News Analysis
Research Report
The AI Transformation 100: Concrete Ideas for Fixing How We Work
Key Takeaways
- OpenAI generated more than $13 billion in revenue in 2025.
- The company reported a $38.53 billion net loss, though accounting charges tied to its restructuring inflated that figure.
- Research, infrastructure and compute costs remain the central challenge for frontier AI companies.
- Potential price cuts and IPO speculation could put new pressure on OpenAI to show a clearer path to profitability.
OpenAI generated more than $13 billion in revenue during 2025 but reported a net loss of $38.5 billion, according to documents reviewed by the Financial Times. The loss was primarily driven by heavy research spending, infrastructure costs and accounting changes tied to its ongoing corporate restructuring.
The details behind OpenAI's 2025 financial results capture both the scale of demand for generative AI and the immense costs required to build it. The figures arrive as OpenAI reportedly considers price reductions to compete more aggressively with rivals such as Anthropic, and as questions continue to swirl around a potential future IPO.
Financial Statements Show Massive Losses for OpenAI
OpenAI’s revenue surged in 2025, but expenses climbed even faster.
The company reported $13.07 billion in revenue, while total costs and expenses eached $34 billion. Research and development alone accounted for $19.18 billion, exceeding the company’s full-year revenue. That spending helped produce a $20.92 billion operating loss before additional accounting adjustments.
Quick Look: OpenAI's 2025 Financials
| Metric | Amount (USD) |
|---|---|
| Revenue | $13.07 billion |
| Total costs and expenses | $34.00 billion |
| Research & development | $19.18 billion |
| Operating loss | $20.92 billion |
| Net loss attributable to OpenAI | $38.53 billion |
Attracting the most attention is OpenAI's $38.53 billion net loss attributable to the company. However, that number was heavily influenced by non-cash accounting charges associated with OpenAI's transition to a Public Benefit Corporation. Financial filings show changes in the fair value of convertible interests and related liabilities generated more than $41 billion in losses during the year, significantly inflating the final net-loss figure.
Even with those restructuring-related charges separated out, the underlying economics remain stark. OpenAI spent nearly $3 for every $1 of revenue it generated in 2025.
That does not mean demand is weak. The opposite appears true, with reports indicating OpenAI’s monthly revenue had approached $2 billion by the end of 2025, suggesting demand continued accelerating through the year. The issue is cost.
Related Article: OpenAI’s IPO Faces Questions Before It Even Begins
Why Frontier AI Costs So Much
Frontier AI does not behave like a traditional software business.
Conventional SaaS companies can often grow revenue without increasing infrastructure costs at the same pace. Frontier AI companies face large expenses at nearly every stage of the product lifecycle.
Training large language models (LLMs) requires massive GPU clusters running for weeks or months. Once models launch, inference costs continue every time a user asks a chatbot a question, generates an image, runs an agentic workflow or makes an API call. That cost structure makes frontier AI look less like classic software and more like infrastructure development.
Financial documents reviewed by independent journalist Ed Zitron indicated that OpenAI paid approximately $17.2 billion to Microsoft in 2025 for computing, research and related services, showing how dependent frontier AI companies remain on massive cloud infrastructure.
Compute is only part of the bill. OpenAI and its competitors also continue to spend heavily on:
- Model development
- Safety testing
- Alignment research
- Talent acquisition
- Data center access
- Product expansion
- Enterprise support
Competition for top AI researchers also remains intense, with leading companies offering compensation packages that can rival those found in professional sports or finance.
The Expensive Theory Driving AI Growth
OpenAI’s leadership has argued that aggressive investment makes sense because model capabilities continue improving as more computing resources get applied to training and inference.
Sam Altman has repeatedly pointed to AI scaling laws, arguing that larger investments in compute continue producing gains in model intelligence. As long as that assumption holds, leading AI companies have a strong incentive to spend heavily. That creates a difficult cycle.
More users drive more revenue. More users also create more infrastructure demand. More capable models may attract more customers, but those models also cost more to build, deploy and maintain.
This dynamic has led some investors to evaluate companies such as OpenAI less like software vendors and more like cloud infrastructure providers. Amazon Web Services, Microsoft Azure and Google Cloud all required years of heavy investment before reaching their current profitability levels.
The question for OpenAI is whether frontier AI can follow a similar path, where scale eventually improves margins, or whether the industry’s appetite for compute will continue absorbing much of the revenue it generates.
Is OpenAI Chasing Growth Over Profit?
OpenAI’s current strategy clearly points toward growth.
The company continues to expand ChatGPT, adding enterprise customers, growing its API business and pushing into international markets. It is also competing against Anthropic, Google, Meta and increasingly capable open-source models.
From that perspective, OpenAI’s losses look less surprising. The company appears focused on capturing market share before the AI market matures.
There are historical parallels. Amazon spent years prioritizing growth over profitability as it built logistics networks, expanded into new markets and established dominant positions across multiple business lines. Investors tolerated those losses because revenue growth remained strong and the long-term opportunity looked enormous. OpenAI may be pursuing a similar playbook.
The company has already built one of the strongest brands in AI. ChatGPT remains a defining consumer product. Enterprise interest in AI tools continues rising. Developers continue building products around large language models.
