ascend 910B against Nvidia's A100, A800, and H200 clusters. Findings show that with tuning, the 910B reaches 93% of an A800 cluster on MoE training, but is about five times slower than an H200 in long-context sampling. Huawei's 950DT, the training-focused chip, was still under testing at Huawei Connect in September 2026, with large-scale supply expected by late 2026 or early 2027. iFlytek's data serves as the most detailed customer-side evidence of Huawei's chips' capabilities and limitations.

Polaris7 AgentPolaris7 Strategic Assessment
High Confidence

This article provides concrete, customer-side performance data on Huawei's Ascend chips versus Nvidia's, crucial for assessing the competitive landscape in AI training infrastructure, especially for markets with export restrictions.

MARKET SIGNALS

Track Huawei Signals & Strategic Shifts

Polaris7 connects regulatory filings, primary sources, leadership changes, and M&A activity into structured intelligence. Set up your Polaris7 workspace to monitor relevant companies in your executive briefing.

  • Verified primary sources
  • Automated briefings
  • Instant watchlist setup

Key Takeaways & Evidence Grounding

  • iFlytek released Spark X2.6 on October 22, 2026, trained entirely on Chinese chips.
  • iFlytek's April 2026 comparisons show the tuned Ascend 910B at 93% of an A800 cluster on MoE training.
  • The Ascend 910B is about five times slower than an Nvidia H200 in long-context sampling.
  • Huawei's Ascend 950DT was still under testing as of September 2026.
  • Large-scale supply of the 950DT is expected by late 2026 or early 2027.

Connected Companies & Entities

Ontology Mapping & Concepts

Large Language Models (LLM) & AIInfrastructure#Ascend 950DT supernodes
Primary Source Grounding & Direct Attribution
Direct Origin Attribution
Primary Reporting: Hello China Tech•Published: Oct 26, 2026
Original Coverage Title: “Huawei Ascend vs Nvidia in AI Training: iFlytek's Numbers”

Related Market Signals & Shifts

Recent verified developments and strategic activity across this market segment.

InfrastructureSep 21, 2026

Huawei Unveils Ascend 960 Chips, Claims System-Level Parity with Nvidia

At Huawei Connect in Shanghai, Huawei unveiled its next-generation AI accelerators—the Ascend 960DT and 960PR—scheduled for Q1 and Q3 2027, respectively, about three quarters earlier than planned. Designed for training and inference of advanced open-weight models, these chips power the Atlas 960E SuperPoD, a system connecting 4,096 processors with 8 EFLOPS FP8 and up to 1 PB of high-speed memory. Huawei emphasizes system-level integration and open-source ecosystems, with over 40 models pre-trained on Ascend, including GLM-5. However, per-chip performance still trails Nvidia's Rubin generation, though system-level parity is closer. DeepSeek has ordered at least 160,000 Ascend 950DT chips for inference but still trains on Nvidia hardware. Huawei's in-house HBM development addresses US export restrictions, and the company projects a 100,000-fold increase in AI token processing by 2035.

Read assessment
AI Infrastructure / SemiconductorsJun 5, 2026

Huawei's He Tingbo Unveils 'Tao' Scaling Law vs Nvidia

Huawei's chip lead architect He Tingbo announced the 'Tao scaling law', a new semiconductor performance framework that emphasizes data movement, chip arrangement, wiring architecture, and memory systems rather than further transistor miniaturization. The article reports Nvidia CEO Jensen Huang has effectively conceded China’s AI chip market to Huawei, which holds roughly 50% of that market in 2026. Morgan Stanley projects China’s AI chip market to grow from about $21 billion in 2026 to $67 billion by 2030. Analysts note a persistent compute gap—U.S. AI chips remain several times more powerful—and estimate Huawei could supply only a small fraction of Nvidia’s aggregate AI compute capacity in optimistic scenarios. The piece also covers related geopolitical and economic trends: EU pressure on Chinese imports, rising Chinese FDI into Europe, and Boston Consulting Group data showing Hong Kong overtook Switzerland as the top offshore wealth hub by cross-border assets under management.

Read assessment
AI InfrastructureSep 28, 2026

Alibaba's T-Head AI Chips: Cloud Customers or Qwen Training?

Alibaba announced at its Apsara conference that its new Zhenwu V900 AI chip will enter mass production and go on sale in Q1 2027, two quarters earlier than planned. This follows Huawei's announcement of its Ascend 960DT chip being ready in Q1 2027. Both companies face high demand and limited supply for their chips. IDC data shows Nvidia holds 55% of China's server AI accelerator shipments, Huawei 20%, and T-Head 7%. Alibaba plans to train Qwen models with 5-10 trillion parameters, but has not disclosed which chips will be used, raising concerns about competition between internal model training and paying cloud customers for scarce chip capacity.

Read assessment

WATCHLIST & BRIEFINGS

Receive future signals for Huawei & competitors in your briefing

Stay ahead of material M&A deals, leadership changes, account wins, and product rollouts. Polaris7 connects primary filings, company newsrooms, and trade reporting directly into your executive briefing.

  • Regular primary-source checks
  • Structured Watchlist briefings
  • Preserves current context