In a bid to challenge $NVIDIA (NVDA.US)$'s AI dominance, Asian startups are developing more energy-efficient and cost-effective chips for specific artificial intelligence applications.
What Happened: Asian startups are challenging Nvidia's AI dominance by developing more energy-efficient and cost-effective chips for specific AI applications. These startups are targeting the gap in the market left by Nvidia's high energy consumption and bulky design, reported Nikkei Asia on Friday.
These startups are focusing on two types of AI chips: "inference" chips, used to operate existing AI models, and "training" chips, high-powered data-processing components used to develop new AI models.
While Nvidia's GPUs continue to dominate the AI landscape, the startups believe that their GPUs' high energy consumption and bulky design leave a gap in the market that they can fill.
These startups believe that Nvidia's GPUs, while powerful, are too energy-intensive and expensive for many applications. Preferred Networks (PFN) CEO Toru Nishikawa stated, "No one has come up with the perfect chip architecture for inference." PFN is developing chips that aim to be more efficient and less costly than Nvidia's offerings.
Nvidia's GPUs are primarily used for training AI models, but their high cost and energy consumption make them impractical for devices like laptops and wearables. Analysts, including Kazuhiro Sugiyama from Omdia, believe that the demand for on-device AI will rise, encouraging new entrants to the market.
Startups such as Edgecortix, led by Sakyasingha Dasgupta, are focusing on solving issues like the "memory wall" problem to create more streamlined and energy-efficient AI chips. These efforts are part of a broader strategy to cater to the growing demand for AI in industrial applications and robotics, particularly in Asia, according to the report.
"Nvidia's GPU is mainly suited for training, but we are seeing more newcomers developing chips which can target both training and inference," Sugiyama said.
Other companies entering the market include U.S.-based SambaNova Systems, backed by $SoftBank Group (9984.JP)$'s Vision Fund; Tenstorrent, founded by a former $Intel (INTC.US)$ engineer; and the British company Graphcore, recently acquired by $SoftBank (94345.JP)$.
Big tech companies like $Alphabet-C (GOOG.US)$, $Meta Platforms (META.US)$, and $Amazon (AMZN.US)$ Web Services are also joining in, along with Nvidia's rival $Advanced Micro Devices (AMD.US)$.
Why It Matters: The competition between Nvidia and emerging Asian startups is heating up as the AI chip market continues to expand. Recently, Eric Schmidt, former CEO of Google, highlighted Nvidia as a major player in the AI sector, noting that large tech companies are planning significant investments in Nvidia-based AI data centers, potentially costing up to $300 billion.
Meanwhile, SoftBank has faced setbacks in its efforts to rival Nvidia with its own AI chip production. Negotiations with Intel reportedly fell through due to Intel's inability to meet production demands, leading SoftBank to turn to $Taiwan Semiconductor (TSM.US)$, a key Nvidia supplier.
This story was generated using Benzinga Neuro and edited by Kaustubh Bagalkote
为了挑战人工智能领域的主导地位,亚洲初创公司正在开发更具能源效率和成本效益的芯片,用于特定人工智能应用。 $英伟达 (NVDA.US)$其他进入市场的公司包括得到Vison基金支持的美国公司SambaNova Systems,由前Nvidia工程师创立的Tenstorrent和最近被英伟达收购的英国公司Graphcore。与此同时,大型科技公司如亚马逊网服务也加入了研发行列,而英伟达的竞争对手AI芯片公司也在加入。
亚洲初创公司正在开发更节能、成本更低的人工智能芯片,挑战英伟达的市场主导地位。据日经亚洲周五报道,这些初创公司瞄准了英伟达能耗高、设计笨重的市场空缺。
这些初创公司的主要关注点是两类人工智能芯片:用于操作现有人工智能模型的“推论”芯片和用于开发新的人工智能模型的高功率数据处理组件“训练”芯片。
虽然英伟达的GPU继续主导人工智能行业,但这些初创公司认为他们的GPU的高能耗和笨重设计留下了市场空缺。
这些初创公司认为英伟达的GPU虽然功能强大,但对许多应用程序来说能耗过高、价格昂贵。Preferred Networks(PFN)的CEO西川亨表示:“没有人想出完美的推论芯片架构。”PFN正在开发的芯片旨在比英伟达的产品更高效、更经济。
英伟达的GPU主要用于训练人工智能模型,但其高成本和能源消耗使其在笔记本电脑和可穿戴设备等设备上不切实际。包括Omdia的Kazuhiro Sugiyama在内的分析师认为,在设备端人工智能的需求将会上升,这将鼓励新进入市场的公司。
Edgecortix等初创公司正在努力解决“内存墙”问题,开发更流畅、更节能的人工智能芯片。据报道称,这些努力是为了满足亚洲工业应用和机器人市场对人工智能需求的增长。
“英伟达的GPU主要适用于训练,但我们看到越来越多的新进公司开发了既能用于训练又能用于推论的芯片。”Sugiyama说。
与此同时,软银在努力开发自己的AI芯片生产,但一直处于不利状态。根据报道,由于英特尔无法满足产量需求,与Intel的谈判失败,导致SoftBank转向作为英伟达重要供应商的其他公司。 $软银集团 (9984.JP)$的Vision基金; Tenstorrent由前工程师创立; 英国公司Graphcore最近被Web服务收购,随后Nvidia的竞争对手也加入进来了 $英特尔 (INTC.US)$ 工程师;英国公司Graphcore最近被亚马逊Web服务收购, $软银第1期债券型优先股 (94345.JP)$.
大型科技公司如… $谷歌-C (GOOG.US)$, $Meta Platforms (META.US)$和$亚马逊 (AMZN.US)$ ,以及英伟达的竞争对手 $美国超微公司 (AMD.US)$.
为了满足人工智能芯片市场的不断扩张,英伟达和新兴的亚洲初创公司之间的竞争日趋激烈。前Google CEO Eric Schmidt最近将英伟达列为人工智能行业的主要参与者,并指出大型科技公司计划在基于英伟达的人工智能数据中心上投入重金,可能达到3000亿美元的投资总额。
与此同时,软银在努力开发自己的AI芯片生产,但一直处于不利状态。根据报道,由于英特尔无法满足产量需求,与Intel的谈判失败,导致SoftBank转向作为英伟达重要供应商的其他公司。 $台积电 (TSM.US)$与此同时,软银在努力开发自己的AI芯片生产,但一直处于不利状态。根据报道,由于英特尔无法满足产量需求,与Intel的谈判失败,导致SoftBank转向作为英伟达重要供应商的其他公司。
这个故事是使用Benzinga Neuro生成的,并由Kaustubh Bagalkote编辑