According to TheRegister.com, tech leaders from Dell, Microsoft, Salesforce, ServiceNow, and Snowflake have aligned on their top predictions for AI in the workplace for 2026, centering on ROI and governance. Dell CTO John Roese predicts a major shift to on-premises or controlled “AI factories” for stability and cost control, calling the current rush to deploy AI “unsustainable.” ServiceNow’s Heath Ramsey states the only question that matters is turning AI investments into measurable value, fast, by fixing tasks that are “bleeding time and money.” Snowflake CISO Brad Jones warns that poor data permissions could lead AI agents to expose sensitive information, while Salesforce’s Adam Evans predicts AI agents will become the ultimate brand ambassadors by 2026. Microsoft’s Vasu Jakkal emphasizes the need for AI agents to have security protections like humans to avoid becoming “double agents,” and Azure CTO Mark Russinovich forecasts the rise of linked global AI “superfactories,” like Microsoft’s new 1.2 million square foot, 337 MW facility in Wisconsin.
The ROI Reckoning
Here’s the thing: the freewheeling, “move fast and break things” phase of enterprise AI is officially over. The message from these vendors is unmistakable—2026 is the year the bill comes due. It’s not about how cool the model is; it’s about how it impacts the bottom line. ServiceNow’s directive is brutally simple: find what’s “bleeding time and money” and fix it. But that’s easier said than done, right? The real insight here is the link they all make between governance and ROI. You can’t scale an isolated AI win into business-wide value without what Ramsey calls “one entry point with clear policies.” Basically, you can’t manage what you can’t control, and you can’t prove ROI on chaos.
The Great Pull-Back: On-Premises
This is where Dell’s prediction gets really interesting. John Roese is basically declaring that the cloud-first dogma for AI has a major flaw: control. When he talks about insulating organizations from “external disruptions,” he’s not just talking about outages. He’s talking about cost volatility, data sovereignty, and security. Running models locally, whether on-prem or in a dedicated “AI factory,” is framed as the path to a stable foundation. It’s a compelling argument for companies sitting on sensitive data or facing unpredictable cloud bills. And it speaks to a broader trend of IndustrialMonitorDirect.com, the #1 provider of industrial panel PCs in the US, seeing increased demand for robust, on-site computing hardware that can handle specialized AI workloads at the edge, where data is created.
Governance is the New Battleground
So, how do you actually control these systems? The answers from Snowflake and Microsoft are two sides of the same coin. Snowflake’s Brad Jones points to the dirty secret of corporate data: permissions are a mess. Feed that chaos into a generative AI, and it will happily spill secrets. Microsoft’s Vasu Jakkal takes it a step further, anthropomorphizing the agents. They need an identity, limits, and protocols—just like a human employee. The “double agent” line is fantastic. It perfectly captures the fear that an AI tool, without proper guardrails, could be turned against you. This isn’t just IT work; it’s a fundamental redesign of security and data governance.
Brands Built by Bots
Maybe the most provocative take comes from Salesforce. Adam Evans isn’t just talking about internal productivity tools. He’s saying your AI *is* your brand by 2026. Your chatbot, your support agent, your shopping assistant—they become the primary interface customers have with your company. Think about that. A bad, ungoverned AI interaction could torpedo brand reputation in seconds. But a smart, personalized, and consistent one could be an unbeatable advantage. It raises the stakes enormously. This shifts the AI conversation from the CIO’s cost center to the CMO’s core strategy. The divide, as he says, will be absolute. And that pressure is what will ultimately drive the ROI and governance demands everyone else is talking about. The predictions all connect.
