Open-weight models capture 72% share in one year: Morgan Stanley warns software gains flow only to companies that control complex workflows

Notice that over the past year, the AI model landscape has been undergoing a quiet yet profound restructuring. In a research report released on October 26, Morgan Stanley presented a set of highly striking data: on the third-party model routing platform OpenRouter, the share of tokens processed by open-weight models has soared from about 22% a year ago to about 72% today; on Vercel's AI gateway platform, this proportion jumped even further, from 7% in December 2025 to 63%. Behind these numbers is a clear signal: developers are accelerating their migration toward lower-cost open-source models to handle an ever-growing number of routine workflows. Morgan Stanley's core judgment is that this wave is "constructive" for the software industry as a whole, but the benefits are by no means universalultimately, value will flow to companies with the deepest moats that control complex workflows. Costs Plunge, but "Saving Money" Is Not the Whole Story Morgan Stanley emphasizes that the significance of open-source models lies not in "models becoming cheaper," but in "a systematic decline in the cost of producing useful intelligence, while consumption of intelligence can expand dramatically." For software companies, this opens up two very different opportunities. The first opportunity is expanding the range of "economically viable workflows," thereby broadening use cases and volumebut this door is equally open to competitors. The second opportunity is more differentiated: cheaper intelligence makes vendor "proprietary assets"enterprise context, trusted data, telemetry, secure execution, transaction processing, and governancemore valuable. Where these capabilities are difficult to replicate, vendors have greater confidence to maintain or even raise prices. In other words, the real winners are not companies that simply pass cost advantages on to customers, but companies that can simultaneously capture "a larger monetizable opportunity" and "stronger pricing power." Three Key Debates: Sovereignty, Unit Economics, and Customer Acquisition Around the implementation of open-source models, the report sorts out three key debates. First, will data sovereignty risks hinder enterprise adoption? The bull case holds that Western open-source models such as Nemotron 4 and Mistral Large 4 are rapidly approaching the capabilities of leading international rivals, which will unlock previously hesitant enterprise customers; the bear case worries that European and American enterprises will remain cautious due to geopolitical, reputational, and security concerns. Morgan Stanley's key distinction is that "where a model is developed" and "how a model is hosted, governed, and monitored" are two different thingshyperscale cloud providers can fully address enterprise concerns through regional hosting, encryption, fine-grained access controls, and continuous monitoring. Second, can open-source models improve the "unit economics" of AI? The bull case lists three major benefits: escaping dependence on expensive proprietary APIs, the ability to fine-tune for specific processes, and control over the deployment environment and inference optimization. The bear case's rebuttal is equally sharp: these advantages also apply to AI-native startups, cloud providers, and customers building in-house capabilities, and may ultimately evolve into price cuts, lower switching costs, and business model erosion as pricing shifts from "per seat" to "per task/per outcome." Third, can open-source strategies unlock new customers and expand the user pool? Morgan Stanley believes that lower inference costs make the "product-led growth" (PLG) model truly work, but the same models are available to everyone, so returns will ultimately accrue to companies with proprietary data, deep customer workflow knowledge, and regulatory barriers. Uneven Heat Across Four Tracks Turning to specific sectors, Morgan Stanley's conclusions show a clear "Matthew effect." In infrastructure software, open-source models are broadly positive. Core data platforms (Microsoft, Databricks, MongoDB, Oracle, Palantir, Snowflake) will see higher query volumes and broader use cases due to AI agents' demand for real-time, permission-controlled enterprise context; observability vendors (Datadog, Dynatrace, Elastic) will benefit from the massive new traces and governance demands generated by agents. Cybersecurity is another major structural winner. The explosive growth in model supply means a surge in the number of AI agents running within enterprises, expanding the attack surfacemodel supply chain integrity, prompt injection, machine identity proliferation, unauthorized autonomous execution, and cross-workflow data leakage all become new security gaps. CrowdStrike, Palo Alto Networks, and Okta are listed as the biggest beneficiaries. The opportunity in horizontal SaaS lies in "shifting from seat-based pricing to outcome/usage-based pricing," with Atlassian, Salesforce, and ServiceNow leading large caps in the Microsoft camp, and Box, Figma, Zoom, and Wix leading small and mid caps. Vertical software, with its triple advantages of "low-automation markets + outcome-aligned pricing + deep workflow control," makes Shopify, Toast, Manhattan Associates, Navan, and ServiceTitan the biggest beneficiaries. The Great Reshuffle Morgan Stanley points out that the spread of open-weight models is essentially rapidly commoditizing "intelligence." When model barriers are flattened, the focus of competition shifts from "who has a better model" to "who controls deeper moats, more complex processes, and data assets that are harder to replicate." For investors, this means a repricing: "point solution" companies with only a single feature, lacking proprietary data, and relying 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 true winners in Morgan Stanley's eyes that can navigate this round of AI reshuffling.