Enterprise AI Spending Shifts Toward Open-Weight Models as Adoption Surges

Open-weight AI models accounted for 56 percent of tokens processed through Vercel's AI Gateway in August 2026, up sharply from just 7 percent in December 2025, according to data reported by AI Weekly.

The rapid shift reflects growing enterprise preference for models that can be deployed on private infrastructure without relying on proprietary APIs from companies like OpenAI or Anthropic. Open-weight models allow organizations to run AI locally, giving them more control over data privacy and costs.

The trend is being driven by improvements in the quality of open-weight models. Recent releases from companies including Meta, Mistral, and several Chinese AI startups have narrowed the performance gap with leading proprietary models.

Goldman Sachs projects that major hyperscalers, including Amazon, Alphabet, Microsoft, Oracle, and Meta, will invest $1.2 trillion in AI infrastructure in 2027, up from $800 billion in 2026. The investment reflects the enormous computing demands of training and running large AI models.

However, high interest rates are creating financial pressure on the data center buildout. With 10-year Treasury yields near 5.17 percent, operators like CoreWeave and Oracle have reported increased interest expenses on the debt-heavy infrastructure projects.

On the supply side, Vanguard International Semiconductor inaugurated a 300mm advanced fabrication facility in Singapore in September 2026. The plant is already fully booked due to high demand for AI-related silicon interposers.

The shift toward open-weight models is also influencing how companies think about AI vendor relationships. Organizations that previously relied on a single proprietary provider are increasingly building systems that can switch between models depending on cost, performance, and availability.

Industry analysts said the trend is likely to continue as open-weight model quality improves and enterprises become more sophisticated in their AI deployments.