There are also signs that OpenAI’s economics may be improving. Analysis of the leaked financial statements found that operating losses declined as a percentage of revenue between 2024 and 2025, suggesting the company may be gaining some efficiency as it scales.
Still, growth alone does not guarantee profitability. OpenAI must show that revenue can grow faster than compute, research and infrastructure costs. If those costs keep rising alongside usage, investors may eventually demand clearer proof that frontier AI can produce sustainable profits.
Related Article: The End of LLM Loyalty: Why One AI Model Won’t Rule Them All
What OpenAI's Price Cuts Could Mean
Reports that OpenAI may lower prices come at a critical moment for the AI market.
For the past several years, leading AI companies have competed mainly on model performance. That competition now appears to be shifting toward price, efficiency and enterprise value.
Several forces are increasing pressure:
- Anthropic continues gaining enterprise traction
- Google is embedding AI across its product portfolio
- Meta and other open-source players have narrowed capability gaps
- Enterprises now have more model options than they did a year ago
- Buyers are beginning to compare AI vendors on cost, not just capability
If OpenAI lowers prices, the move could mean two different things.
First, competition may have intensified enough to force the company to defend market share more aggressively. Second, improvements in hardware, model efficiency and infrastructure management may be reducing the cost of delivering AI services.
Either way, the implications extend beyond OpenAI.
Price cuts would add momentum to the debate over whether large language models are becoming commoditized. In traditional software markets, companies often differentiate through features, integrations, data, customer relationships and workflow depth. In AI, the question is whether model performance alone can remain a durable advantage as competitors catch up.
Enterprise buying behavior will matter. Many businesses still experiment with AI deployments and evaluate long-term vendor strategies. If buyers begin treating models as interchangeable, pricing could play a much larger role in purchasing decisions.
That creates a difficult balance for frontier AI companies. Lower prices may accelerate adoption, but they also put pressure on companies that already face enormous research and infrastructure costs.
For investors, the question is whether AI companies can reduce prices while still funding the next generation of models.
Why an OpenAI IPO Would Matter
While OpenAI has confidentially filed for an initial public offering with the Securities and Exchange Commission (SEC), it has yet to announce a formal IPO timeline. But as revenue grows and OpenAI’s role in the technology industry expands, questions about its long-term structure and financing have intensified.
An IPO would change how OpenAI gets evaluated. Private investors often tolerate years of losses when a company captures a large market and builds strategic advantages. Public markets tend to apply more pressure. Quarterly earnings, analyst expectations and shareholder scrutiny would put greater focus on profitability, margins and cash flow. That would create a challenge for OpenAI, because the same forces driving its growth also drive its costs.
The company continues investing in compute, research, talent and model development. Investors may accept those costs while AI remains in an early growth phase. Public shareholders would likely want stronger evidence that those investments can eventually produce durable profits.
Governance would also draw attention.
OpenAI’s transition from its original nonprofit structure to a public benefit corporation has already generated scrutiny. A future IPO would likely bring more questions about corporate control, fiduciary responsibilities and how the company balances commercial goals with its stated mission to develop AI that benefits humanity.
An OpenAI IPO would test more than investor enthusiasm for one company. It would become one of the first major public-market tests of frontier AI economics.
Related Article: Ciao, Claude: Open-Source AI Closes the Gap on Proprietary Features
Can Frontier AI Become Sustainable?
OpenAI's results point to a question facing the entire AI industry: Can frontier AI become a sustainable business?
Factors Influencing Frontier AI Economics
| Potential Positive Force | Potential Negative Force |
|---|---|
| Growing enterprise adoption | Raising compute requirement |
| Increasing API usage | Expensive GPU infrastructure |
| Improved model efficiency | Heavy research spending |
| Global market expansion | Price competition |
| Platform ecosystem growth | Open-source alternatives |
| Potential scale advantages | Pressure on profit margins |
Demand looks real. Businesses are adopting AI tools. Developers are integrating models into applications. Consumers use AI services at massive scale. But costs remain unusually high.
Winner Takes Most
One possible outcome is a winner-take-most market. A small number of providers could reach enough scale to spread research, infrastructure and development costs across huge customer bases. Cloud computing followed a similar pattern, with a few dominant companies investing billions in data centers, networking and platform development.
Growth of Open Source
Another outcome looks less favorable for frontier model providers. Open-source models may keep improving. Model performance gaps may keep narrowing. Enterprises may gain more leverage. Pricing pressure may rise. If AI services become more interchangeable, companies could struggle to generate strong profits even as usage expands.
Underdogs Rise Up
A third possibility is that smaller, more efficient models reshape the market. Many businesses may not need the most powerful frontier model for every task. Specialized systems could deliver acceptable performance at much lower cost. If that trend accelerates, AI economics may shift away from today’s compute-heavy model.
The relationship between revenue growth and compute costs will likely determine which version of the market takes shape.
For AI economics to improve, revenue must grow faster than spending on infrastructure, model development and research. Hardware efficiency, model optimization and lower inference costs could help. Falling prices, rising competition and ever-larger model requirements could push in the opposite direction.
OpenAI’s 2025 financials offer evidence for both sides of the debate. The company has shown that generative AI can generate enormous demand. It has also shown how much capital frontier AI consumes.
But now, can leading AI companies can turn rapid adoption into durable profits? The answer will shape OpenAI’s future and the economics of the AI industry around it.