Anthropic’s Flagship Model Is Struggling to Capture Spending

Anthropic’s flagship model has not captured the level of enterprise spending that many would expect from a newly released top-tier model. Its share reportedly sits around 11% of total Anthropic spending more than two months after launch. That is significant because businesses usually move toward newer models when they offer a clear improvement in performance.

Cost appears to be one important factor. High token prices can make a powerful model difficult to use for everyday workloads. Token limits can also restrict how much work users can complete. For businesses, the best AI model is not always the smartest model. It is often the model that delivers the right performance at an acceptable cost.

The Old AI Upgrade Cycle Is Breaking

The traditional AI market followed a predictable pattern. A company released a smarter model, developers tested it, and businesses gradually moved their workloads to the new system. The latest model became the default option because better intelligence usually justified the higher price.

That pattern is becoming less certain. Businesses now compare AI models based on cost, speed, privacy, flexibility, and deployment options. A company can use a powerful model for complex planning and a smaller model for routine tasks. It can also combine cloud AI with local AI. This gives companies more control over how they spend their AI budgets.

Open-Weight Models Surge on Vercel

Open-weight models are gaining significant traction. Vercel gateway data reportedly showed their share of token usage rising from 28% to 62% in only two months. That is a major increase in a short period. It suggests that developers are becoming more comfortable using models outside the biggest closed AI platforms.

The economic appeal is clear. Open-weight models can be deployed on infrastructure controlled by the user. Companies do not have to purchase every token from an external provider. They can invest in hardware and use it across multiple workloads. As open models continue to improve, more businesses may consider this approach for tasks that do not require the most advanced frontier model.

Why Enterprises Are Moving Away From Expensive Frontier Models

Enterprise AI decisions are becoming more focused on return on investment. Companies need to know how much each AI task costs and what value it creates. They also need to consider where their data goes and how much control they have over the technology.

Many routine business tasks do not require the largest available model. Document processing, coding assistance, summarization, classification, and other standard workloads can often run on smaller models. A larger model can handle planning and reasoning while smaller models complete simpler tasks. This layered approach can reduce costs without forcing companies to sacrifice useful AI capabilities.

Thomson Reuters and the Push for AI Independence

Thomson Reuters provides an important example of this changing approach. The company reportedly invested $40 million to develop its own AI model using a modified version of Alibaba’s Qwen. The objective was to reduce dependence on a single frontier AI provider.

The company has not completely abandoned Claude. Instead, it is adding more control and flexibility to its AI strategy. This approach reflects a broader concern among enterprises. Depending entirely on an external AI provider can create long-term risks. Pricing, policies, infrastructure, and product decisions remain outside the customer’s control. Open models can give businesses another path.

The $30 Trillion AI Market Opportunity

Anthropic is reportedly presenting a potential $30 trillion total addressable market around AI during its IPO preparations. This figure represents the broad amount of work AI models could potentially perform. It should not be confused with a $30 trillion company valuation.

The size of this opportunity has also attracted criticism. A total addressable market is a theoretical measure. It does not mean companies will actually spend that amount on AI products. AI can transform productivity without turning the entire value of economic activity into revenue for AI model providers. Investors therefore need to look beyond large market estimates and examine actual customer spending, margins, adoption, and long-term economics.

Open Source vs. Closed AI: The Economics Are Changing

The economics of AI could change significantly as open-weight models become more capable. One possibility is that open models could eventually handle 60% to 90% of AI tokens while representing only 15% to 25% of the economic value. High token volume does not always produce high revenue.

This creates a challenge for companies that depend on expensive model access. Open-weight models can handle large volumes of work without charging a provider for every token. Businesses still pay for hardware, electricity, maintenance, and infrastructure, but they gain more control over those costs. For organizations with heavy AI usage, that trade-off can become increasingly attractive.

Why Local AI Is Becoming More Attractive

Local AI gives companies greater control over their technology stack. Privacy is one major reason. Businesses may want sensitive information to remain inside their own infrastructure rather than passing through an external AI service.

Vendor lock-in is another concern. A company that builds its entire product around one API becomes dependent on that provider’s pricing and policies. Local deployment provides another option. Businesses can choose the model they want, run it in their own environment, and change models when their requirements evolve. This flexibility is especially useful for organizations that operate in regulated environments.

OpenMonoAgent and the Local-First AI Approach

OpenMonoAgent.ai represents a local-first approach to AI development. It is designed as a terminal-native AI coding agent that can work with local LLMs. This allows developers to build with AI while maintaining greater control over the underlying infrastructure.

The system can run on a single machine or use separate machines for the agent and inference. A secure relay can connect those environments. The hardware requirements can also be practical. A system using an RTX 3090 can run a Qwen model locally, giving developers access to capable AI without requiring an expensive enterprise GPU setup. This makes local AI more accessible to smaller teams.

Why Businesses Should Own Their AI Stack

AI infrastructure can become an important part of a company’s competitive advantage. When a product depends completely on an external API, changes to pricing or policies can directly affect the business. A provider can also change its models or services, forcing teams to adjust their products.

A model-agnostic architecture gives businesses more flexibility. Teams can switch models based on cost, performance, or availability. They can use different models for different workloads and decide which tasks should run locally or in the cloud. A fractional CTO can help companies make these architecture decisions and build an AI strategy that fits their technical and business requirements.

The Future of AI May Be Model-Agnostic

The future of enterprise AI may not revolve around one dominant model. Businesses increasingly have access to different models with different strengths, costs, and deployment options. This gives them the ability to choose technology based on the task instead of committing everything to one provider.

This shift can also help smaller companies. A small team can combine open models, local hardware, AI agents, and strong engineering practices to build useful products. The competitive advantage comes from having a flexible architecture. Instead of asking which AI company should control the entire stack, businesses can build systems that allow them to change models when needed.

What Anthropic’s IPO Story Could Mean for the AI Market

Anthropic’s IPO plans are arriving at a time when enterprise AI behavior is changing. Expensive frontier models still have an important role, but open-weight models are gaining attention because they offer lower-cost and more controllable alternatives.

If this trend continues, frontier AI companies could face greater competition for enterprise workloads. Businesses may continue paying premium prices for complex tasks where advanced intelligence creates clear value. However, routine workloads could increasingly move to smaller models or local infrastructure. That would reduce the share of AI spending captured by the most expensive providers and create a more fragmented AI market.

Conclusion: Build AI You Can Control

The AI market is moving toward greater choice. Businesses no longer need to use one frontier model for every task. They can combine advanced models with smaller open-weight systems and local AI infrastructure. This approach can improve flexibility, reduce vendor dependence, and give companies more control over their data and costs.

The bigger opportunity is not simply choosing between cloud AI and local AI. It is building an architecture that can use both when necessary. Model-agnostic systems allow companies to adapt as AI models, prices, and technologies change. This local-first philosophy is also central to OpenMonoAgent. For businesses and developers exploring practical AI infrastructure, startuphakk provides a broader view of how technology teams can build useful systems without depending entirely on one vendor. The future of enterprise AI may belong to companies that own their architecture and control how intelligence becomes part of their products.