The open source weight model swallows up 72% of the year: Damo warns that software dividends only go to companies that control complex workflows

The Zhitong Finance App noticed that in the past year, the AI model landscape has been quietly but profoundly restructured. In a research report released on October 26, Morgan Stanley gave an impressive set of data: on the third-party model routing platform OpenRouter, the share of tokens processed by the open source weight model has soared from about 22% a year ago to about 72% today; on Vercel's AI gateway platform, this ratio has jumped from 7% in December 2025 to 63%.

Behind this set of numbers is a clear sign that developers are rapidly flocking to lower cost per unit open source models to handle an increasing number of routine workflows. And Morgan Stanley's core judgment is that this wave is “constructive” for the software industry as a whole, but the dividends are by no means inclusive; the ultimate value will flow to the companies that have the deepest moat and control complex workflows.

Costs have plummeted, but “saving money” isn't the whole story

Damo emphasized that the meaning of the open source model is not that “the model has become cheaper,” but that “the cost of producing useful intelligence has been systematically reduced, and the consumption of intelligence can be dramatically increased.” For software companies, this opens up two very different opportunities.

The first opportunity is to expand the scope of “economically viable workflows” to broaden usage scenarios and usage — but this door is also open to competitors. The second opportunity is more differentiated: cheaper intelligence makes enterprise context, trusted data, telemetry, secure execution, transaction processing, and governance, the “exclusive control assets” of these vendors more valuable. Where these capabilities are difficult to replicate, manufacturers are more motivated to maintain and even increase pricing.

In other words, the real winners are not companies that simply hand over cost advantages to customers, but companies that can seize “greater monetization opportunities” and “stronger pricing power” at the same time.

Three major differences: sovereignty, single-customer economy, and customer acquisition

The report sorted out three key debates surrounding the implementation of the open source model.

First, will data sovereignty risk hinder enterprise adoption? Many believe that Western open source models such as Nemotron 4 and Mistral Large 4 are rapidly approaching the ability to lead international rivals and will leverage previously hesitant corporate customers; bearish parties worry that European and American companies are still cautious due to geographical, reputational, and security concerns.

Damo's key difference is that “where the model is developed” and “how to host, govern, and monitor the model” are two different things — hyperscale cloud vendors can fully resolve enterprise concerns through regional hosting, encryption, fine-grained access control, and continuous monitoring.

Second, can the open source model improve the “single-customer economy” of AI? Let's take a look at the three major benefits: getting rid of dependency on expensive proprietary APIs, being able to fine-tune specific processes, and controlling the deployment environment and inference optimization. The bears' objections are equally sharp: these advantages apply equally to AI-native startups, cloud vendors, and self-built customers, and may eventually evolve into lower prices, lower switching costs, and the erosion of the business model from “price per seat” to “price by task/result”.

Third, can an open source strategy leverage new customers and expand the user pool? Damo believes that lower reasoning costs make the “Product Driven Growth” (PLG) model work, but the same model can be used by everyone, so the return ultimately still belongs to companies with proprietary data, deep customer workflow awareness, and regulatory barriers.

The heat and cooling of the four tracks are uneven

When it comes to specific sections, Morgan Stanley's conclusion shows a clear “Matthew effect.”

In the field of infrastructure software, the open source model is generally beneficial. Core data platforms (Microsoft, Databricks, MongoDB, Oracle, Palantir, Snowflake) will usher in higher query volumes and wider use cases due to AI agents' need for real-time, permission-controlled enterprise context; observability vendors (Datadog, Dynatrace, Elastic) will benefit from proxies generating massive new trace and governance requirements.

Cybersecurity is another major structural winner. The explosive growth in model supply meant a surge in the number of AI agents running within the enterprise, and the attack surface expanded along with it — model supply chain integrity, prompt injection, proliferation of machine identities, unauthorized autonomous execution, and data leakage across workflows all became new security gaps. CrowdStrike, Palo Alto Networks, and Okta were listed as the biggest beneficiaries.

The opportunity for horizontal SaaS is “shifting from seat pricing to results/usage pricing”. Atlassian, Salesforce, and ServiceNow lead the market, while Box, Figma, Zoom, and Wix lead the small to medium markets in the Microsoft camp. Vertical software leverages the triple advantage of “low automated market+result alignment and pricing+deep workflow control”, making Shopify, Toast, Manhattan Associates, Navan, and ServiceTitan the biggest beneficiaries.

Big shuffle

Damo pointed out that the popularity of the open source weighting model is essentially a rapid commercialization of “intelligence.” When model barriers are removed, the focus of competition shifts from “who has a better model” to “who has a deeper moat, more complex processes, and more difficult data assets to be replicated.”

For investors, this means a repricing: “point solutions” companies with only a single function, lack proprietary data, and rely on simple workflows will be marginalized in this wave of declining costs. Atlassian, Cloudflare, CrowdStrike, Datadog, Microsoft, MongoDB, Nutanix, Palantir, Palo Alto Networks, SAP, and ServiceNow are the winners that can actually get through this round of AI reshuffle in Morgan Stanley's eyes.