Morgan Stanley: Open-Source Model Share Surges to 72%, Software Gains Concentrate on Companies with Deep Moats
nashnova research
Morgan Stanley cites OpenRouter data showing open-source models' share of processed tokens jumped from 22% to 72% in one year, marking a structural shift in the AI model landscape — but the payoff is not evenly spread, flowing instead to software companies with the deepest moats and control over complex workflows.
Open-source model share tripled in a year — what happened?
OpenRouter data shows open-weight models' share of processed tokens surged from roughly 22% to about 72%.
On Vercel's AI gateway, the ratio leapt from 7% in December 2025 to 63%.
This means → developers are shifting routine workloads en masse to lower-cost open-source models, eroding closed-source models' status as the default choice.
Costs are falling — where does the money go?
Morgan Stanley stresses the point is not "models got cheaper" but that the cost of producing useful intelligence is dropping systemically, so consumption of intelligence can scale dramatically.
This opens two paths for software companies: widen the range of "economically viable workflows," and make proprietary assets — enterprise context, trusted data, security enforcement, transaction processing, governance — more valuable.
In plain terms = the model itself is losing value, but whoever holds data and processes that others cannot replicate gains pricing power.
Why does infrastructure software benefit across the board?
AI agents — software programs that execute tasks autonomously — need real-time enterprise context, driving higher query volumes for core data platforms. Morgan Stanley names Microsoft, Databricks, MongoDB, Oracle, Palantir, and Snowflake.
Observability vendors — tools that monitor how software runs — also benefit: agents generate massive new tracing and governance demands. Datadog, Dynatrace, and Elastic are listed as beneficiaries.
This reflects a deeper shift: once the model layer is commoditized, competition moves down to "who controls the data pipeline and runtime monitoring."
Why is cybersecurity a structural winner?
Open-source model supply explodes → the number of AI agents running inside enterprises surges → the attack surface widens accordingly.
New security gaps include model supply-chain integrity, prompt injection — hijacking AI behavior via malicious instructions — machine-identity sprawl, unauthorized autonomous execution, and cross-workflow data leakage.
CrowdStrike, Palo Alto Networks, and Okta are named the top beneficiaries. In plain terms = more AI agents means more doors to guard, and security companies' business only grows.
Where are the opportunities in SaaS and vertical software?
The core shift in horizontal SaaS: moving from per-seat pricing to outcome- or usage-based pricing. Large-cap leaders are Atlassian, Salesforce, and ServiceNow; mid-cap leaders are Box, Figma, Zoom, and Wix.
Vertical software benefits from a triple advantage — low-automation markets, outcome-aligned pricing, and deep workflow control. Shopify, Toast, Manhattan Associates, Navan, and ServiceTitan are listed as top beneficiaries.
This means → the more deeply a software product is embedded in industry processes and the harder it is to replace, the stronger its pricing power becomes as model costs fall.
Who did Morgan Stanley name as winners — and who gets marginalized?
Morgan Stanley's list of companies that can ride out this AI reshuffle: Atlassian, Cloudflare, CrowdStrike, Datadog, Microsoft, MongoDB, Nutanix, Palantir, Palo Alto Networks, SAP, and ServiceNow.
The profile of companies at risk: point-solution vendors with a single function, no proprietary data, and reliance on simple workflows.
This reflects an accelerating winner-take-more dynamic: with model barriers leveled, competition shifts from "who has the better model" to "who holds the deeper moat and the harder-to-replicate data assets."
市场有风险，内容仅供研究参考，不构成投资建议。