Rule, Britannia? – Nay! – Rule Nvidia!!
2023-11-29 - Nvidia's Game to Lose – Barron’s
Every major technology company is chasing Nvidia.
Last week, the semiconductor company posted stellar financial results. Revenue in its latest quarter nearly tripled, with the company citing surging demand for its chips that enable artificial intelligence applications.
This year, developers have been clamoring for the company’s GPUs, or graphics processing units. They're well suited for the parallel computations needed for AI projects, including large language model training and inference, the process of generating answers from those AI models.
Rivals are racing to compete against Nvidia. Earlier this month, Microsoft unveiled its in-house designed Azure Maia AI Accelerator chip, which is scheduled to be rolled out early next year. On Tuesday, Amazon announced the next version of its Trainium AI chip. Advanced Micro Devices, Intel, and Google are actively working on improved products.
It's going to be an uphill battle for them all. Jefferies analyst Mark Lipacis analyzed the September AI workloads from the six top cloud computing companies and found that Nvidia had an 86% share -- a figure that hasn’t changed much over the past year.
He tracked Alibaba Aliyun, Amazon Web Services, Microsoft Azure, Google Cloud Platform, Oracle Cloud, and Tencent Cloud.
There are several reasons why customers don’t want alternatives to Nvidia’s chips, even when they face a long wait to receive their orders.
First, Nvidia has the most mature technology offering for AI. The company has spent over a decade fixing software and driver issues for its software programming ecosystem, CUDA. It means the company has already fixed technical issues that other less experienced vendors may still need to iron out.
Second, Nvidia is cloud-agnostic. Customers have the flexibility to take their Nvidia-powered workloads from one cloud to another. Rival AI chip offerings from Amazon or Google, on the other hand, lock users into their cloud platforms. That reduces flexibility to switch to another provider offering a cheaper service or better technology.
Third, developers stick with Nvidia because of its decades of platform stability, large market share, access to industry specific tools, and its reputation for backward compatibility.
“All the invention of technologies that you build on top of Nvidia accrue,” Jensen Huang, the CEO of Nvidia, said last week.
Then there’s performance. Nvidia still offers the best overall capability when customers assess the company's combination of software, systems hardware, and networking hardware.
Ultimately, developers want the technology that empowers them to build the best AI applications as fast as possible with the fewest technical risks.
It's Nvidia's game to lose. |
