Nvidia has become synonymous with AI processors thanks to its cutting-edge blueprint for graphics processing units (GPUs) and rapid innovation at Taiwan Semiconductor Manufacturing, a contract manufacturing company that produces 90% of the world's advanced AI chips.
That's starting to change. New entrants into the AI chip design business, including Google and Amazon, are talking about selling cutting-edge chips that rival Nvidia's GPUs in power and efficiency to large numbers of external customers.
Smaller rivals such as Advanced Micro Devices, Qualcomm and Broadcom are introducing products to focus more on AI data center computing. Even some of Nvidia's biggest customers, including ChatGPT makers OpenAI and Meta Platforms, have begun designing their own custom chips, posing new challenges to the company's ubiquity.
While Nvidia is unlikely to experience a mass exodus of customers, the AI company's efforts to diversify its suppliers could make it harder for the market leader to generate the top-notch sales growth that investors are used to seeing.
The landscape is changing rapidly. It seems like every week brings a new big technology infrastructure deal or the release of a new generation of powerful AI chips. Below is an overview of the major companies vying for a position in the rapidly growing AI chip market.
top dog
Nvidia's dominance in AI computing power has made it the world's most valuable company and catapulted its leather-jacketed CEO Jensen Huang to celebrity status. Investors will be parsing every word from Huang and will be watching the company's quarterly profits as a barometer for the overall AI boom.
Nvidia likes to describe its business as more than just chips, emphasizing that it offers “rack-scale server solutions” and calling the data centers that use it “AI factories.” But the base product Nvidia offers, accelerated computing, is the same thing every AI company wants.
From February to October, Nvidia sold $147.8 billion worth of chips, network connections, and other hardware that powers the explosive growth of AI. This is up from $91 billion in the same period last year.
In July, NVIDIA became the first company on Earth to surpass $4 trillion in market capitalization. Five months later, it briefly topped $5 trillion, before concerns of a bubble spread across the AI industry. Nvidia's stock price, like that of most of its rivals, has moved slightly closer to reality. Even with these adjustments, the company is worth more than twice the $1.8 trillion value of its closest competitor, Broadcom.
Nvidia's beginnings were modest. The now legendary company was founded in 1993 by Hwang, Curtis Priem and Chris Malachowski, three friends who were all electrical engineers, to serve up Grand Slam breakfast plates at Denny's in San Jose, California.
Their initial goal was to develop a chip that could produce more realistic 3D graphics for personal computers. Unlike the central processing unit (CPU) that powers most PCs, GPUs are capable of parallel computing and can perform millions or billions of simple tasks at the same time. Nvidia's GPUs were originally used by video game developers, but the company later realized they were ideal for deep learning and AI.
In 2006, Nvidia released CUDA, its own software library that allows developers to use the company's chips to build applications and run them faster. As the AI gold rush took hold, thousands of developers became locked into Nvidia's hardware and software ecosystem.
Nvidia is increasing its pace with each new generation of advanced AI chips. Late last year, the company began shipping its Grace Blackwell series of servers, its most powerful AI processors to date with cutting-edge chips, and they sold out almost instantly. At an October conference in Washington, D.C., Huang said the company will have sold 6 million Blackwell chips by 2025, with orders for 14 million more, representing a total of $5 trillion in sales.
Challenges still remain. Nvidia has been effectively banned from selling its chips in China for the past three years, a problem because Huang claims the rival superpower is home to half of the world's AI developers. Without the billions in sales from Chinese customers, the company's growth is likely to be constrained and China's tech sector accustomed to working with domestically produced chips instead.
Nvidia is also currently facing increasing pressure domestically.
Show full image
rival designers
AMD made a significant policy change three years ago to set up a classic David vs. Goliath challenge against Nvidia.
As it became clear that demand for advanced AI processors was surging, AMD CEO Lisa Su told the board that the company plans to reorient the entire company around AI. She predicted that the “insatiable demand for computing” will continue. The bet has paid off handsomely so far, with AMD's market cap nearly quadrupling to more than $350 billion, and the company recently inked a major deal to supply chips to OpenAI and Oracle.
Another chip design company, Broadcom, once part of Hewlett-Packard, has also emerged as a strong competitor. A series of mega-mergers has expanded the company into a $1.8 trillion leviathan. Broadcom now makes custom chips called XPUs designed for specific computing tasks and networking hardware that helps data centers string together huge racks of servers.
Intel, originally one of the giants of Silicon Valley, is in trouble. The company largely missed the AI revolution due to a series of strategic mistakes, but recently it has invested heavily in both its design and manufacturing businesses and is attracting customers for advanced data center processors.
Qualcomm, known for designing chips for mobile devices and cars, saw its stock rise 20% after announcing in October that it would launch two new AI accelerator chips. According to the company, the new AI200 and AI250 feature extremely high memory capacity and energy efficiency.
Show full image
giant intruder
Competition has intensified in recent weeks. Armed with large sums of money from other business areas, Alphabet Inc.'s Google unit and Amazon.com's cloud-computing arm Amazon Web Services are investing in AI chips, which are also seeing increasing demand from third-party customers.
For more than a decade, Google has designed and used chips known as tensor processing units (TPUs) internally. The company first offered them to third parties in 2018, but they weren't widely sold to large customers for several years. Currently, major companies such as Meta, Anthropic, and Apple are purchasing or negotiating access to TPUs to train and run their models.
In late November, Dylan Patel, founder of influential AI infrastructure consulting firm Semianalysis, mused that the growing popularity of Google's chips could mean “the end of Nvidia's dominance.”
Meanwhile, Amazon is expanding its Anthropic data center cluster, which will eventually include more than 1 million of Amazon's Trainium chips, and AWS has just begun widespread sales of chips that it says are faster and use much less energy than their Nvidia counterparts.
Show full image
DIY person
Even Nvidia's customers are beginning to erode that advantage by developing their own application-specific integrated circuits (ASICs). Co-designed by AI companies and big silicon companies, this class of chips is optimized for highly specialized computing tasks.
OpenAI and Broadcom recently entered into a multibillion-dollar partnership to develop custom chips to address the computing needs of ChatGPT makers. A few months ago, Meta announced the acquisition of chip startup Rivos to strengthen its in-house AI training chip development efforts.
Microsoft's chief technology officer said in October that the company plans to rely more on its custom accelerator chips in its data center business. And over the summer, Elon Musk's xAI posted a job opening for a chip designer to help “design and refine new hardware architectures” to help train AI models.
Most industry watchers say Nvidia is unlikely to lose its dominant position in the market, with Nvidia arguing that its computing systems are more flexible and have a wider range of uses than custom chips. But it's no longer the only game in town, as demand is growing rapidly.
Email Robbie Whelan at robbie.whelan@wsj.com.
