DeepSeek Raised Its Prices and Its Customers Stayed. That Is the Warning.
Business cases built on the lowest rates assume those rates will last, and the cheapest major model has just shown they may not.
Sign up for my Substack daily AI newsletter here.
See my AI Training course portfolio for corporate Business Leaders here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Since early 2025, DeepSeek has been the name most associated with cheap artificial intelligence. Its models competed with the leading laboratories at a fraction of the price, and many businesses built their plans around its rates. In August, it raised those rates by between 2.3 and 4.5 times, depending on the model, and its annualised revenue has since more than doubled to about $1 billion, according to Reuters, citing The Information. A price rise of that size would normally send customers looking elsewhere. They did not, and that should concern every South African board that has built a business case on today’s AI prices.
DeepSeek gave notice of the increase in early August, only a week after launching one of the cheapest capable models on the market, and observers linked the move to a surge in demand that was straining the computing power behind the service. Even after the rise, its prices remain among the lowest of the major models, and demand appears to have held. The company now also charges twice as much during peak hours as it does off-peak, pricing intelligence much as a utility prices electricity.
The wider market is moving in two directions at once. OpenAI has just cut prices on its newer mid-range and budget models by about half, saying efficiency gains allow it to pass savings on. At the same time, GitHub has replaced flat Copilot allowances with billing based on consumption, explaining that long, multi-step agentic sessions made the old model unsustainable.
INSIGHT AND ANALYSIS
The warning is not that AI prices will rise across the board. Some will keep falling. The warning is that a price is not a promise. Whether DeepSeek raised its rates as a deliberate strategy or because demand outgrew its capacity, the outcome for its customers was the same. A rate they had built on multiplied, and leaving was harder than paying.
That is what dependence does. Once a model sits inside customer service, coding or document workflows, switching means rewriting prompts, retesting outputs, retraining staff and accepting new risk. Those costs rarely appear in a business case, yet they decide who holds the bargaining power when prices change. The customers who stayed were not being irrational. They were doing the arithmetic, and the arithmetic favoured paying more.
I have previously written about this. In an article on AI spending in customer service, I showed that the function allocating the largest share of its budget to AI was also one where only about a quarter of leaders could demonstrate a positive financial return. Rising or unpredictable prices make weak business cases weaker still.
IMPLICATIONS
For South African organisations, the exposure is sharper because these prices are set in dollars and can grow quietly under consumption billing. Gartner has warned that AI spending per developer is starting to rival developer pay, with some firms paying more for AI each month than a junior South African developer earns. Boards should ask three questions before approving any AI business case. What happens to the return if the price doubles? How long would it take, and what would it cost, to move this workload to another model? What protections does the contract provide, such as caps on increases, notice periods and the right to take data and configurations elsewhere? Businesses that design for the ability to switch keep their bargaining power. Those that do not have already given it away.
CLOSING TAKEAWAY
Cheap AI has been one of the great gifts of this technology cycle, and in many parts of the market prices are still falling. A low price, however, only tells you what a supplier charges today. It tells you nothing about what it will charge once you cannot easily leave. DeepSeek’s customers learnt that in August. South African boards can learn it now, while they still have the freedom to negotiate.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments