1. Introduction

Two companies dominate the chatbots this presentation discusses, and neither one makes money. Both sell subscriptions and programming access, both grow revenue several times over each year, and both spend far more than they take in. Investors cover the gap. Three terms carry the sections that follow. Revenue is what customers pay. Operating loss is what the business loses running itself, meaning revenue minus the cost of operating. Net loss adds accounting charges that move no cash, so a net loss can dwarf the operating loss without anyone writing a larger check. Both companies show exactly that pattern in 2025. Neither company publishes financial statements. The figures below come from audited documents that reporters obtained: OpenAI_Financials_2015-2025.pdf rests on statements verified by the Financial Times in June 2026, and Anthropic_Financials_2021-2025.pdf rests on a confidential draft filing reviewed by Reuters in September 2026. Figures marked n/d were never disclosed, and early-year revenue is often an outside estimate rather than a booked number. One figure frames both sections. In 2025 the two companies together reported about 18 billion dollars of revenue and about 29 billion dollars of operating loss. For every dollar a customer paid, the companies spent roughly 2.60 dollars.

2. OpenAI, 2015 to 2025

OpenAI began as a nonprofit with a billion dollars in pledges and reached 13 billion dollars of revenue ten years later. Spending rose faster than revenue at every step.

Year	Funding raised	Revenue	Compute	Total opex	Operating loss	Net loss
2015-18	$1B pledged	negligible	n/d	n/d	n/d	n/d
2019	$1B (Microsoft)	n/d	n/d	n/d	n/d	n/d
2021	~$1B	n/d	n/d	n/d	n/d	~$0.27B est.
2022	-	~$28M	n/d	n/d	n/d	~$0.54B
2023	$10B (Microsoft) + ~$0.3B	~$1B est.	n/d	n/d	n/d	n/d
2024	$6.6B (Series E)	$3.7B	~$7B est.	$12.48B	$8.78B	~$5B
2025	$40B (Series F)	$13.07B	~$17.2B	$34.0B	$20.92B	$38.53B

Revenue grew 3.5 times between 2024 and 2025, and the operating loss grew 2.4 times alongside it. Compute is the reason. OpenAI paid Microsoft about 17.2 billion dollars in 2025 for computing, research, and related services, which exceeded the company’s entire revenue that year. Research and development ran 19.18 billion dollars and sales and marketing came to nearly 6 billion. The 38.53 billion dollar net loss needs a qualifier. Roughly 30 billion of it is a one-time charge with no cash attached, tied to the conversion of investor interests during the move to a Public Benefit Corporation. Excluding that charge, the loss is about 8 billion. The operating loss of 20.92 billion is the number that reflects running the business. Funding kept pace with the losses. The 40 billion dollar Series F led by SoftBank arrived in 2025, and a 122 billion dollar Series G at an 852 billion dollar valuation followed in March 2026, after the period this table covers.

3. Anthropic, 2021 to 2025

Anthropic started five years later, at a smaller scale, and shows the same shape. Revenue multiplied about twelve times in one year, and the company still spent nearly three dollars for every dollar it earned.

Year	Funding raised	Revenue	Compute	Total opex	Operating loss	Net loss
2021	$124M	n/d	n/d	n/d	n/d	n/d
2022	$580M	~$10M est.	n/d	n/d	n/d	n/d
2023	~$2.6B	~$100M annualized	n/d	n/d	n/d	n/d
2024	~$7.5B	$386M	~$2.5B	~$3.37B	$2.98B	$8.31B
2025	~$20B + up to $15B committed	$4.59B	$7.33B	$12.65B	$8.06B	$41.97B

Compute dominates the cost here too. The 7.33 billion dollars Anthropic spent on computing and infrastructure in 2025 came to 1.6 times its revenue, and accounted for 58 percent of total operating expense. The same qualifier applies to the net loss. About 34 billion of the 41.97 billion is a non-cash charge on financing that converts into shares. The operating loss of 8.06 billion is the figure that measures the cost of running the business. The funding came largely from companies that sell computing. Amazon put in 1.25 billion in 2023 and 6.75 billion in 2024, Google added roughly 1.8 billion across 2023 and 2025, and Microsoft and Nvidia committed up to 15 billion in November 2025. Money flows from the cloud providers to the model builders, and much of it flows back as payment for computing.