AI has commoditized the two things SaaS used to compete on: features and models. Both can be copied now. What can’t be copied is a proprietary dataset that gets more valuable every time another customer touches the platform, which is why HubSpot, Salesforce, Snowflake, and Microsoft are all racing to become the trusted home for your data rather than just another tool you use. HubSpot’s July stumble and four-day reversal on shared contact data was a preview of the real fight ahead: customers will contribute to these networks only if vendors answer, up front, what gets collected and who benefits. Trust is becoming the feature that decides which platform you join.
For nearly twenty years, the software-as-a-service industry competed on features.
Who had the best CRM. Who had the smartest marketing automation. Who had the easiest user experience.
That formula built some of the largest software companies in history.
AI has rewritten the rules.
Software can be copied now. AI models can be matched. What can’t be replicated overnight is a proprietary dataset that gets more valuable every time another customer touches the platform.
Last week, HubSpot made that shift visible, whether it meant to or not.
The week HubSpot changed the conversation
On July 1, HubSpot rolled out updated terms and a new “Contact Discovery” capability meant to improve contact enrichment across customers. The idea was that business contact information and related enrichment signals would feed a shared dataset, improving prospecting and data quality for everyone in it. HubSpot called the feature opt-in. Plenty of customers experienced it as something closer to opt-out, because the controls were scattered across settings pages and hard to find.
The reaction was immediate.
Customers asked whether data they had spent years building could now improve records used by other companies. Privacy people questioned the consent model. LinkedIn filled up with complaints from marketers, consultants, and CRM admins.
Four days later, HubSpot reversed course.
Chief Product and Technology Officer Duncan Lennox published a post titled “We Got This Wrong. And We Are Fixing It.” It opened with a sentence you almost never see from a public software company: “We made a mistake.”
HubSpot pulled the updated terms and said any future shared-data feature would be rebuilt as genuinely opt-in, with clear disclosure.
The reversal was fast. That’s not the interesting part.
The interesting part is why they tried it in the first place.
AI changed what a moat looks like
Twenty years ago, SaaS companies sold software. Today they’re trying to build data networks.
Features are easy to copy. Foundation models are available from half a dozen vendors now. Code generation keeps shrinking the time it takes to build something.
Data doesn’t work that way.
Every new customer improves identity resolution. Every sales email sharpens engagement signals. Every CRM update makes enrichment more accurate. Every support conversation teaches an AI agent something it didn’t know yesterday.
A dataset that keeps compounding becomes a real advantage. One that’s hard to buy your way around.
But only if customers know they’re contributing to it.
HubSpot isn’t the only one doing this
The rest of enterprise software is heading the same direction.
Salesforce repositioned Data Cloud as Data 360, the layer that feeds Agentforce. The pitch isn’t “we store your records” anymore. It’s “we unify your data so AI agents can act on something trustworthy.” Salesforce keeps describing unified enterprise data as the fuel for AI, and its partnerships with Snowflake and Databricks say the same thing a different way.
Snowflake went from cloud data warehouse to “AI Data Cloud.” Databricks made the same bet on the lakehouse. Microsoft keeps pushing Microsoft Graph as the context layer sitting behind Copilot in Microsoft 365.
Different architecture, same finish line.
Whoever becomes the trusted home for an organization’s knowledge wins the AI era. Everyone building toward that knows it.
The economics are shifting under this
This is also why the old data brokers are getting squeezed.
For years, companies paid real money for proprietary contact databases. AI now makes a lot of that information cheap to find, verify, and enrich on your own.
As raw data gets commoditized, vendors need a different kind of moat, one that keeps improving instead of sitting static on a server. Network effects start to matter more than ownership.
The goal stopped being “sell software.” It became “build an ecosystem where every customer makes the platform better for the next one.”
Trust is the product now
HubSpot’s reversal wasn’t really a PR story. It was a preview of the next governance fight in enterprise AI.
Most customers accept that AI needs data to work. What they want in return is a straight answer to a few questions: What’s being collected? Who benefits from it? Can I actually opt out, or is that theoretical? Does my participation improve only my own AI, or everyone else’s too?
Those questions are becoming product requirements, not fine print.
Companies that answer them up front will earn trust. Companies that bury the answer three settings menus deep will get the same backlash HubSpot got, and they’ll deserve it.
A new question for every CIO
The first question in an enterprise AI evaluation used to be “how capable is your model?”
I think that question is going away. The one replacing it: how does our data make your product better, and who else gets to benefit from that?
That’s the real story here. AI is turning SaaS from software you use into a network you choose to join, and trust is becoming the feature that decides who joins.
Frequently Asked Questions
Why is data becoming a bigger moat than software?
Because software and AI models can now be copied or matched quickly. Features are easy to replicate, and foundation models are available from half a dozen vendors. A proprietary dataset is different. It compounds, getting more accurate and more valuable every time another customer adds to it, which makes it much harder for a competitor to buy their way around.
What happened with HubSpot’s Contact Discovery feature?
On July 1, HubSpot rolled out updated terms and a Contact Discovery capability that would feed business contact information and enrichment signals into a shared dataset used across customers. HubSpot called it opt-in, but many customers experienced it as opt-out because the controls were scattered and hard to find. After four days of backlash, HubSpot pulled the updated terms, admitted it got the change wrong, and said any future shared-data feature would be rebuilt as genuinely opt-in with clear disclosure.
Are other software companies building data networks too?
Yes. Salesforce repositioned Data Cloud as Data 360, the layer that feeds its Agentforce agents. Snowflake rebranded itself as an AI Data Cloud, Databricks made the same bet on the lakehouse, and Microsoft keeps pushing Microsoft Graph as the context layer behind Copilot. The architectures differ, but the goal is the same: become the trusted home for an organization’s knowledge so AI agents have something reliable to act on.
Why does trust matter so much for AI data platforms?
A data network only compounds if customers keep contributing to it, and they will only contribute if they trust how the data is handled. Most customers accept that AI needs data to work. What they want in return is a straight answer on what gets collected, who benefits, whether they can actually opt out, and whether their participation improves only their own AI or everyone else’s. Those questions are becoming product requirements rather than fine print.
What should CIOs ask AI software vendors now?
The old opening question was “how capable is your model?” That one is fading. The question replacing it is “how does our data make your product better, and who else gets to benefit from that?” As models and features commoditize, the honest answer to that question tells a buyer more about the real value and the real risk of joining a vendor’s network than any capability benchmark will.
