Running a standard AI benchmark task now costs about 15 cents. That number alone doesn’t mean much until you realize the same task cost roughly 97 times more just a year and a half ago.
ARK Invest unpacked this staggering cost decline on Episode 140 of its podcast, The Brainstorm, laying out what it means for AI competition, enterprise adoption, and the broader market.
The 97x collapse in AI costs
The core data point from ARK’s analysis is hard to overstate. AI inference costs have dropped by up to 97 times within approximately 18 months. During peak periods, historical cost reductions have exceeded 90% per year.
ARK’s analysts noted that this cost compression is fundamentally reshaping the competitive landscape. The firms duking it out at the top of the AI market, OpenAI, Anthropic, Meta, and xAI, are no longer differentiated primarily by capability. They’re increasingly competing on performance-per-dollar.
What 15-cent tasks mean for businesses
ARK predicts that as cost per task approaches low single-digit cents, enterprises could collectively save hundreds of billions of dollars on knowledge work. That’s a forecast grounded in observable cost curves that have been declining at 90%-plus annually.
This threshold effect is what ARK sees driving the next wave of agentic AI adoption, where AI systems don’t just assist human workers but handle entire workflows autonomously. The podcast highlighted that this agentic future is anticipated but hasn’t fully arrived yet.
The bottleneck isn’t the models themselves. It’s real-world execution. Benchmarks keep improving, but translating benchmark performance into reliable, autonomous action in messy, unstructured business environments remains the critical gap.
The competitive landscape is shifting
For the major AI players, this cost collapse creates a strategic dilemma. OpenAI, Anthropic, Meta, and xAI each bring different approaches to the table, from proprietary APIs to open-source models to vertically integrated platforms. As raw capability becomes table stakes, the winners will likely be determined by distribution, integration, and pricing strategy rather than benchmark bragging rights.
ARK’s framing suggests that investors should be watching not just who builds the best model, but who builds the best business model around cheap AI. If the cost of running an AI task approaches zero, the premium shifts to whatever is scarce: proprietary data, domain-specific fine-tuning, reliable integration with existing enterprise systems, and the trust that comes with consistent real-world performance.