Leaked financials show OpenAI’s 2025 revenue topped $4 billion in just nine months, but the company’s costs are escalating even faster. While investors celebrate this historic growth, the underlying numbers reveal a cautionary tale: a compute spend curve that is outpacing sales and could redefine the economics of generative AI.
Revenue sprint meets cost climb
OpenAI’s revenue growth is on a historic trajectory, projected to reach a $15-20 billion run rate in 2025. However, this explosive growth is challenged by even faster-climbing infrastructure costs for model training and inference, putting significant pressure on the company’s profitability and long-term financial model.
According to SaaStr, annual recurring revenue (ARR) rocketed from $3.7 billion in 2024 to $12 billion by July 2025, marking the company’s first $1 billion revenue month. This trajectory suggests a 2025 run rate of $15-20 billion, with ChatGPT subscriptions (Plus, Team, and Enterprise) accounting for roughly 75% of sales.
A TechCrunch report shows OpenAI paid Microsoft $866 million in revenue share for the first three quarters of 2025. This payment, representing 20% of top-line sales under their Azure partnership, implies gross revenue of at least $4.3 billion during that period.
The widening compute bill
OpenAI’s demand for training and inference compute power is growing faster than its revenue. Analyst Tomasz Tunguz projects annual infrastructure costs could jump from ~$6 billion in 2025 to $173 billion by 2029, backed by over $1 trillion in multiyear vendor commitments. The majority of this spending goes directly to Microsoft Azure, suppressing gross margins.
Major cost drivers include:
- Intensive R&D and training for next-generation foundation models.
- Soaring inference demand from millions of paid ChatGPT subscribers.
- Fulfilling uptime guarantees for global enterprise contracts.
- Foundational investments in autonomous AI agents scheduled for a 2026 release.
As a result, the cost of revenue reached $2.5 billion in the first half of 2025 alone, nullifying operating profit despite record sales. Analysts predict margins will likely stay under 50% until breakthroughs in custom silicon or cheaper energy become available.
Strategic levers for 2026
To balance its finances, OpenAI’s strategy for 2026 focuses on diversifying revenue. Projections from Epoch AI indicate that enterprise licensing and API usage are expected to double their revenue share, while consumer subscriptions will decrease to less than 60% of the total mix. A revised agreement with Microsoft also reduces vendor lock-in by permitting OpenAI to use alternative clouds for certain government contracts.
Ultimately, the core challenge is whether revenue can continue to outpace compute inflation. The answer will not only determine OpenAI’s financial viability but also influence the entire market’s investment thesis for generative AI.
How much revenue did OpenAI generate in the first three quarters of 2025?
Leaked documents show the company handed $866 million to Microsoft as its 20% revenue-share payment for the January-September period. Under the long-standing Azure-for-models agreement, that cut implies OpenAI booked at least $4.3 billion in top-line revenue during those nine months.
With July marking the first-ever $1 billion revenue month, OpenAI’s full-year 2025 run-rate is now tracking toward $15–20 billion, well ahead of the $12.7 billion forecast it gave investors at the start of the year.
Why are OpenAI’s costs rising faster than its sales?
Demand for GPUs is outrunning every other line item. In 2024 the firm spent an estimated $7 billion on cloud compute; the first half of 2025 alone already shows $6.7 billion in R&D-driven compute bills, and the company has signed more than $1 trillion of infrastructure commitments for 2025-2035. Even though annualized revenue has roughly tripled since 2024, the compute tab is growing well into double-digit percentages each quarter, squeezing gross margin to 48% this year.
What does OpenAI’s ballooning Microsoft bill cover?
Two separate meters are spinning:
- Revenue share: Microsoft keeps 20¢ of every dollar OpenAI earns on ChatGPT subscriptions and API calls.
- Azure consumption: OpenAI buys raw capacity to train and serve models; leaked figures show over $12 billion paid to Microsoft for compute power since 2024, on top of the revenue share.
The partnership remains mutually beneficial – Microsoft’s cloud revenue line benefits while OpenAI secures scarce GPU supply – but it also locks the startup into ever-larger Azure purchase commitments ($250 billion in future obligations disclosed in October 2025).
When will OpenAI become profitable?
Not soon. Internal forecasts shared with investors project annual losses through 2028, driven by:
- Rising inference cost per additional user
- Model-training cycles for next-gen releases
Profitability is pencilled in around 2029–2030, assuming revenue scales to $125 billion and gross margin widens to 70% as software efficiencies and dedicated chips come online.
How does the Microsoft revenue share shape OpenAI’s strategy?
The 20% payout, plus the need to keep buying Azure capacity, influences every product decision:
- Pricing: Enterprise tiers must cover both cloud bill and the royalty, nudging list prices upward.
- Architecture: Models are optimised for Microsoft’s hardware roadmap, making it harder to switch providers.
- Government sales: Recently negotiated carve-outs let OpenAI sell API access to U.S. agencies outside Azure, a small but growing channel that could diversify revenue and margin pressure.
For now, the partnership remains the fastest route to GPU supply and global reach; OpenAI’s challenge is to grow revenue quickly enough that the 20% slice, however large, stays manageable against a projected $30–40 billion top line in 2026.
