User Content as ‘Modern Oil’ for AI Training
AI training data value is no longer an abstract idea; it is becoming a central bargaining chip in the AI economy. Reddit CEO Steve Huffman describes Reddit’s user-generated discussions as “modern oil” for AI, arguing that large language models would “not exist as we know them” without that content. He highlights that LLMs rely on natural, wide-ranging human conversations, and Reddit’s archives provide exactly that, across nearly every topic imaginable. This framing recasts casual posts, comments, and debates as critical raw material for AI systems. It also underscores a broader shift: user content AI economics are moving from free, open web scraping to tightly controlled, monetised access. By treating conversations as strategic assets rather than public commons, platforms are redefining who captures value in the AI boom—and challenging the assumption that AI innovation can scale without paying for the data that trains it.
How LLM Data Licensing Deals Create New Revenue Streams
Reddit’s recent LLM data licensing deals with Google and OpenAI show how platforms are monetising their archives. Huffman says Reddit has moved from open, permissive access to a “commercial use requires commercial terms” stance. The company now charges for commercial API access while still offering free access to researchers and universities. This selective openness lets Reddit collaborate with partners like Google and OpenAI, putting guardrails on how data is used while tapping new revenue streams. At the same time, Reddit is pursuing lawsuits against firms such as Anthropic and Perplexity, accusing them of using Reddit data without proper licenses. The result is a new tiered ecosystem: firms that pay for structured, compliant access, and others facing legal and reputational risks. These dynamics suggest that AI training data value is crystallising into formal markets, where contracts and court cases decide who can turn conversations into AI products—and at what price.
Bubble Economics: When AI Revenue Is a Circular Loop
While data is becoming a salable asset, some experts argue that AI company valuations rest on fragile economics. Zoho founder Sridhar Vembu calls AI “clearly an investment bubble,” pointing to what critics describe as round-trip revenue. A large cloud provider invests in an AI startup, but much of the investment is issued as cloud credits rather than cash. The startup then spends those credits on the investor’s own cloud services, which the provider records as new revenue. The same loop reportedly underpins major relationships such as Microsoft–OpenAI and Amazon–Anthropic, where investor, supplier, and customer roles blur. Corporate filings suggest that a significant chunk of future cloud backlogs is tied to a handful of AI firms funded by the same giants booking the revenue. This raises uncomfortable questions: how much of the current AI boom reflects real, diversified demand versus internally recycled spending and paper gains?
Paper Gains, Real Capex—and the Question of Sustainable Value
Beyond circular revenue, AI company valuations are being buoyed by paper profits. When AI startups raise fresh rounds at higher valuations, their big-tech investors mark up the value of those stakes and report the unrealised gains as profit. Alphabet, for example, recently reported a record profit figure, with nearly half attributed to a markup on its Anthropic investment. Amazon likewise reported a large profit number, with more than half linked to Anthropic-related gains, even as its free cash flow reportedly dropped sharply due to heavy spending on data centres. These accounting dynamics amplify the perception of booming profitability while masking how dependent the numbers are on a few high-priced AI bets. When paired with platforms asserting that their data is indispensable, the tension is clear: user-generated content is becoming more valuable, yet the structures built on top of that data may be more fragile than headline valuations suggest.