DeepSeek is turning to Huawei chips for training its AI models, a move that helps the company work around U.S. export restrictions on advanced semiconductors. The startup has invested 30 billion yuan to build an 8-trillion-parameter model and plans to use more domestically made chips going forward. Huawei chips are expected to arrive by late 2026 or early 2027, and CEO Liang Wenfeng believes Huawei's performance will catch up with Nvidia within a few years. Training a 2-trillion-parameter model requires about 60,000 Huawei Ascend chips compared to 30,000 Nvidia H100/H200 chips.
Key Takeaways
- DeepSeek is adopting Huawei Ascend chips for AI training to circumvent U.S. export restrictions on advanced semiconductors to China
- Training a 2-trillion-parameter model requires roughly 60,000 Huawei Ascend chips versus 30,000 Nvidia H100/H200 chips
- DeepSeek has invested 30 billion yuan to build an 8-trillion-parameter model using domestic chips
- Huawei chips are expected to arrive by late 2026 or early 2027, with CEO Liang Wenfeng calling this one of the company's biggest strategic bets
- A Reuters/Ipsos poll found 73% of Americans believe AI companies are not doing enough to prevent serious harm
- About 39% of poll respondents said AI is having a negative impact on society, and 55% said it would be good to slow AI development
- Salt Security launched native AI Detection and Response to protect against prompt injection, jailbreak attempts, and unsafe model behavior
- About 45 data center projects worth $68 billion were blocked or delayed by local opposition between April and June
- The first quarter of 2026 saw 75 disrupted data center projects worth about $130 billion across the U.S.
- European AI spending is projected to reach nearly $470 billion by 2030, growing at a compound annual rate of 35%
DeepSeek turns to Huawei chips for AI training
Chinese AI startup DeepSeek is now using Huawei chips to train its AI models. This move helps the company work around U.S. restrictions on advanced chip exports to China. DeepSeek has invested 30 billion yuan to build an 8-trillion-parameter model. The shift is also seen as a boost for Huawei's chip business.
DeepSeek plans wider use of Huawei chips for AI
DeepSeek founder Liang Wenfeng said the company will use more domestically made chips for AI training. Huawei chips are expected to arrive by late 2026 or early 2027. Liang called this one of the company's biggest strategic bets. Training a 2-trillion-parameter model requires about 60,000 Huawei Ascend chips, compared to 30,000 Nvidia H100/H200 chips.
DeepSeek chooses Huawei chips over Nvidia for AI
DeepSeek is widely adopting Huawei chips for AI model training instead of Nvidia. CEO Liang Wenfeng said Huawei's chip performance will catch up with Nvidia within a few years. The company is currently training a 2-trillion-parameter model and plans to build an 8-trillion-parameter model. Training large models with Huawei Ascend 950 processors requires far more chips than Nvidia alternatives.
Most Americans worry AI firms are not doing enough
A Reuters/Ipsos poll found that 73% of Americans believe AI companies are not doing enough to prevent serious harm. The online poll of 1,277 U.S. adults was conducted over four days ending September 22. About 39% of respondents said AI is having a negative impact on society. Some 55% of people also said it would be good to slow down AI development.
Americans say AI companies need to do more
A Reuters/Ipsos poll of 1,004 adults found that 73% of Americans think AI companies are not doing enough to prevent harm. The poll was conducted September 15-17 and had a margin of error of 3 percentage points. About 64% of respondents believe AI will have a major impact on the job market. Only 21% said AI companies are doing enough to prevent harm.
Salt Security Adds Native AI Detection to Agentic Platform
Salt Security launched native AI Detection and Response within its Agentic Security Platform. The new feature provides real-time protection against prompt injection, jailbreak attempts, and unsafe model behavior. It helps security teams see how attacks move from AI models to downstream tools and APIs. The capabilities are available now in Salt AG-DR.
Salt Security Adds AI Detection and Response to Platform
Salt Security introduced native AI-DR capabilities to its Agentic Detection and Response platform. The update protects LLM interactions and maps attacks across agents, MCP servers, and APIs. A survey of over 300 security leaders found that 66% saw API growth above 50% in the past year. Only 18% felt extremely confident in detecting generative AI attacks. The new features work alongside existing AI guardrails and security products.
