Enterprise AI has hit a financial wall. For the past two years, the conversation centered on capability—whether a model could write a complex function or summarize a thousand-page document. But as companies move from experimental pilots to production-scale deployments, the focus has shifted from what the AI can do to what it costs to run. CFOs are now staring at monthly token bills from OpenAI and Anthropic that scale linearly with usage, creating a sustainability crisis for high-volume AI pipelines. This shift in the corporate psyche is exactly where Databricks has found its latest surge of momentum.
The Capital Shift Toward AI Orchestration
Databricks recently saw its valuation climb to 188 billion dollars, a figure that signals a massive bet by the market on the infrastructure surrounding the model rather than the model itself. This valuation emerged from a new funding round led by Coatue. While the company has not officially disclosed the exact amount raised, industry reports suggest the round was approximately 3 billion dollars. This influx of capital is not merely a vote of confidence in a data company, but a recognition that the next phase of AI value lies in optimization and data management.
The growth trajectory of Databricks has been aggressive. In December 2024, the company was valued at 62 billion dollars. By September 2025, that number jumped to 100 billion dollars. The momentum continued into February of this year, when the company closed a Series L funding round of 5 billion dollars, pushing its valuation to 134 billion dollars. The leap to 188 billion dollars represents a doubling of value in less than two years. This rapid ascent suggests that investors are pivoting away from the "model wars" and toward the tools that make those models economically viable for the Fortune 500.
Central to this strategy is the adoption of open-weight models. Databricks has moved aggressively to support models where the weights are public, allowing companies to avoid the "token tax" of proprietary APIs. Specifically, the company has championed Z.ai's GLM 5.2, an open-source based large language model. Internal benchmarks conducted by Databricks reveal a critical insight: in specific coding tasks, open models are not just cheaper, but often more cost-effective without sacrificing quality. These benchmarks show that GLM 5.2 can handle high-difficulty coding assignments while maintaining a significantly lower total cost of ownership compared to the proprietary offerings from Anthropic or OpenAI.
The Harness as the New Unit of Value
If the model is the engine, the harness is the chassis and the fuel injection system. The industry is realizing that the model itself is often less important than the harness—the agentic coding tools and frameworks that wrap the model, manage prompt context, and dictate instructions. The harness determines how much context is sent to the model and how that information is structured. Because LLM pricing is tied to token volume, an inefficient harness can double or triple the cost of a project even if the underlying model remains the same.
This is where the distinction between a model-centric and a harness-centric strategy becomes clear. Databricks is betting that the ability to manage context efficiently is the real competitive advantage. For example, the open-source harness Pi has emerged as a preferred choice for those looking to lower costs without degrading output quality. By optimizing how prompts are surrounded by context, Pi reduces the number of unnecessary tokens processed, directly lowering the final invoice.
To institutionalize this approach, Databricks has expanded its product ecosystem to move beyond its identity as a big data warehouse. The company has launched Lakebase, a database specifically designed for AI agents, and Unity, an AI gateway that manages the flow of requests. To tie these together, they introduced Omnigent, a meta-harness designed to integrate and manage multiple agents across a corporate environment. This suite transforms Databricks from a place where data is stored into a command center where AI agents are orchestrated.
This strategic pivot aligns with a broader 2026 trend where enterprises are increasingly adopting low-cost, open-weight models to escape vendor lock-in. By providing the Lakebase, Unity, and Omnigent stack, Databricks ensures that regardless of which open model a company chooses—such as GLM 5.2—the orchestration layer remains a Databricks product. The value has shifted from the intelligence of the model to the efficiency of the delivery system.
The era of chasing the highest benchmark score at any cost is ending. The success of an AI deployment is now measured by the ratio of performance to token spend, making the harness the most critical component of the modern AI stack.
