Almost a day goes by without hearing about some AI company whose market value is skyrocketing. Earlier this month, the stock price of Dell, a hardware manufacturer, jumped by more than 30 percent in a single day due to hopes that AI technology will boost sales. A few days later, Together AI, a cloud computing startup, raised funds at a valuation of $1.3 billion, up from $500 million in November. One of the investors in Together AI is Nvidia, the AI chip manufacturer that is also on a significant rise. And for probably the largest rise, of course, is responsible Open AI and its ChatGPT.
When it was introduced in November 2022, its market capitalization was around 300 billion dollars, while today it stands at 2.3 trillion dollars, which is about 300 billion dollars less than Apple, writes The Economist. The increasing number of various AI companies and those that present themselves as such makes it difficult to assess the true winners in this boom of the artificial intelligence market.
Early Days
Technological advances usually elevate new tech giants. The computer boom in the 1980s and 1990s brought Microsoft, which created the Windows operating system, and Intel, which produced the chips needed to run it, to the top of the corporate hierarchy. By the 2000s,’Wintel’ captured four-fifths of the operating profit from the computer industry, according to investment bank Jefferies. The smartphone era did the same for Apple. Just a few years after launching the iPhone in 2007, it accounted for more than half of the global operating profit of mobile phone manufacturers.
Now the world is still in the early days of the generative AI era, but even in these early days, the AI business has become very lucrative. Around a hundred AI companies analyzed by The Economist, both large and small, have already created a value of $8 trillion from October 2022 to today.
However, some are more successful than others, while the greatest value is concentrated in a handful of leading companies. In the hardware, model development, and application sectors, the three largest companies have increased their share of total value by a median of 14 percentage points in the last year and a half. In the cloud computing sector, Microsoft, which has a partnership with ChatGPT’s producer, OpenAI, has pulled ahead of Amazon and Alphabet. Its market capitalization now accounts for 46 percent of the total ‘cloud trio’, compared to 41 percent before the release of ChatGPT.
Wealth of Hardware Manufacturers
The most ‘wealth’ has gone to hardware manufacturers. This group includes chip manufacturing companies (such as Nvidia), server manufacturers (Dell), and those producing networking equipment (Arista). In October 2022, 27 public hardware companies analyzed by The Economist were valued at around $1.5 trillion. Today that figure stands at $5 trillion. This is what you would expect in a tech boom: first, you need to build the underlying physical infrastructure to offer software. In the late 1990s, as the internet boom was growing, hardware providers like modems and other telecommunications products, such as Cisco and WorldCom, were early winners.
The far biggest winner is undoubtedly Nvidia, which accounts for about 57 percent of the increase in the market capitalization of hardware companies. The company produces more than 80 percent of all AI chips, according to research firm IDC. It also enjoys almost a monopoly in the networking equipment used to connect chips within AI servers in data centers. Revenue from Nvidia’s data center business has more than tripled in the 12 months ending January, compared to the previous year. Its gross margins have increased from 59 to 74 percent.
Nvidia’s chip manufacturing competitors want a piece of that wealth, and established ones, such as AMD and Intel, are launching competing products. Nvidia’s largest customers, the three cloud giants, are also designing their own chips—as a way to reduce reliance on a single service provider and to steal some of Nvidia’s juicy margins for themselves. Lisa Su, CEO of AMD, predicted that revenues from AI chip sales could grow to $400 billion by 2027, up from $45 billion in 2023. ‘That would be too much for Nvidia to digest alone,’ Su believes.
Increasing AI Applications
As AI applications become more widespread, an increasing share of that demand will shift from chips needed for training models, which consist of analyzing mountains of data to teach algorithms to predict the next word or pixel in a sequence, to those actually needed to use them for responding to queries (‘inference’, in tech language).
In the past year, about two-fifths of Nvidia’s AI revenue came from customers using its chips for inference. Experts expect that some inference will begin to shift from specialized graphics processing units (GPUs), which are Nvidia’s strong suit, to general-purpose central processing units (CPUs) like those used in laptops and smartphones, dominated by AMD and Intel. Soon, even some training could be done on CPUs instead of GPUs.
Nevertheless, Nvidia’s grip on the hardware market seems secure for the next few years. Startups without a track record will find it hard to convince large clients to reconfigure corporate hardware systems for their new technology. The introduction of proprietary chips by cloud giants is still limited. And while hardware is winning the race for value increase in absolute terms, independent model manufacturers have enjoyed the largest proportional gains. The combined value of 11 such companies that The Economist observed jumped from $29 billion to about $138 billion in the last 16 months. OpenAI is estimated to be worth around $100 billion, up from $20 billion in October 2022. The valuation of Anthropic rose from $3.4 billion in April 2022 to $18 billion. Mistral, a French startup founded less than a year ago, is now valued at around $2 billion.
Part of that value is tied to hardware. Startups buy piles of chips, mostly from Nvidia, to train their models. Imbue, which like OpenAI and Anthropic is based in San Francisco, has 10,000 such chips. Cohere, a Canadian rival, has 16,000. These semiconductors can sell for tens of thousands of dollars each. As models become more sophisticated, they require more and more, and GPT-4 reportedly costs around $100 million to train. Some doubt that training its successor could cost OpenAI ten times more.
Intellectual Property
However, the true value of model manufacturers lies in their intellectual property and the profit it can generate. The actual scope of that profit will depend on how fierce the competition among model providers becomes—and how long it lasts.
The abundance of models has also enabled the growth of companies working on AI applications. The value of 19 public software companies from The Economist’s analysis has jumped by $1.1 trillion, or 35 percent, since October 2022. This includes major software providers adding generative art to their services. Zoom uses the technology to allow users to summarize video calls. ServiceNow, which provides technical, HR, and other support to companies, has introduced chatbots to help resolve customer IT inquiries. Adobe, the maker of Photoshop, has an application called Firefly that uses AI for image editing.
And newcomers add more diversity. For example, the website ‘There’s An AI For That’ lists over 12,000 AI applications, compared to fewer than 1,000 in 2022.
Then there are the cloud companies. The combined market capitalization of Alphabet, Amazon, and Microsoft has jumped by $2.5 trillion since the beginning of the AI boom, but actual profits are even higher, so many believe that these three companies will be the long-term winners of the AI frenzy.