Generative AI and Microdramas Highlight Busan Market
The Asian Contents & Film Market in Busan will feature generative AI and microdramas as major themes. Thailand is the Country of Honor, and Japan is the Focus Country. The 21st ACFM runs Oct. 10-13 as part of the 31st Busan International Film Festival. The previous edition drew 1,222 companies and over 30,000 attendees. Seven sections and 11 InnoAsia programs will span four days across eight venues.
Missouri's Hawley Supports Federal AI System Testing
Missouri Senator Josh Hawley voiced support for federal testing of advanced AI systems. He said the Trump administration has been testing AI models and agents for national security concerns. Hawley added that Congress should help fund and structure AI oversight. The comments came as experts warn that powerful AI systems could become hard to control. President Trump and Chinese President Xi Jinping are expected to discuss AI later this week.
AI Backlash Puts $68 Billion in Data Center Projects on Hold
About 45 data center projects worth $68 billion were blocked or delayed by local opposition between April and June. Data Center Watch reported 843 opposition groups across 49 states. Some 30 statehouses introduced or adopted rules on data center siting, electricity, and water use. The first quarter of 2026 saw 75 disrupted projects worth about $130 billion. The research group also tracked data center protests outside the U.S. for the first time.
AI handles routine work but human judgment becomes more valuable
As AI takes over routine tasks in customer support, the value of human judgment may actually increase. The article argues that meeting customers where they want to be requires more than automation. Human decision-making and empathy remain essential as AI handles simpler duties.
AI investment drives unexpected UK growth despite global tensions
The UK economy expanded by 0.4% in July, driven by a surge in artificial intelligence investment that exceeded economists' predictions. The International Monetary Fund upgraded its 2026 UK growth forecast to 1% from 0.8%. However, the Iran conflict has closed the Strait of Hormuz, spiking oil prices and threatening future growth. Rising energy costs are expected to squeeze household incomes and business finances in the coming months.
European AI spending projected to reach nearly $470 billion by 2030
IDC forecasts European AI spending will reach nearly $470 billion by 2030, growing at a compound annual rate of 35%. Agentic AI is the main driver, with companies moving from single-purpose copilots to multiple agents working together. Banking is the largest spending sector, while healthcare providers grow fastest at 41.0% per year. The NHS plans to scale AI ambient scribing to 20,000 clinicians. IDC identifies risks including fragmented EU regulations, AI talent shortages, and cloud cost pressures.
Trump Nvidia deal makes Armenia a surprising AI hub
A deal between the Trump administration and Nvidia aims to establish Armenia as a major hub for AI development and deployment. Armenia offers low costs, a skilled workforce, and a favorable business environment with tax incentives for foreign investors. The agreement is expected to create thousands of new jobs and inject billions into the local economy. Armenia also plans to open a new AI research center in Yerevan focusing on healthcare, finance, and education.
Sources
- DeepSeek Shifts to Huawei Chips for AI Model Training
- DeepSeek to use more domestic chips like Huawei for AI training
- DeepSeek to use Huawei chips widely for AI training over Nvidia
- Three out of four Americans say AI firms not doing enough to prevent disaster, Reuters/Ipsos poll finds
- Three out of four Americans say AI firms not doing enough to prevent disaster, Reuters/Ipsos poll finds
- Salt Security Extends Its Agentic Security Platform with Native AI Detection and Response
- Salt Security adds native AI detection and response to agentic security platform
- Generative AI, Microdramas to Take Center Stage at Busan Market
- Missouri's Hawley Backs Federal Testing Of Advanced AI Systems as Safety Concerns Grow
- AI backlash puts $68 billion in U.S. data center projects on hold
- As AI takes on more routine tasks, the value of human judgment may actually increase
- AI Investment Drives Surprise UK Growth, but Geopolitical Storm Clouds Gather
- European AI spending is on track to reach nearly $470 billion by 2030
- Trump’s Nvidia Deal Turns Armenia Into Surprising AI Hotspot
Comments
Please log in to post a comment.