Enterprise AI has a dirty secret: most of the money isn’t spent on models, it’s spent on plumbing. Databricks just unveiled OpenSharing, an open protocol that lets companies share AI models, agent skills, dashboards, and unstructured data across platforms without copying or moving any of it — and if it gains traction, it could quietly redraw how every CIO budgets for AI in 2026.
The pitch is simple but loaded. OpenSharing uses what Databricks calls a zero-copy credential vending model, which lets recipients securely access shared assets directly from a provider’s cloud storage through temporary, scoped credentials. No duplication. No replication. No ‘send me a copy and I’ll wire it up on my end.’ For enterprises trying to operationalize AI across business units, partners, and customers, that’s a structural shift, not a feature release.
Why the Integration Tax Is Strangling Enterprise AI
Ashish Chaturvedi, leader of executive research at HFS Research, framed the problem bluntly: every organization building multi-agentic systems is hitting the same wall, where the model, the skill, and the consumer all live on three different platforms. He called the integration tax ‘enormous’ and warned that it grows exponentially with every new partner, customer, or internal team added to the mix. Dion Hinchcliffe of The Futurum Group echoed the point, noting that today’s hidden AI costs aren’t just model development — they’re the ‘endless packaging, translation, sync, and governance effort’ needed to move AI assets across organizational boundaries.
This matters because the integration tax is the line item nobody puts on a slide. Procurement sees the model cost. Finance sees the cloud bill. Engineering eats the rest. If you’re a mid-sized fintech running a credit-risk model in Databricks while your fraud team operates on Snowflake and your customer-success team lives in Tableau, every cross-team AI initiative becomes a quarter-long integration project. OpenSharing’s promise is that those handoffs collapse into a protocol call. The prediction: integration tax will become a board-level metric within 18 months, and protocols like OpenSharing will be how CIOs argue they’ve cut it.
From Data Sharing to AI Supply Chains
Stephanie Walter, practice lead of the AI stack at HyperFRAME Research, made the sharpest observation in the briefing: enterprises are realizing the value is no longer just in the dataset, but in ‘the governed context, logic, and intelligence built around the dataset.’ Existing approaches share datasets well, she said, but they rarely address the broader AI package — the models, agent skills, dashboards, and applications that turn raw data into a decision. Hinchcliffe summarized the shift in one line: ‘CIOs increasingly want AI supply chains, not isolated data lakes like before.’
That reframing changes how procurement evaluates AI vendors. The question is no longer ‘can you give us the data?’ but ‘can you ship us the reasoning?’ Picture a regional bank that licenses a fraud-detection agent skill from a fintech partner. Under the old model, the bank’s team rebuilds the skill in their stack, drifts away from the upstream version, and absorbs the maintenance burden. Under OpenSharing, the skill is referenced in place, governed centrally, and updated by the provider. The same logic applies to anyone weighing AI agents against traditional automation pipelines — the unit of distribution is no longer a script, it’s a governed asset. Expect M&A diligence checklists to start asking ‘which AI sharing protocols do you support?’ within a year.
The Monetization Angle CIOs Are Quietly Tracking
Chaturvedi flagged the commercial implication most analysts buried: ‘For CIOs, the speed at which you can share AI assets across partners, subsidiaries, and customers determines the speed at which you can monetize your AI investments.’ If sharing an agent skill takes six weeks of integration work, the window is gone. If it takes a protocol call, AI becomes a distribution business.
That’s not marketing language. That’s a P&L statement. Consider a SaaS company that built a forecasting model for its enterprise tier. Today, exposing that model to a strategic partner means an API project, a security review, a contract amendment, and roughly a quarter of engineering time. Under a zero-copy protocol with scoped credentials, the same exposure becomes a configuration task. For product teams building AI-integrated software platforms, this is the difference between AI as a feature and AI as a revenue channel. The prediction: the first cohort of vendors to wrap their AI assets in OpenSharing-compatible packaging will capture an outsized share of partner-led revenue in 2026.
How OpenSharing Stacks Up Against Snowflake and What Developers Get
Walter was careful to note the novelty isn’t sharing itself, zero-copy access, or marketplace-style distribution — those exist in various forms across the market, including Snowflake’s Zero-Copy integrations. The difference, she pointed out, is that Snowflake allows data to be copied only when both provider and receiver are on Snowflake. OpenSharing is designed to work across platforms. Chaturvedi went further, arguing that no other open protocol covers agent skills and AI models as shareable, governed objects.
For developers, the practical payoff is the connector list. OpenSharing, an evolution of Databricks’ existing Delta Sharing protocol, is currently a sandbox project under the Linux Foundation AI & Data Foundation and lives on GitHub. Generally available connectors already include Python, Apache Spark, Tableau, PowerBI, Snowflake, DuckDB, Clojure, Node.js, Java, Arcurate, Rust, Go, C++, and R. Google Spreadsheet, Excel, Airflow, and Lakehouse Sharing are expected to follow. If you’re a developer who has spent the last two years rebuilding the same model wrapper for every consuming environment, that breadth matters more than any benchmark. The cynical read: Databricks is making a long bet that being the standard-bearer for cross-platform AI asset sharing is worth more than locking customers in. The optimistic read: they might be right.
FAQ
Q: What is OpenSharing? A: OpenSharing is an open protocol unveiled by Databricks that lets enterprises share AI models, agent skills, dashboards, and unstructured data across platforms without copying or moving the underlying assets. It uses a zero-copy credential vending model so recipients can access assets directly from the provider’s cloud storage using temporary, scoped credentials. Currently a sandbox project under the Linux Foundation AI & Data Foundation.
Q: How is OpenSharing different from Snowflake’s Zero-Copy integrations? A: According to HyperFRAME Research’s Stephanie Walter, Snowflake’s approach allows data to be copied only when both the provider and receiver are on Snowflake. OpenSharing works across platforms and extends the sharing model beyond datasets into AI models, agent skills, dashboards, applications, and unstructured data.
Q: Which connectors does OpenSharing support today? A: Generally available connectors include Python, Apache Spark, Tableau, PowerBI, Snowflake, DuckDB, Clojure, Node.js, Java, Arcurate, Rust, Go, C++, and R. Databricks has signaled that Google Spreadsheet, Excel, Airflow, and Lakehouse Sharing connectors are on the roadmap.
Key Takeaways
- Treat integration tax as a budgeted line item, not invisible overhead — CIOs who can quantify it will be the first to justify protocol-level investments.
- Audit your AI assets the way you audit APIs: which models, skills, and dashboards are candidates for governed cross-platform sharing within the next two quarters?
- Vendors that ship AI assets behind proprietary wrappers will face renewed procurement pressure as enterprises start asking about open sharing protocols in RFPs.
- Expect agent skills to emerge as a distinct, monetizable asset class, separate from the underlying models that power them.
- Snowflake-only or Databricks-only sharing strategies will look increasingly fragile as Linux Foundation governance pulls cross-platform interoperability into the mainstream conversation.