Jensen Huang did not mention a fall in the share price when he spoke last week at the Nvidia Annual General Meeting.
The US chipmaker briefly became the most valuable company in June, but it quickly lost the title. Nvidia lost about $550bn from its $3.4tn peak market value (£2.68tn), as investors combined profit-taking and doubts about rocketing development to apply the brakes.
Huang spoke, however, like the CEO of an organization that has taken 30 days to move from a $2tn valuation to $3tn this year – and is now looking at $4tn.
He said that a group of powerful, new chips called Blackwell could be “the most successful products in our history”, and possibly in the history of computers. He said that automation of $50tn in heavy industry would be the next wave of AI. He described what sounded as an endless loop robotic factories orchestrating robotics that “build robotic products”.
He concluded by saying: “We have reinvented Nvidia and the computer industry, as well as the world.”
This is the kind of language that supports a $4tn value and the AI buzz cycle. Nvidia’s shares have been gaining ground, and this week they returned above $3tn. This is because it is still the best way to invest in the AI boom. Does that mean it will reach $4tn, despite investor doubts?
Alvin Nguyen is a senior analyst with the research firm Forrester. He said that “only the collapse of the genAI markets” could prevent Nvidia reaching $4tn. But whether they got there before their tech rivals, was another question. Microsoft, another major AI player, and Apple currently hold the first and second positions in terms of market share, respectively, with Nvidia coming in third.
Nguyen said that if OpenAI’s GPT-5 and other new AI models are amazing, then the share price will stay buoyant, and could reach $4tn at the end of 2025. If they disappoint, the share price may be affected due to its position as the flag-carrier of the technology. He added that a technological breakthrough might result in less computing power needed to train the models or consumer and business interest in generative AI could be lower than expected.
Nguyen said that there are many unknowns and factors outside of Nvidia’s control that may impact the company’s path to $4tn. “Such things as disappointments with the new models, model improvements which reduce computational requirements, and a weaker demand than expected from consumers and enterprises for genAI.”
The private AI labs, such as OpenAI or Anthropic (the entities behind ChatGPT and Claude chatbots), are not traded on the public market. This means that investors have no access to the major players in the generative AI frenzy.
The cost of buying shares in a multinational company like Microsoft or Google can be high, and the investment only represents a small fraction of what is being invested. Google’s search ad business, for instance, could suffer if there is a huge AI boom.
Nvidia is, in contrast, selling spades during a gold rush. It continues to sell top-end processors faster than it is able to make them, despite years of investment. Nvidia reaps the benefits of huge investments made in AI research.
This chip was sold by the company to allow gamers to enjoy crisp and smooth graphics when playing 3D games. But, through a stroke of luck, it turned out that this type of chip is exactly what researchers needed to create massive AI systems like GPT-4 and Claude 3.5.
GPUs can perform complex calculations at high speeds and volumes, which are essential for the operation and training of AI tools like chatbots. Any company that wants to create or run a generative AI tool, like chatGPT, or Google’s Gemini needs GPUs. It’s the same for free AI models like Meta’s Llama. They also require a lot of chips during their training phase. For systems called large language models, training requires crunching huge blocks of data. The LLM learns to recognize patterns in language, and to predict what the next word or phrase should be in response to the chatbot’s query.
Nvidia never did dominate the AI chip market. Google has relied on TPUs, or “tensors”, which are features of AI models. Now others want in. Amazon has its Trainium2 chip available to AWS users, and Intel’s Gaudi 3 is a Meta Training and Inference Accelerator.
Nvidia is the only big competitor that hasn’t competed with it at the top. There are other places where there is competition. The Information – a tech news website – highlighted the growth of “batch process”, which offers businesses cheaper AI models, as long as they are willing to wait for periods of low demand to run their queries. This allows OpenAI, for example, to purchase cheaper and more efficient chips in their datacentres, rather than spending all of their money on the fastest hardware.
Smaller businesses are now offering more specialised products than Nvidia in a race to the top. Groq, not to be confused Elon Musk’s Grok AI (whose launch has sparked a trademark dispute), makes chips that can’t even be used to train AI – but run the models they produce blazingly quickly. Etched is also building a chip which only runs one kind of AI model, a “transformer”, or the T in the GPT (generative-pre-trained-transformer) acronym.
Nvidia needs to thrive, not just compete with the competition. It must thrive to reach the next milestone. Market fundamentals have fallen out of favor, but even a $3tn valuation would require the company to sell a billion of its high-end GPUs every year at a 30% margin for all eternity, according to an expert.
Nvidia may have a harder time defending its own profit margin, even if the AI sector grows to the point where it can justify this. It has the chip designs that will allow it to lead the market, but its supply chain is hampered by the same bottlenecks as the rest of industry. These are foundries like those operated by Taiwanese TSMC, America’s Intel, China’s SMIC, and very few others. Nvidia, a TSMC customer, is not included in this list. If Nvidia needs to eat up the rest of TSMC’s order book in order to meet demand, the profits will flow to them too.
Neil Wilson, chief analyst of the brokerage firm Finalto said that the bear case against Nvidia, which is market jargon to describe a sustained drop in share prices, rested on an argument that the company would return to less frenetic demand levels once it had worked through its order books.
Wilson said that “all their customers are ordering GPUs in a rush, but this won’t last forever.” Wilson said that customers over-order, then cancel. The sweet spot is now, but it can’t be sustained. He sees Nvidia reaching $4tn or beyond but “maybe at the current rate”.
Jim Reid, Deutsche Bank’s head of global economy and thematic analysis, wrote a note last week in which he asked if Nvidia is “the fastest-growing large company ever?” He noted that Nvidia grew from $2tn up to $3tn within 30 days. Reid also said it took Warren Buffett more than 60 years to bring Berkshire Hathaway to close to $1tn.
Reid said that AI’s economic promise was welcomed in a world with low productivity, a measure for economic efficiency, and a declining population of working-age people and increasing government debt.
He wrote: “If AI is the catalyst of a fourth Industrial Revolution that would be a very good thing.” “If not, then the markets will have a major problem.”
There is more at stake than just winning the race to $4tn.
Post Disclaimer