DeepSeek cut its V4-Pro model prices by 75%, but enterprise vendors are finding that cheaper pricing does not automatically improve profitability. Agent systems—AI that performs multi-step reasoning rather than single-turn responses—consume tokens so rapidly that the per-token savings are being outpaced by higher total token usage per task, threatening the margin gains vendors had expected from falling inference costs.
Summaries like this, in your inbox every morning.
Sign up free →What happened
DeepSeek drastically cut pricing on its V4-Pro model by 75%. However, the cost reduction is being offset by a fundamental shift in how AI systems operate—agent systems (AI that chains together planning, retrieval, tool use, verification, and follow-up decisions) consume tokens far faster than traditional chatbots, which turn one user question into one model call.
Why it matters
For two decades, software economics followed a pattern where infrastructure became cheaper each year while applications became more capable, and many assumed AI would follow the same path. Instead, the shift toward agents is reversing that dynamic—even as per-token costs plummet, total token consumption per user task is rising sharply, making it harder for vendors to maintain healthy margins despite price cuts.
What to watch
The gap between token price declines and agent token consumption rates. The body describes this as an exponential problem that "has begun crumbling," signaling that the business model assumption underlying AI margin improvement may no longer hold.
For the past two decades, the economics of software have been defined by a reliable pattern: infrastructure costs fell steadily each year while applications became more capable. When frontier AI models began to improve and token prices dropped, the industry assumed inference would follow the same trajectory—becoming a negligible operating expense over time. DeepSeek's 75% price cut on V4-Pro appeared to validate that expectation.
However, the article reveals a critical breakdown in that assumption. The shift from simple chatbots to agent systems fundamentally changes token consumption patterns. A traditional chatbot processes a user's single question in a single model call. An agent, by contrast, must perform multiple internal steps—planning the approach, retrieving relevant information, using tools, verifying results, summarizing findings, and deciding on follow-up actions—all before presenting the user with a single answer. Each of those steps consumes tokens, and the cumulative token cost per user task can be orders of magnitude higher than a single inference. The body explicitly states this problem "has begun crumbling exponentially," indicating not a gradual shift but an accelerating mismatch between falling per-token prices and rising total token consumption. This suggests that vendors cannot rely on price declines alone to maintain or improve margins; the architecture of AI applications themselves has become the primary cost driver.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
