Microsoft Maia 300 vs. Huawei Ascend: DeepSeek Deploys 160,000 Chips, Reshaping the Compute Landscape · 1013-735

2026-10-13 00:00

Author: RadarAI Editorial Editor: RadarAI Editorial Last updated: 2026-10-13 Review status: Editorial review pending Brief 速报官方 AI动态开源

Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.

## 🔍 Core Insights This week the AI infrastructure race entered a white-hot phase, as **Microsoft** officially unveiled its **Maia 300** chip and announced plans to achieve **gigawatt-scale** inference deployment by 2027, aiming to reshape the cost structure of compute with its in-house silicon [15][17][20]. Meanwhile, **Huawei Ascend**'s market share in China may have already surpassed **Nvidia**'s, and **DeepSeek** plans to deploy **160,000** Ascend chips [23]. At the application layer, **OpenStocks**' AI agent beat Wall Street consensus in earnings forecasts, winning 60 out of 90 earnings reports [12]. ## 🚀 Key Developments - **Microsoft unveils Maia 300 chip, targeting gigawatt-scale inference deployment** [15][17]: TSMC N2P process, 288GB HBM4 memory, 13.8 PFLOPS FP4 compute, and 40% better energy efficiency than GPU systems. - **Huawei Ascend's China share may surpass Nvidia's, as DeepSeek deploys 160,000 chips** [23]: Bernstein estimates Huawei could capture about 50% of the Chinese market this year, with Nvidia at only 8%. - **NVIDIA and Cisco introduce the new USFP optical module form factor** [11]: The first to support 400Gb/s-per-lane electrical signaling, with a single module capacity of 12.8Tb/s, aimed at scaling AI factories. - **OpenStocks uses an AI agent to forecast earnings, beating Wall Street consensus** [12]: It won 60 out of 90 earnings reports, and 17 out of the most recent 20, with a typical error of just 2.5%. - **NVIDIA paper proposes a new evaluation method to predict agent post-training performance** [0]: Evaluating the base model at the first decisive edit that makes the test pass predicts post-training potential better than step-by-step scoring. - **Blueprints AI claims it can generate permit-ready architectural blueprints in minutes** [2]: It generates code-compliant construction blueprints from 2D sketches or PDFs, supporting synchronized output across multiple disciplines. - **DigitalOcean releases previews of Signals, Simulations, and Insights** [3]: Signals can read real-time agent traffic and flag failing sessions, and supports converting problematic sessions into test cases. - **AMD releases Helion, an agentic AI compute rack** [24]: The CPU-only version can scale to 32,768 cores, is based on OCP design specifications, and lets partners flexibly combine CPU and GPU resources. ## 🔗 Sources [0] NVIDIA paper proposes evaluating the base model at the first edit that makes the test pass, to predict coding agent post-training performance — https://aihot.news/items/h0g8pz7j7qx8xdjxhuan75vq2 [2] Blueprints AI claims it can generate permit-ready architectural blueprints in minutes — https://aihot.news/items/mbt8bp7p1iriwfm1qxxqsrw5c [3] DigitalOcean releases previews of Signals, Simulations, and Insights — https://aihot.news/items/mrcuz6clvtvvow56fcqv1axo5 [11] NVIDIA and Cisco introduce the new USFP pluggable optical module form factor, aimed at scaling AI factories — https://aihot.news/items/n1ixq8s5hctqa70n6ucn879ue [12] OpenStocks uses an AI agent to forecast earnings, beating Wall Street consensus — https://aihot.news/items/ebk0al5gqgcww5hlgb4khx273 [15] Microsoft Maia

This week the AI infrastructure race entered a white-hot phase, as Microsoft officially unveiled its Maia 300 chip and announced plans to achieve gigawatt-scale inference deployment by 2027, aiming to reshape the cost structure of compute with its in-house silicon [15][17][20]. Meanwhile, Huawei Ascend's market share in China may have already surpassed Nvidia's, and DeepSeek plans to deploy 160,000 Ascend chips [23]. At the application layer, OpenStocks' AI agent beat Wall Street consensus in earnings forecasts, winning 60 out of 90 earnings reports [12].

🚀 Key Developments

Microsoft unveils Maia 300 chip, targeting gigawatt-scale inference deployment [15][17]: TSMC N2P process, 288GB HBM4 memory, 13.8 PFLOPS FP4 compute, and 40% better energy efficiency than GPU systems.
Huawei Ascend's China share may surpass Nvidia's, as DeepSeek deploys 160,000 chips [23]: Bernstein estimates Huawei could capture about 50% of the Chinese market this year, with Nvidia at only 8%.
NVIDIA and Cisco introduce the new USFP optical module form factor [11]: The first to support 400Gb/s-per-lane electrical signaling, with a single module capacity of 12.8Tb/s, aimed at scaling AI factories.
OpenStocks uses an AI agent to forecast earnings, beating Wall Street consensus [12]: It won 60 out of 90 earnings reports, and 17 out of the most recent 20, with a typical error of just 2.5%.
NVIDIA paper proposes a new evaluation method to predict agent post-training performance [0]: Evaluating the base model at the first decisive edit that makes the test pass predicts post-training potential better than step-by-step scoring.
Blueprints AI claims it can generate permit-ready architectural blueprints in minutes [2]: It generates code-compliant construction blueprints from 2D sketches or PDFs, supporting synchronized output across multiple disciplines.
DigitalOcean releases previews of Signals, Simulations, and Insights [3]: Signals can read real-time agent traffic and flag failing sessions, and supports converting problematic sessions into test cases.
AMD releases Helion, an agentic AI compute rack [24]: The CPU-only version can scale to 32,768 cores, is based on OCP design specifications, and lets partners flexibly combine CPU and GPU resources.

🔗 Sources

[0] NVIDIA paper proposes evaluating the base model at the first edit that makes the test pass, to predict coding agent post-training performance — https://aihot.news/items/h0g8pz7j7qx8xdjxhuan75vq2 [2] Blueprints AI claims it can generate permit-ready architectural blueprints in minutes — https://aihot.news/items/mbt8bp7p1iriwfm1qxxqsrw5c [3] DigitalOcean releases previews of Signals, Simulations, and Insights — https://aihot.news/items/mrcuz6clvtvvow56fcqv1axo5 [11] NVIDIA and Cisco introduce the new USFP pluggable optical module form factor, aimed at scaling AI factories — https://aihot.news/items/n1ixq8s5hctqa70n6ucn879ue [12] OpenStocks uses an AI agent to forecast earnings, beating Wall Street consensus — https://aihot.news/items/ebk0al5gqgcww5hlgb4khx273 [15] Microsoft Maia

← Back to Updates