OpenAI’s projected $5 billion operating loss for 2024 illustrated the unusually high cost of competing at the frontier of artificial intelligence, where running popular products can consume billions of dollars in cloud computing before a company reaches durable profitability. The estimate, reported by The Information in July 2024, placed OpenAI’s revenue for the year at $3.5 billion to $4.5 billion while forecasting spending that would substantially exceed those sales.
The largest expense was expected to be computing capacity. The Information estimated that OpenAI would spend nearly $4 billion during 2024 renting servers to operate ChatGPT and the large language models behind it. Training new models, including data preparation and payments for data, was projected to cost another $3 billion.
Those figures help explain why OpenAI’s relationship with Microsoft has been central to its expansion. Building and serving advanced AI models requires thousands of specialized chips, vast data-center capacity and dependable electricity supplies. Companies can sell subscriptions and enterprise access, but their costs rise sharply as more users generate text, code, images and other outputs.
Microsoft’s role extends beyond funding
Microsoft’s first major investment in OpenAI came in 2019, when the software company committed $1 billion to the organization. OpenAI and Microsoft said at the time that the arrangement included an exclusive computing partnership and a plan to build an Azure-based AI supercomputer platform.
Greg Brockman, OpenAI’s co-founder and then chief technology officer, said the 2019 investment was made in cash. Microsoft’s accompanying announcement said it would receive rights to commercialize “pre-AGI” technologies developed from OpenAI research, while the two companies would collaborate on Azure AI infrastructure.
Microsoft expanded the partnership in January 2023 through what it described as a multiyear, multibillion-dollar investment. Reuters and other outlets reported the investment at $10 billion, with deal terms widely reported to include a profit-sharing arrangement and an ownership stake that could reach 49% after Microsoft recovered its initial outlay.
Part of Microsoft’s contribution was reportedly delivered through Azure cloud-computing credits rather than unrestricted cash. That structure gives OpenAI access to the infrastructure it needs, but it also ties the company’s operating model closely to Microsoft’s cloud platform and pricing.
The relationship has raised strategic questions as both companies develop their own AI products. Microsoft has embedded OpenAI models across its Copilot offerings, Azure services and productivity software, while reports have said the company is also building an in-house foundation model known as MAI-1. The Information reported in May 2024 that Mustafa Suleyman, Microsoft AI chief and a co-founder of DeepMind, was leading work on the model.
Microsoft hired Suleyman and much of the team from Inflection AI in March 2024 under an arrangement that gave Microsoft access to parts of Inflection’s technology. The move gave Microsoft a prominent internal AI research group while preserving its commercial relationship with OpenAI.
Revenue growth faces an expensive race
OpenAI’s estimated costs reflect a difficult equation confronting developers of frontier models. Each generation requires more computing power, more training data and larger technical teams. Serving models after training can also be expensive, particularly when products gain mass-market adoption and users expect fast responses.
Early estimates had already shown the scale of the burden. Analyst firm SemiAnalysis estimated in early 2023 that ChatGPT could cost roughly $700,000 a day to operate, or about $235.3 million annually, depending on usage and infrastructure assumptions. The later projection reported by The Information suggested that OpenAI’s spending had grown far beyond the expense of operating one chatbot.
Personnel costs also added to the total. The Information estimated OpenAI had around 1,500 employees in 2024 and projected roughly $1.5 billion in annual employee-related and other operating expenses. AI researchers, systems engineers and infrastructure specialists command some of the highest compensation packages in technology, especially as major companies compete for a limited pool of talent.
OpenAI had raised more than $13 billion over roughly five years, according to figures reported at the time. Yet fundraising totals can obscure the difference between capital for operating expenses and transactions that allow existing shareholders to sell stock.
A February 2024 secondary share sale valued OpenAI at about $80 billion, according to The New York Times. Such secondary transactions can provide liquidity to employees and early backers, though they do not necessarily add new cash to a company’s balance sheet for training models or paying cloud bills.
Model costs are colliding with infrastructure limits
The financial pressure is linked to technical constraints that extend beyond OpenAI. Reports on next-generation GPT systems indicated that future models could require four to five times as much training data as GPT-4. Publishers have increasingly restricted web scraping, while lawsuits over copyrighted training material have added legal uncertainty to the data supply.
Synthetic data—material generated by AI systems and then used to train later models—has been proposed as one response. Researchers have warned that excessive dependence on synthetic data can degrade a model’s output over time, a risk often described as model collapse. The extent of that risk depends on how data is generated, filtered and combined with human-created material.
Power supply presents a separate bottleneck. Data centers housing AI chips require large and reliable electricity connections, while new generation projects and transmission upgrades can take years to complete. Lawrence Berkeley National Laboratory found that the typical wait for U.S. electricity-generation projects seeking grid interconnection had reached roughly five years by 2022, up from less than two years in 2008.
Google reported in its 2024 environmental report that its greenhouse-gas emissions had risen 48% from 2019 levels, citing the energy demands of data centers and supply chains as contributing factors. The company’s disclosure showed how AI infrastructure can complicate climate targets even for technology groups that have invested heavily in clean-energy procurement.
OpenAI’s projected 2024 loss therefore placed its valuation beside a demanding operational reality: revenue needed to rise fast enough to cover an expanding bill for servers, chips, data and power. Its ability to convert enterprise demand for generative AI into recurring, high-margin revenue would shape whether the company’s enormous infrastructure commitments become a sustainable business rather than a continuing dependence on outside capital.
Curious how AI reshapes trading? Explore our smart AI copy trading guide to automate strategies efficiently.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
