Amazon unveiled its latest AI chips last month in a bid to reduce its dependence on market leader Nvidia and take a share of a multibillion-dollar market.
Central to this effort is the introduction of Trainium 2, Amazon’s newest chip built for training massive AI models. Amazon is hardly alone. A growing cohort of Big Tech companies are eager to challenge the commanding lead of Nvidia in designing cutting-edge AI chips.
Nvidia has been at the forefront when it comes to supplying chips that power large language models, such as the one used by OpenAI’s ChatGPT. Nvidia’s near monopoly has propelled the company’s valuation past $3.4 trillion, leaving competitors including AMD scrambling to close the gap.
In November, Nvidia reported an impressive 94 per cent annual revenue growth for the third quarter, reaching a record $35.1 billion. Questions remain, however: how long can Nvidia stay on top? And how can it do so?
For one thing, the annals of business history remind us just how quickly business advantage can slip. Consider Intel’s once-formidable position in desktop computing – when mobile computing arose, Intel stumbled. And thus, the challenge facing Nvidia’s chief executive Jensen Huang is: how can the company keep growing when it already has the largest market share of AI chips?
Some of Nvidia’s biggest customers, including Amazon, Microsoft and Google, are spending billions of dollars to build their own custom chips. In many ways, Big Tech’s push to unseat Nvidia is a familiar story: develop in-house hardware to reduce reliance on outside suppliers, cut costs and achieve tighter control over one’s own technology.
But overthrowing Nvidia is no small feat, even for these tech giants. They all rely on the same manufacturing partner: Taiwan Semiconductor Manufacturing Company (TSMC), the world’s largest chip manufacturer. Because TSMC produces chips for so many companies, no single rival gains a manufacturing edge over Nvidia.
Furthermore, TSMC’s pricing structure favours those placing larger orders. Companies such as Nvidia benefit from lower per-unit costs, reinforcing an already sizeable advantage.
That’s why, as long as the market depends on high-performance graphics processing units, Nvidia’s dominance remains secure because it leads in both hardware and software innovation.
Let me explain. Founded about 30 years ago with a focus on video games, Nvidia is famous for its cutting-edge processors, but the real game-changer is its software platform, Cuda. This “secret sauce” allows GPUs, originally built for graphics, to make running AI applications much faster and easier.
Nvidia’s hardware is in such high demand because of the software it has built over nearly two decades, a combination that competitors have found nearly impossible to breach – so far.
But what if the advance in AI doesn’t require such cutting-edge GPUs in the future? Will there be a time when upstarts, as well as Big Tech, could be nipping at Nvidia’s heels?
History repeating?
A cautionary tale comes from Intel. The company was once the pride of Silicon Valley for its x86 processors, dominating the market for personal computer chips.
Yet the rise of mobile computing in the early 2000s exposed a painful oversight: Intel clung to high-performance but power-hungry processors, just as demand shifted towards more efficient designs made for smartphones. The result? ARM Holdings, Qualcomm and others swiftly captured the mobile arena, leaving Intel sidelined at a critical turning point in tech.
This provides a stark parallel for Nvidia. Its GPUs, notably the flagship H100, command a premium that can be up to four times higher than AMD’s competing MI300X. That premium is sustainable only as long as top-tier performance remains essential to AI models.
But what if future AI systems demonstrate comparable results with cheaper, lower-tier processors? In that instance, Nvidia’s steep pricing could prove to be a serious liability.
Already, emerging competitors are betting on more economical approaches. They are focusing on cost-effective, “good enough” solutions.
One example is DeepSeek, a Chinese AI developer, which is creating AI models that rival the performance of models such as OpenAI’s GPT-4, some experts say. But it uses advanced techniques to process information faster and use less computer memory.
In other words, DeepSeek is achieving similar performance through limited hardware by tweaking software to get the best performance out of processors. This saves the energy required for computing and keeps costs low.
This “good enough” approach is reminiscent of Chinese smartphone brands including Huawei, Oppo and Xiaomi, which emerged in the early 2010s. They offered affordable devices with decent performance, gradually chipping away at western mobile phone makers’ historic dominance. It’s not inconceivable that Nvidia could face a similar erosion of market share if cost-sensitive solutions become widely adopted.
Intel’s decline – rooted in an insistence on high-performance chips for traditional PCs, even as the industry pivoted to mobile – highlights the peril of failing to adapt.
If Nvidia continues to focus only on top-of-the-line GPUs, it may risk losing relevance one day.
Looking ahead, one promising strategy for Nvidia lies in expanding its cloud-based AI offerings. By selling its services directly to enterprises that do business with Amazon Web Services and Google, Nvidia can tap into additional revenue streams that do not hinge solely on GPUs.
Car makers, drug manufacturers and consumer goods-companies all need AI. Nvidia’s latest foray involves rolling out its AI-enabled cloud services designed to help these businesses build, fine-tune and run custom AI models in their operating environments. It’s the ultimate response when your customer turns into your competitor: you go after your customer’s customer.
This is the lesson about staying on top. Even when you are the most successful, you should already be scaling up the next growth engine. Why? Andy Grove, the renowned former chief executive of Intel, had the answer some 25 years ago: “Success breeds complacency. Complacency breeds failure. Only the paranoid survive.”
