Cost pressures drive enterprises toward open-source AI alternatives

Organizations are increasingly switching from expensive proprietary AI models to cheaper alternatives as token costs become unsustainable, with companies like Harvey and AT&T shifting significant workloads to open-weight models they can run independently. Pricing strategies are also shifting on the vendor side, as providers like Anthropic eliminate volume discounts once contracted usage thresholds are reached. The trend reflects a broader industry shift toward treating AI models as commodities that can be swapped based on cost-performance tradeoffs rather than locked-in relationships.

The economics of token-based pricing have become untenable for many AI-dependent businesses. Harvey, a legal tech company, experienced a dramatic swing from positive to negative margins as its usage of paid API services scaled, forcing leadership to pursue independence through open-source alternatives. This pattern extends across industries—AT&T now routes two-fifths of its AI operations through open models it controls internally, reflecting a fundamental reassessment of deployment strategy among enterprise users.

Pricing practices from model providers are simultaneously tightening, with vendors like Anthropic structuring contracts to eliminate volume discounts once thresholds are met. This shift creates pressure for customers to either accept higher per-token costs or migrate workloads elsewhere, accelerating the commoditization of AI capabilities and reducing vendor lock-in.