ChatGPT is about to turn three years old and continues to stand out as a unique startup. The most promising project in the world, and all the competitors created in its wake have made one thing clear to users, companies, and markets in its short existence: predicting the future of artificial intelligence (AI) is not easy, but it tends to exceed expectations. This has become the key word for measuring technology in the market. And no one knows how to fuel these fantasies more than Sam Altman, the creator of OpenAI, especially when he has the support of the most powerful man in the industry, someone who doesn't create models or chatbots, but the chips that enable their design: Jensen Huang, CEO of Nvidia.
The agreement of the world's most valuable company to invest $100 billion in OpenAI has disrupted the playing field where moves have not been linear for a long time. The impact of this agreement has also raised suspicions, reinforced by the exaggerated forecasts about the future of OpenAI's infrastructure that Altman has outlined and currently lacks the funds to support. Despite ChatGPT becoming the fastest adopted digital product in history and almost a monopoly.
There are currently five major battlegrounds in the world of AI: infrastructure; the future of the Internet and everything we do on it; the world of businesses and code; audiovisual creation; and solving complex problems. Paradoxically, the one attracting headlines, massive investments, and dominating the thoughts of top executives seems to be the first of these, which is undoubtedly the most boring.
Microsoft, Google, Amazon, and Meta have already shown their willingness to put all their efforts into deploying the infrastructure that will enable the AI era. Hundreds of data centers will require investments exceeding $350 billion annually, an unparalleled effort on paper for any other company, but one that has found a response in the new powers of AI and Oracle, their new preferred partner.
Altman has taken it a step further. In recent weeks, he announced a major agreement with Nvidia: $100 billion investment aimed at purchasing computing capacity of 10 gigawatts (GW). The agreement also benefits a third party: Oracle, which has doubled its stock value in six months. The math seems clear: OpenAI will receive $100 billion from Nvidia and will invest it in renting capacity in the data centers that Oracle is building and will fill with Nvidia GPUs, thus bringing the money back to their coffers.
This is only possible if Nvidia meets its profit forecasts, which involve generating so much liquidity that it is difficult to find a destination for it, according to financial analysts. By investing in OpenAI, Jensen Huang's company is betting that it will continue to be a winning horse in the world of AI and a long-term customer. The second issue is not trivial, as Meta, Amazon, Microsoft, and Google already have their own chips and are developing others to rely less on Nvidia, whose agreement with OpenAI will be very profitable. According to Bank of America's estimates, it will result in a return of between three and five times the investment, between $300 billion and $500 billion for OpenAI, or in other words, between $30 billion and $50 billion per gigawatt just in GPUs (the rest needs to be paid for).
And this is where Altman's ability to sell expectations comes into play again. In a blog post, he framed the agreement within a long-term vision for the group. "If AI continues on the trajectory we believe, incredible things will be possible. With 10 gigawatts of computing power, AI might be able to cure cancer or tutor every student on the planet," he stated, before revealing his final vision.
"We want to create a factory that produces a gigawatt of new AI infrastructure every week," he asserted, a vision that he himself acknowledges is "extremely difficult" to achieve. The final touch came with the leak of a projection of the capacity that OpenAI plans to use, reaching its peak in 2033 with 250 gigawatts, 125 times the current capacity. A monumental project that would face an unprecedented challenge in terms of power generation.
Marta Sánchez, EY's Energy partner, puts into perspective the magnitude of OpenAI's promise. "About a year ago in Spain, we started making estimates and recovering data center projects. We had a scenario of 1,800 megawatts (1.8GW) by 2030, and the data center industry told us we were underestimating, as they believed it would be 3,500 MW (3.5GW). Some communities needed to double their energy demand for this," she explains, highlighting two things: the scaling up of data center projects due to AI and the dimensional change that would represent for the entire electrical infrastructure. The capacity promised by Altman is equivalent to what is available at the highest points of energy consumption in India.
Although each country and region varies significantly, Sánchez also points out that the challenge goes beyond access to electrical grids, involving immense construction projects to build these data centers and the issue of producing enough energy. She emphasizes that this is only possible with renewables and batteries for companies committed to green initiatives, or in other cases, solutions like small modular reactors (SMRs) or gas. "In countries like the US and Germany, there is already a waiting list for turbines," explains the EY partner.
Therefore, to build the 'super-AI,' ChatGPT needs to rapidly increase its revenues, and it won't be solely from customers. According to its latest update, it has around 700 million weekly users, which means that almost 10% of the world's population connects to OpenAI's tool at least one day out of every seven. This figure represents 76.4% of the chatbot market, according to the consultancy SimilarWeb, almost a monopoly, but with trends that cannot allow the company to rest on its laurels, especially when leadership does not translate into financial success.
Following ChatGPT is Gemini in second place. Google's chatbot has around 450 million monthly users (not weekly), but according to the consultancy, it surpassed a 10% market share for the first time in September. Just six months ago, while ChatGPT's dominance was similar, Gemini had only a 4.9% share, ranking third after DeepSeek, which has lost half of its share in the same period - dropping from 8% to 4%.
From the headquarters of the world's most used search engine, they see this as their moment after reading dozens of articles predicting their demise. In addition to Gemini's users, who have made it the number one downloaded app in multiple markets, there are over 2 billion users consuming AI-generated content on the search engine, and the future deployment of AI Mode globally, a series of launches that enhance a well-oiled business like Google's digital advertising, which also benefits from the growing influence of a YouTube powered by AI.
An income stream that Altman has not yet built. Currently, OpenAI generates around $12 billion in annual recurring revenue (ARR, an indicator that takes into account organic revenues from the last month and projects that figure twelve times), far from what is needed to materialize its projects. According to an extensive study published by the company, 70% of its users use the tool for personal purposes, while 30% use it in their work environment. Among the most used functions are writing and editing texts (28.10%), seeking practical advice on health, beauty, or education (28.3%), and information searches (21%).
These pillars are ideal for building an advertising business similar to Google's or Meta's, one of the paths the startup is exploring. However, there is another more immediate one that the company has already publicly launched: purchases through ChatGPT, a model in which, thanks to its partnership with Shopify and the payment company Stripe, users can buy directly from the chat, a move that could shake up e-commerce and increase the company's revenues.
The extensive business sector. However, the market that all AI players are targeting is the business sector, and the clearest example is Anthropic, perhaps OpenAI's other major technological competitor. Referring back to the previous data, according to SimilarWeb, Claude, their AI assistant, did not reach 2% of users, a penetration almost 40 times lower than Sam Altman's company. However, the company founded by former OpenAI employees announced in September that in eight months, its annual recurring revenue had increased from $1 billion in January to $5 billion in August, almost half of its top rival with significantly fewer users.