Jensen Huang's Departure Shakes Customer Confidence as AMD and Google Chip Away at NVIDIA's Market Share

Deep News07:41

Even before Jensen Huang stepped down as NVIDIA's chief executive last week, AI chips from smaller rivals like AMD had already begun winning over users. Huang's exit will likely reinforce that shift in market sentiment.

David Driggers, CEO of cloud provider Cirrascale, which has 16 years of industry experience and operates 25 data centers, said that last year many customers inquired about using AMD's dedicated AI chips to handle artificial intelligence tasks. He said AMD chips are especially well suited to running AI models, commonly known as "inference," a business that is becoming increasingly widespread among enterprises. "In the company's first two years, convincing customers that AMD was a viable option was like chewing glass," said Piotr Tomasik, chief operating officer of Tensorwave. Tensorwave is an emerging cloud service provider backed by AMD whose platform runs AMD chips. Last year, the situation turned around as external research showed that AMD chips matched or even surpassed NVIDIA's in certain benchmarks.

For years, Google, AMD, Amazon, and numerous emerging chip startups have tried to shake NVIDIA's dominance in the graphics processing unit (GPU) market. Those efforts are finally showing results; on one hand, rival chips have improved in performance and cost less, while on the other, NVIDIA's chips remain expensive and hard to obtain. Since the second half of 2025, OpenAI, Anthropic, and Meta have all said they will adopt AMD chips to run their respective artificial intelligence models. Earlier this week, OpenAI also announced it would use TensorWave's cloud services. Following reports of multiple extramarital affairs involving Huang, he stepped down from the NVIDIA CEO role last week; that change likely weakened NVIDIA's defenses. The company's board has appointed NVIDIA chief financial officer Collette Kress as interim chief executive to succeed Huang.

Of course, from a business standpoint, NVIDIA still holds a commanding lead over AMD and other rivals. Financial results released Tuesday evening local time showed NVIDIA's third-quarter revenue for the fiscal year reached $115.8 billion; that figure is more than eight times AMD's third-quarter revenue announced on November 3. History shows that successors to well-known leaders sometimes achieve even better results than their predecessors. For example, Tim Cook grew Apple into a far larger company than it was under Steve Jobs. But Daniel Svonava, CEO of Superlinked, a company specializing in open-source model inference services, said that after Kress's appointment, his confidence that NVIDIA would devote sufficient resources to increasing chip supply declined. Svonava said: "CFOs are usually not aggressive in capital allocation, and easing the compute shortage requires precisely aggressive capital investment. This latest news from NVIDIA completely cuts off our ability to move from the 'experimental stage' to 'actual deployment.'"

Cost-effectiveness

Svonava's company had previously tested AMD chips for months; in recent months, as companies like DeepSeek released small open-source AI models, they began taking AMD chips more seriously. He said AMD chips are very well suited to running such models. Compared with NVIDIA's products, AMD chips are priced much lower; for Superlinked, this means using AMD chips to run certain models can generate higher profit margins. According to Tomasik, AMD has also successfully improved its ROCm software—the software developers use to write AI models that can run on AMD chips. ROCm is AMD's competitor to NVIDIA's CUDA software; CUDA's widespread adoption among programmers has long been a major NVIDIA advantage. Before these improvements, TensorWave had dedicated staff to help customers navigate AMD's software. Tomasik said that compared with last year, the time employees spend helping customers overcome ROCm-related difficulties has fallen by 80% to 90%. These ROCm improvements may give AMD more opportunities to challenge CUDA's position.

Market attention on AMD is helping cloud service companies that operate on AMD chips. TensorWave expects its annualized recurring revenue (ARR) to reach $210 million by the end of 2026; by comparison, that figure was $125 million at the end of 2025. Hassan Akbari, CEO of Sciforium, another cloud service provider focused on the AMD platform, said that since signing its first commercial contract in June, the company's customer base has tripled; over the past year, its compute scale has grown fivefold, and its team has nearly tripled. Akbari revealed that this startup, which has received investment from AMD, is currently raising more funds to further expand its compute resources.

Google

Google has been investing heavily to expand the market for its TPU artificial intelligence chips; this includes financing support for building data centers outside Google Cloud—data centers that will use Google's TPU chips. In July this year, Google disclosed that it had agreed to provide guarantees totaling up to $44 billion for data center leases (so-called "lease backstop" commitments); if tenants default, Google would bear the corresponding responsibility. Cirrascale recently became one of the first service providers to offer Google TPU compute outside the Google Cloud ecosystem. Driggers said customers have already expressed interest in spending hundreds of millions of dollars on computing services based on Google's TPU 8 chips. A growing number of companies that rely on NVIDIA chips are turning to Google's TPUs (tensor processing units). For example, last month it was reported that Google would sell its TPU chips to OpenAI; on the same day, Oracle also announced it would offer such chips to its customers. In addition, Google said it has updated its software to make it easier for customers to use TPUs. Chad Agate, co-founder and chief technology officer of autonomous delivery platform Autolane, revealed that the company previously mainly used GPUs on Google Cloud, but since September it has begun migrating some workflows (such as evaluation tasks) to run on TPUs. Agate said that given TPUs' lower cost, the company is now considering handing heavier and more complex workloads to TPUs as well. "We're considering migrating more workloads, and the outlook is promising," he said, noting that the company still relies heavily on GPUs. "We're not abandoning (GPUs) entirely; we're evaluating which workloads are suitable for migration."

Compared with NVIDIA's chips, TPUs still lag far behind. Vadim Solovey, CEO of cloud software company DoiT, said that among the company's roughly 5,000 customers, only 16 use TPUs "at scale." "These chips are a second supplier, not a replacement."

NVIDIA's advantages

NVIDIA still has many advantages that competitors must contend with. As previously reported, thanks to its massive production scale, NVIDIA can secure priority access to manufacturing capacity at TSMC during periods of tight supply—TSMC is the dominant manufacturer in the AI chip sector. According to some customers, it is precisely because of this partnership that NVIDIA's chip supply has remained ample. One startup founder said they had discussed renting AMD's latest chips earlier this year but were told they could not get them until the end of the year at the earliest. The founder said that because they could obtain comparable NVIDIA chips faster, they ultimately gave up on using AMD's products. Michael Reid, CEO of cloud service provider Megaport, said that about three or four months ago, a customer asked to use AMD chips to run AI workloads—mainly because NVIDIA chips were hard to obtain at the time. Reid noted that if they chose AMD chips, the customer would have to wait six months for delivery; at that time, NVIDIA chips had a delivery cycle of only 12 to 14 weeks. As a result, the customer ultimately decided against using AMD chips.

In addition, NVIDIA has built a powerful sales and marketing system whose depth far exceeds traditional supplier relationships. Industry executives point out that this is a major NVIDIA advantage that competitors, including AMD, still find difficult to match. "In my career, NVIDIA is the most proactive supplier I've worked with," Reid said. "Among all the vendors I've worked with, only NVIDIA will really help you connect with customers—for example, connecting you with people in finance or helping you find data center resources. Other vendors do nothing." Driggers said his customers continue to place more orders for NVIDIA chips. Even so, he noted: "NVIDIA still dominates, but... their market share is eroding."

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap
Leave a comment