Yazılar

Huawei outlines chip roadmap, challenges Nvidia with AI supernodes

/in /tarafından

Huawei unveiled its long-term semiconductor strategy at the Huawei Connect conference in Shanghai, presenting detailed timelines for its Ascend AI chips, Kunpeng server processors, and next-generation computing systems. The move signals China’s determination to reduce reliance on foreign chipmakers like Nvidia while intensifying the U.S.-China tech rivalry.

Rotating chairman Eric Xu announced that Huawei will release new Ascend chips annually, doubling compute power with each iteration. The company launched its Ascend 910C earlier this year and plans to follow with the Ascend 950 in 2026, the 960 in 2027, and the 970 in 2028. Alongside, Huawei is preparing high-performance “supernodes” designed to interconnect thousands of chips at high speed. The Atlas 950 system, expected in late 2026, will link 8,192 Ascend chips, while the Atlas 960 in 2027 will support 15,488 chips—leaps over the existing Atlas 900.

Huawei also revealed it has developed proprietary high-bandwidth memory (HBM), a field dominated by South Korea’s SK Hynix and Samsung, strengthening its position in advanced computing. Updated versions of its Kunpeng server chip are scheduled for 2026 and 2028.

The timing of the announcement coincides with escalating tensions: Chinese regulators have accused Nvidia of antitrust violations and ordered local firms to halt purchases of its AI chips. Analysts say Huawei’s show of strength reflects growing confidence that U.S. export controls will not derail China’s domestic chipmaking progress.

Despite Huawei’s advances, engineers acknowledge Nvidia’s chips still outperform Chinese alternatives. However, Huawei is betting on leveraging China’s networking and power infrastructure advantages to offset manufacturing gaps and push large-scale systems forward.

Chinese semiconductor stocks rallied after reports of Beijing’s restrictions on Nvidia sales. The geopolitical backdrop looms large, with President Xi Jinping set to meet U.S. President Donald Trump amid renewed trade negotiations.

Microsoft expands Wisconsin AI data center investment to $7 billion

/in /tarafından

Microsoft announced plans to build a second major artificial intelligence data center in Wisconsin, raising its total investment in the state to more than $7 billion. The new $4 billion facility will join a $3.3 billion data center already under construction in Mount Pleasant, Racine County, first unveiled last year.

The initial data center is expected to open in 2026 and employ about 500 people, while the addition of the second will expand staffing to around 800. Microsoft says the combined site will eventually host the world’s most powerful AI supercomputer, linking together hundreds of thousands of Nvidia chips.

The development comes on land once earmarked for Foxconn’s highly publicized $10 billion factory, a project dramatically scaled back after initial political fanfare during Donald Trump’s presidency. When President Joe Biden attended Microsoft’s first announcement last year, he highlighted Foxconn’s retreat as a cautionary tale while framing Microsoft’s plan as a sign of renewed investment.

Microsoft said it will pre-pay for electrical infrastructure to avoid burdening local customers with higher power bills and will use Wisconsin’s cold climate for energy-efficient cooling. Annual water consumption will be capped at roughly the level of an average restaurant. To offset its energy use, the company will also build solar power elsewhere in Wisconsin, though Microsoft President Brad Smith noted new fossil fuel generation—specifically liquefied natural gas—will still be part of the mix.

While the 800 permanent jobs fall short of the thousands promised by Foxconn, Smith emphasized the importance of ongoing skilled labor positions, including pipefitters and electricians, needed for both construction and long-term maintenance.

Nvidia invests $900 million to acquire Enfabrica talent and technology

/in /tarafından

Nvidia has reportedly spent more than $900 million to bring Enfabrica’s CEO, Rochan Sankar, and other staff into the company, while also securing a license for Enfabrica’s technology, according to CNBC. The deal, a mix of cash and stock, closed last week, and Sankar has already taken up his new role at Nvidia.

Enfabrica, a Silicon Valley chip startup founded by former Broadcom and Alphabet engineers, specializes in solving one of AI’s biggest bottlenecks: interconnecting massive numbers of chips efficiently. Its networking technology allows around 100,000 AI chips to work together as if they were one computer—minimizing costly downtime caused when processors wait for data to move across networks.

The startup had previously raised $260 million in venture capital and in July unveiled a chip-and-software system designed to reduce memory chip costs in large-scale AI data centers. Nvidia’s acquisition echoes a recent trend of tech giants pulling in specialized startups and their leaders to strengthen AI infrastructure. Meta recently took a 49% stake in Scale AI while elevating its CEO Alexandr Wang to a strategic role, and Google hired top staff from AI code generation startup Windsurf after OpenAI attempted to acquire it.

Neither Nvidia nor Enfabrica has publicly commented on the reported deal.