Model-Serving Platforms Reach 6.1% of AI-Spending Businesses, While Frontier Spend Holds

The share of AI-spending businesses paying for platforms that offer open-source and Chinese-developed models rose to 6.1% in July. The movement may indicate growing enterprise interest in lower-cost, adaptable model options, even as direct spending remains concentrated with frontier-model vendors.

Ramp’s AI Index put the share at 4.5% in January 2026. The increase measures businesses paying for model-serving platforms, not the volume of work they send through them; those platforms offer access to open-source and Chinese-developed models.

Ramp tracks token and subscription spending across its customer base, so the figure is a directional measure of its customers’ buying rather than a census of enterprise AI.

Open foundations offer a different operating model

Fortune describes open-weight models as generally cheaper than proprietary alternatives, with more opportunity for fine-tuning on particular datasets and greater control over how company data is handled. Moonshot AI’s Kimi K3 is an open-weight model characterized as close to leading proprietary systems in coding and agentic work, or multistep tasks a model carries out.

Z.AI said its GLM-5.3-Flash costs $0.15 per million input tokens and $0.50 per million output tokens, adding another low-cost Chinese option.

Platform buyers increased over six months

AI-spending businesses paying for model-serving platforms

4.5%→6.1%

share of businesses

Ramp recorded a 1.6-percentage-point increase in the share of AI-spending businesses paying for these platforms.

Two specialized deployments show the appeal

Thomson Reuters built Thomson-1, an in-house document-review model based on Snowdon, which adapted Alibaba’s open-source Qwen model. The company said Thomson-1 will take over document-review tasks that previously ran on Anthropic’s Claude.

Harvey took a related but different route. It said Harvey Tenet was post-trained on Kimi K3 and outperformed its base model and named U.S. frontier systems on complex legal agentic tasks. Harvey had previously customized closed models from Anthropic, OpenAI and Google.

The leading vendors still command direct spending

Ramp said open-source growth had not yet directly reduced spending shares for either leading provider. In July, Anthropic held 43.5% of business AI market share after gaining 1.1 percentage points, while OpenAI held 39.7% after a 0.23-point gain.

The contrast may be sharpest at premium pricing. Fable 5 was priced at roughly $10 per million tokens, about twice the cost of GPT-5.6 Sol. It accounted for 6% of Anthropic tokens and 11.4% of its dollars spent, versus 25% of OpenAI tokens and 23% of spending for GPT-5.6 Sol. Ramp lead economist Ara Kharazian argued that businesses may not pay a premium for the best model when a cheaper system is sufficient.