在人工智能芯片领域，英伟达（Nvidia）是市值3万亿美元的行业巨头。然而，这家巨头在市场上投下的巨大阴影并未吓退众多挑战者，这些挑战者从老牌芯片制造商到新兴初创企业，不一而足。
AMD执行副总裁兼首席技术官马克·帕普马斯特在周二举行的《财富》人工智能头脑风暴大会上表示：“市场需要竞争，而我们正因此获得了巨大的推动力。”
AMD，这家以生产个人电脑微处理器起家的公司，在图形处理器（GPU）设计与销售领域也展现出了强大实力。图形处理器作为一种对人工智能友好的芯片，是英伟达蓬勃发展的业务的核心。尽管AMD的图形处理器起初主要面向视频游戏和娱乐应用程序，但如今已直接瞄准人工智能培训市场。
帕普马斯特说：”我们从2023年人工智能收入几乎为零，到今年预计达到50亿……市场需要竞争，这正是我们关注的重点。”
Cerebras Systems联合创始人兼首席执行官安德鲁·费尔德曼认为，之所以有如此多的参与者涌入，是因为人工智能市场过于庞大。不仅仅是英伟达或AMD，超大规模公司或初创公司也“发现了爆炸式增长机会，希望自研芯片”。
费尔德曼说：“这是由于需求激增。我们在训练领域和推理领域都观察到了这一现象——坦白说，我认为我们所有人都难以跟上这一发展速度。”
费尔德曼继续说，竞争不仅对竞争对手有利，更对消费者大有裨益，因为这意味着他们应用程序中的人工智能将以更快的速度运行，且成本更低。“马克的图形处理器运行速度比英伟达的更快，”他说，指的是AMD的帕普马斯特。他声称：“在很多情况下，我们的加速器运行速度比英伟达的快75倍。”
速度至关重要。费尔德曼提到了宽带出现之前的互联网时代，称那是一场灾难。
他表示：”一旦宽带出现，人们突然间就拥有了新应用程序、流媒体等有趣且参与度高的产品。我认为，人工智能目前正处于这样的发展阶段。随着速度的不断提升，人们将迈入人工智能推理的宽带时代。”
同样值得记住的是，生成式人工智能仍处于早期阶段。Altera首席执行官桑德拉·里维拉表示：“未来，我们将拥有无限的创新机遇。”在里维拉看来，这并不是要与英伟达一较高下，而是要致力于为任何想要完成的任务打造最佳设备，因为芯片并不是能解决所有问题的万能钥匙。例如，随着人工智能模型和应用程序的不断演变，运行人工智能所需的处理能力将驻留在智能手机、家用电器等“边缘”设备及数据中心内部。
她说：“竞争是好事，能让你更敏锐、更专注，诚然，也能满足极其广泛的客户需求。”
市场对人工智能芯片的需求也为投资者带来了机会。Cerebras在9月份提交了首次公开募股文件，但费尔德曼拒绝就上市的具体状况或时间发表评论。里维拉表示，英特尔（Intel）旗下的Altera公司仍计划在2026年上市。英特尔于2015年收购了Altera，该公司一直受到业务方面挑战的困扰，导致首席执行官帕特·格尔辛格（Pat Gelsinger）于上周辞职。
里维拉表示，尽管英特尔内部近期发生了变动，但该公司仍计划出售Altera的股权，并在2026年完成首次公开募股。她说：“我们目前正处于这一进程的中间阶段，市场表现出了浓厚的兴趣，这令人倍感振奋。”她后来补充道：”从诸多角度来看，上周的新闻都令人难过，但我竭力引导公司及员工专注于力所能及之事。”（财富中文网）
译者：中慧言-王芳
In the world of AI chips, Nvidia is the $3 trillion Goliath. But the massive shadow Nvidia casts over the market isn’t scaring off a gang of challengers ranging from longtime chip players to young startups.
“The market demands competition, and so we’re getting a tremendous pull,” said Mark Papermaster, executive vice president and chief technology officer of Advanced Micro Devices, at Fortune’s Brainstorm AI conference on Tuesday.
AMD began decades ago as a maker of PC microprocessors, but has also developed a robust business designing and selling GPUs, the AI-friendly class of chips that are at the heart of Nvidia’s booming business. While AMD’s GPUs were once geared for video gaming and entertainment applications, the company is now aiming straight for the AI training market.
“We went from almost zero in 2023 of AI revenue to five billion projected this year…the market needs competition, and that’s what we’re focused on,” said Papermaster.
Andrew Feldman, co-founder and chief executive of Cerebras Systems, suspects there are so many players because the AI market is so big. It isn’t just Nvidia or AMD, it’s hyperscalers or startups “seeing opportunity explode and wanting to do their own chips.”
“It’s because you have this explosion of demand,” Feldman said. “We’re seeing it on the training side, we’re seeing it on the inferencing side—and I think, truth be told, none of us can keep up.”
Competition isn’t just better for competitors, Feldman continued, it’s better for consumers because it means the AI in their applications will run faster and cost less. “Mark’s GPUs are running faster than Nvidia’s,” he said, referring to AMD’s Papermaster. “Our accelerators are running, in many instances, 75 times faster than Nvidia’s,” he claimed.
Speed matters. Feldman mentioned the days of the internet before broadband, calling it a disaster.
“Once we got broadband, all of a sudden, you had new applications, you had streaming, you had all these things that were fun, and the engagement was high,” he said. “And I think that’s what’s happening right now with AI, is that as you get faster, you move into the sort of the broadband era of AI inference.”
It’s also worth remembering that generative AI is still in its very early stages. “There’s so much innovation ahead of us,” said Sandra Rivera, chief executive of Altera. To Rivera, it isn’t about taking on Nvidia, it’s about creating the best device for whatever you’re attempting to do because chips aren’t one size fits all. As AI models and applications evolve into different versions, for instance, the processing power necessary to run it will reside on “edge” devices like smartphones and appliances, as well as inside data centers.
“Competition is good,” she said. “It makes you sharper, makes you more focused, and it certainly addresses the very, very broad customer requirements that are out there.”
The market demand for AI chips is creating an opportunity for investors too. Cerebras filed paperwork in September for an initial public offering, though Feldman declined to comment on the status or timing for the listing. Rivera said that Altera, which is owned by Intel, is still planning for a floatation in 2026. Intel, which acquired Altera in 2015, has been roiled by business challenges that led to CEO Pat Gelsinger resigning last week.
Despite the changes at Intel, Rivera said that the plan was still for Intel to sell an equity stake in Altera and for a 2026 IPO. “We’re in the middle of that process now with a lot of great interest, which is encouraging,” she said, later adding, “the news from last week was sad on so many levels, but what I try to keep the company and the employees focused on is the things that we control.”