The six largest U.S. tech players will spend more than $900 billion on AI infrastructure and development in 2025, a sum equal to roughly 3% of U.S. GDP, according to industry analyst Josh Bersin.

That single number, drawn from capital expenditure forecasts for Google, Nvidia, Meta, Microsoft, OpenAI, and Amazon, frames the strategic stakes for HR. If AI investment is now operating at near-macroeconomic scale, the people managing that technology will need a different skill stack than yesterday's line manager.

Bersin, in a September 2025 column on his industry blog, framed the moment bluntly: "I've never seen such a massive, rapid, and optimistic investment in technology as we've seen with AI."

What is the Supermanager?

The label is provocative on purpose. Bersin's thesis is a new archetype: a leader who combines people judgment with hands-on fluency in AI tools, and who treats model outputs the way earlier generations treated quarterly forecasts.

Bersin argues this is not a forecasting exercise but an observable shift, driven by three forces his column links to the $900 billion figure:

• A workforce already experimenting with copilots, transcription tools, and code assistants
• Vendors selling manager-facing AI features — scheduling, performance summaries, coaching prompts — on monthly contracts
• Boards asking which roles will absorb AI work before they authorize headcount (the total number of employees a company pays to keep on its payroll) growth

For HR, that list is an org-design prompt, not a tech roadmap.

What changes for HR if Bersin is right?

The first implication is job description (JD) surgery — a JD being a written account of a role's duties, responsibilities, and required skills. Compared with the manager of 2015, who ran weekly one-on-ones and approved expense reports, the Supermanager decides when a process is delegated to a model and when it stays human. Current JD libraries do not yet capture that judgment call.

The second is compensation (pay structure). If AI capability sits on the manager's desk, salary bands (predefined pay ranges that group jobs of similar value) will bend toward candidates with prompt design (writing structured instructions that steer an AI model's output), workflow automation, and AI-output review. Talent acquisition teams will need to source these skills without poaching the data engineering bench.

The third is performance management. A manager who allocates 30% of a workflow to AI needs new metrics — output quality, model-error rate, hallucination incidents (instances where an AI generates plausible but factually incorrect content). Standard HR scorecards (standardized evaluation forms used to rate employee performance) will need rebuilding.

What should HR watch for in the next 12 months?

Three early signals will tell readers whether the Supermanager framing has teeth:

• A revised manager competency framework (a structured list of skills and behaviors a manager is expected to demonstrate) issued by a Fortune 500 employer or a major consultancy
• A pilot program pairing line managers with embedded AI tools, with HR co-designing the metrics
• Updated employment-law guidance from the Department of Labor or a state agency — the regulatory layer the source does not address, but which compensation committees (governing bodies that set executive pay) will eventually demand

Reading the data carefully

The $900 billion covers capital spending, not HR budgets, and the 3%-of-GDP benchmark is a ceiling measure across six firms, not an industry average. Readers should note that the Bersin post is a column, not a peer-reviewed study, and that the body of the argument sits behind a paywall. The Supermanager concept is a working label, not a defined competency.

That is not a reason to ignore it. It is a reason for HR to ask sharper questions of vendors before signing contracts that lock in manager workflows for three years.

By the time the first wave of corporate AI deployments is fully reflected in headcount data, the Supermanager job description will already be set — by procurement teams, not HR.