Open-source weighted models take 72% market share in one year: Morgan Stanley warns that software dividends only flow to companies controlling complex workflows

English Financial Newswire (EFN) has noted that over the past year, the landscape of AI models has undergone a quiet but profound restructuring. In its October 26 research report, Morgan Stanley presented a set of striking data: on third-party model routing platform OpenRouter, the market share of open-source weights models processing tokens has surged from about 22% a year ago to around 72% today; on Vercel’s AI Gateway platform, this proportion jumped from 7% in December 2025 to 63%.

This data points to a clear signal: developers are accelerating their migration to lower unit-cost open-source models to handle an ever-increasing volume of routine workflows. Morgan Stanley’s core projection is that this wave is “constructive” for the software industry as a whole; however, the dividends are by no means universal and ultimate value will flow to those companies with the deepest moats and control over the most complex workflows.

Costs Plummet, but “Saving Money” Is Not the Whole Story

Morgan Stanley emphasized that the significance of open-source models lies not in making models cheaper, but in the systemic reduction of the cost to produce useful intelligence—while consumption of intelligence can expand dramatically. For software companies, this opens up two very different opportunities.

The first is to expand the scope of “economically viable workflows,” thus broadening usage scenarios and volume — yet this door is equally open to competitors. The second opportunity is more differentiated: cheaper intelligence increases the value of enterprise context, trusted data, telemetry, secure execution, transaction processing, and governance—assets exclusively controlled by certain vendors. Wherever these capabilities are hard to replicate, vendors are better positioned to maintain or even increase pricing power.

In other words, the true winners are not simply those who pass cost advantages to customers, but those able to simultaneously capture both “greater monetization opportunities” and “stronger pricing power.”

Three Key Debates: Sovereignty, Unit Economics, and Customer Acquisition

Regarding the adoption of open-source models, the report outlines three critical debates.

First, does data sovereignty risk impede enterprise adoption? Bulls point out that Western open-source models like Nemotron 4 and Mistral Large 4 are rapidly approaching the capabilities of their leading international peers, set to win over previously hesitant enterprise customers; bears worry that US and European companies will remain cautious due to geopolitical, reputational, and security concerns.

Morgan Stanley makes a key distinction: “Where the model is developed” and “how it is hosted, governed, and monitored” are two different matters—hyperscale cloud vendors can fully address enterprise concerns via regional hosting, encryption, fine-grained access control, and continuous monitoring.

Second, can open-source models improve the “unit economics” of AI? The bull case identifies three main benefits: freeing from reliance on expensive proprietary APIs, fine-tuning for specific processes, and control over deployment environments and inference optimization. The bear rebuttal is sharp: these advantages apply equally to AI-first startups, cloud vendors, and self-built clients, and may ultimately lead to price cuts, reduced switching costs, and a shift in business models from seat-based pricing to task/result-based pricing.

Third, can open-source strategies unlock new customers and expand user pools? Morgan Stanley argues that lower inference costs make “product-led growth” (PLG) models truly scalable, but since the same models are available to all, value still accrues to companies with proprietary data, deep workflow understanding, and regulatory barriers.

Uneven Heat Across Four Tracks

At the sector level, Morgan Stanley’s conclusions reveal a clear “Matthew effect.”

In the infrastructure software sector, open-source models are broadly positive. Core data platforms (Microsoft, Databricks, MongoDB, Oracle, Palantir, Snowflake) will see higher queries and broader use cases as AI agents drive demand for real-time, permissioned enterprise context; observability vendors (Datadog, Dynatrace, Elastic) will benefit from the massive volume of new traces and governance needs 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 inside enterprises, expanding the attack surface—the integrity of the model supply chain, prompt injection, machine identity sprawl, unauthorized autonomous execution, and cross-workflow data leaks all become new security gaps. CrowdStrike, Palo Alto Networks, and Okta are listed as the top beneficiaries.

For horizontal SaaS, the opportunity is to “shift from seat-based to results/usage-based pricing.” Within the Microsoft ecosystem, Atlassian, Salesforce, and ServiceNow lead the large-cap segment, while Box, Figma, Zoom, and Wix lead among mid and small caps. In vertical software, “low-automation markets + results-aligned pricing + deep workflow control” confer triple advantages, making Shopify, Toast, Manhattan Associates, Navan, and ServiceTitan the biggest beneficiaries.

A Major Reshuffle

Morgan Stanley points out that the proliferation of open-source weights models is essentially accelerating the commoditization of “intelligence.” As model barriers are leveled, competition shifts from “who has a better model” to “who possesses deeper moats, more complex processes, and harder-to-replicate data assets.”

For investors, this signals a repricing: companies with only single features, lacking proprietary data, and relying on simple workflows—these “point solution” firms—will be marginalized in this cost-down cycle. Atlassian, Cloudflare, CrowdStrike, Datadog, Microsoft, MongoDB, Nutanix, Palantir, Palo Alto Networks, SAP, and ServiceNow are, in Morgan Stanley’s view, the true winners capable of navigating this round of AI reshuffling.