Why Salesforce is Hoarding Your Slack Messages
Sign up for ARPU: Stay ahead of the curve on tech business trends.
Salesforce-owned Slack, the ubiquitous workplace messaging app, recently made a quiet but significant policy change. According to a Reuters report, the company blocked other software firms from searching or storing Slack messages, even if their own customers grant them permission. The stated reason was to protect customer privacy. But this move signals a much larger shift in the tech landscape, highlighting a new front in the AI arms race where proprietary data has become the most valuable resource of all.
Why is enterprise data suddenly so valuable?
For years, AI companies trained their massive models by scraping the public internet—a seemingly infinite trove of text and images. But as models have grown exponentially, a new reality has set in: the industry is hitting a "data wall." As one analyst put it, we have "already exhausted the world's accumulated set of high quality training data." Public data from the web is often of low quality, and the best sources have already been thoroughly mined.
This scarcity makes unique, high-quality datasets immensely valuable. Corporate data—like the trillions of contextual, professional conversations, project plans, and customer interactions stored within Slack—is a goldmine. Unlike the random chaos of the open internet, this data is structured around specific business workflows, making it the perfect fuel to train AI models that can automate enterprise tasks, a core ambition for Salesforce and its competitors. By restricting access, Salesforce is effectively locking down its own private data well.
How does this create a competitive "moat"?
As foundational AI models from companies like OpenAI, Google, and Anthropic become increasingly capable and, in some respects, commoditized, the real competitive advantage is shifting. The new "moat" isn't just about having the biggest or best general-purpose model; it's about having the best data to fine-tune that model for specific, high-value tasks.
By preventing AI rivals from training their models on Slack's rich conversational data, Salesforce is building a defensive wall around its own AI ambitions. It ensures that only Salesforce has the ability to leverage that data to build superior AI agents for the workplace. This strategy could give its Einstein Copilot a distinct advantage in understanding and automating the nuances of corporate communication, making it harder for competitors to match its performance in the lucrative enterprise market.
Who else is building these data walls?
Salesforce is not alone in recognizing the strategic value of proprietary data. This is a tactic being deployed by major tech platforms across the board:
- Microsoft leverages data from its massive ecosystem, including LinkedIn's professional network and the vast repository of code on GitHub, to train and refine its Copilot and other AI offerings.
- Google's two-decade dominance in search has given it an unparalleled dataset on human intent and knowledge, which is the foundational asset for its Gemini models and AI-powered search features. The company also owns YouTube, which is the ultimate repository of video data.
- Apple, with its intense focus on privacy, is pursuing an "on-device" AI strategy. While limiting its access to broad cloud data, it creates a moat around the personal data on a user's iPhone (photos, messages, calendar), making Apple Intelligence uniquely capable of performing personalized tasks that competitors can't access.
Each of these companies is transforming its platform from a service into a proprietary data source, creating a series of walled gardens that will define the next era of AI competition.
What does this mean for the future of AI?
The trend toward walling off data suggests a future where AI development becomes more fragmented. Instead of a single, all-powerful AI, we may see a landscape of specialized models, each excelling in the domain where its owner has the best proprietary data. The best AI for coding might come from Microsoft/GitHub, the best for general knowledge from Google, and, if Salesforce has its way, the best for workplace productivity will come from Salesforce. For businesses, this means that choosing a software vendor will increasingly mean choosing a data ecosystem, a long-term bet on which "walled garden" will yield the most intelligent and useful AI for their specific needs.