WhiteLine Daily, published by WuBlockchain, says AI is showing a pattern that resembles the early internet: the cost of a unit of intelligence is falling, but agent working time and token consumption are rising even faster. In that setup, the market is no longer judging companies only by model strength. Usage, distribution efficiency and gross margin are moving closer to the center of the debate.
Codex is moving beyond developers
OpenAI’s latest disclosure shows Codex weekly active users have passed 5 million. About 20% of them are knowledge workers, and that segment is growing more than three times faster than developers. WhiteLine Daily says users are no longer relying on Codex only for coding. They are also using it to build reports and spreadsheets, organize contracts, complete research and process internal data.
The report places more weight on depth of use than on user count alone. Among users in external organizations, 17.3% used Codex in the past month. Even so, Codex generated 63.3% of the combined output tokens from Codex and ChatGPT. In other words, agent penetration is still limited, but token consumption jumps once users begin relying on it.
WhiteLine Daily explains the difference this way: a chatbot mainly answers questions, while an agent is assigned a task. That task can require reading files continuously, calling tools, editing content and checking the result. More than 10% of Codex users run more than three agents at the same time each week. Based on the amount of human time needed to complete similar work, the share of individual users handing off jobs worth more than eight hours of labor to Codex has grown nearly tenfold this year.
That makes token counts less of a measure of conversation volume and closer to a measure of actual work performed.
Lower prices can still produce higher revenue
Since mid-July, the effective usage price of leading U.S. models has fallen by nearly one-quarter, according to the report. After OpenAI cut Luna pricing by 80%, usage on OpenRouter rose 14x over a short period, while revenue still increased 34%. Terra posted about 5x usage growth and a 45% increase in revenue.
WhiteLine Daily also notes that this dataset covers only about two weeks after the price cuts. Whether that trend can hold over a longer period remains unclear. Lower pricing can bring in more calls, but the report says it is still hard to tell whether the added demand comes from stable users or from developers temporarily switching between models.
DeepSeek has taken a different approach. The peak output price for V4 Pro was raised from $0.87 per million tokens to $3.96 per million tokens, while peak and off-peak pricing was introduced at the same time. The idea is to offer discounts during low-demand periods and charge more during peak periods to ration limited compute capacity.
Different pricing paths lead back to the same calculation: revenue depends on price and volume. If usage grows faster than prices fall, cutting prices can still lift token revenue. If usage fails to keep pace, a price war will feed directly into lower gross margins.
Usage growth > price decline → token revenue growth
The next question is whether all those tokens are profitable
WhiteLine Daily says Anthropic has stated that its annualized revenue run rate in May exceeded $47 billion. The report adds an important caveat: annualized figures simply project the current pace of revenue across a full year and are not the same as audited full-year revenue.
Anthropic and OpenAI confidentially submitted S-1 filings on June 1 and June 8, respectively. A confidential filing does not mean an IPO date has been set, the report says, but a public prospectus would give the market several data points that remain in short supply: model gross margin, long-term compute commitments, the cost of a single agent task and how much new compute is needed to sustain revenue growth.
Fast token growth can support revenue, but that alone does not settle the business model. If inference costs and compute spending rise faster than revenue, larger scale can also mean larger cash burn. WhiteLine Daily concludes that revenue growth will still matter in the next phase, but unit economics and cash flow will decide whether valuations can hold up.