Palantir CEO Karp Warns AI Token Economics Are Broken
Alex Karp argues that soaring token costs are pushing enterprises away from OpenAI and Anthropic toward leaner, open-weight AI models.
Palantir CEO Alex Karp is sounding the alarm on what he sees as a fundamental flaw in the commercial AI strategies of OpenAI and Anthropic, arguing that their reliance on token-based pricing is creating unsustainable cost structures for enterprise customers. In Karp's framing, the current trajectory is not a minor inefficiency — it is a systemic problem. "Something has gone completely wrong," he said, a rare moment of blunt public criticism from a tech executive toward two of the industry's most prominent players.
At the heart of Karp's critique is the practice sometimes called "tokenmaxxing" — the tendency to maximize token consumption as a metric of AI engagement or capability, which in turn drives up operational costs for companies deploying these systems at scale. Rather than treating token volume as a proxy for value, Karp argues that enterprises are increasingly demanding the opposite: doing more with fewer tokens, prioritizing efficiency as a competitive necessity rather than an optional optimization.
Read more Boeing Confirms IT Outage Disrupted Internal Systems and Apps →
The commercial consequence, according to Karp, is a visible market shift. Skyrocketing inference costs are nudging businesses toward open-weight models — AI systems whose parameters are publicly available — which offer far greater flexibility for cost control and on-premise deployment. This dynamic puts pressure on the closed, API-gated model that OpenAI and Anthropic have built their revenue structures around, suggesting that the current pricing paradigm may not hold as enterprise buyers grow more sophisticated.
Karp's comments carry strategic weight beyond mere competitive sniping. Palantir has positioned itself as the infrastructure layer for AI deployment inside large organizations and government agencies, meaning its fortunes are directly tied to whether enterprises can afford to run AI workloads at scale. If token costs remain prohibitive, that is arguably good for Palantir's pitch — but it is also a genuine constraint on the broader adoption of AI tools that the entire industry depends on to justify its valuations.
The tension Karp is highlighting reflects a deeper unresolved question in enterprise AI: whether the current generation of frontier model providers can build durable business models before customers route around them. Continue reading at US Top News and Analysis.
