The AI "Doing" Era: Amazon and Salesforce’s Billion-Dollar Power Play

The Librarian Just Became an Employee

For the last two years, Enterprise AI has felt like having a really smart librarian. You could ask it to summarize a PDF, draft an email, or brainstorm a marketing plan, but at the end of the day, you still had to do the actual work. You were the one clicking the buttons, moving the data, and filing the reports. That era just ended.

We are officially entering the "Agentic Era." The industry is pivoting from AI that suggests to AI that executes. When Amazon drops a staggering $25 billion into Anthropic and Salesforce builds a "Fabric" to orchestrate AI agents from a dozen different vendors, they aren't just upgrading their software—they are building a digital workforce. This is the shift from the 'Copilot' (the sidekick) to the 'Agent' (the employee).

Amazon’s War on the Nvidia Tax

The headline number is eye-watering: Amazon isn't just investing cash; they've secured a $100 billion AWS spend pledge from Anthropic over the next decade. But if you think this is just a venture capital play, you're missing the forest for the trees. This is a strategic strike against the "Nvidia Tax." For years, every AI company has been paying a massive premium to Nvidia for the GPUs required to run their models. Amazon is tired of paying the landlord, so they're building their own apartment complex.

By co-developing custom silicon like Trainium and Inferentia with Anthropic, Amazon is vertically integrating the entire stack. They own the chips, they own the cloud (AWS), and they have a tight alliance with the intelligence (Claude). This creates a massive advantage in Inference Economics—which is essentially the cost of running a single AI query. If Amazon can slash the cost of a token by 50% using their own chips, they can underprice every other cloud provider in the market.

The Brain Meets the Hands

While Amazon is winning the infrastructure war, Salesforce and UiPath are fighting for the "Execution Layer." Think of it this way: Salesforce provides the Brain (the customer data and reasoning), and UiPath provides the Hands (the ability to actually click buttons in legacy software).

Most big companies are still running on ancient, "green-screen" software from the 90s that doesn't have a modern API. An AI can decide that a customer deserves a refund, but it can't actually log into a 30-year-old banking system to send the money. The Salesforce-UiPath partnership bridges this gap. Through "Agent Fabric," Salesforce can now trigger a UiPath bot to perform the physical action of a wire transfer or a database update. This is the "Model Context Protocol" in action—turning a chat window into a command center.

This shift is fundamentally changing how we measure productivity. We're moving away from "seats sold" (how many people have a login) to Agentic Work Units (AWUs). In this new world, Salesforce doesn't care how many humans use their software; they care how many tasks their agents successfully complete. It's a move from a subscription model to a utility model.

The Rise of the Compute Gentry

However, this rapid expansion is creating a dangerous divide in the market: the "Compute Gentry" versus the "Compute Poor." When a company like Amazon can guarantee 5 gigawatts of power and a fleet of custom chips to a partner like Anthropic, smaller AI startups are left fighting for scraps. Compute is the new oil, and the pipelines are being owned by a very small number of players.

This concentration of power is already triggering red flags for regulators. The EU AI Act is now in full effect, and the FTC is eyeing these "cloud-for-equity" deals. Regulators are worried that the "closed loop" of chips, cloud, and models creates a vertical foreclosure where no new competitor can ever enter the market because they can't afford the electricity or the silicon.

The Bottom Line for Investors

The "hype phase" of AI is officially over. Institutional investors are no longer impressed by a CEO mentioning "AI" in an earnings call. They are looking for Agentic ROI. They want to see "Human Hours Reclaimed" and expanded inference margins.

For the retail investor, the play is clear: watch the orchestrators. The companies that can manage multiple agents from different vendors—the "Fabrics" of the world—are the ones that will own the enterprise. Whether it's Amazon owning the silicon or Salesforce owning the workflow, the goal is the same: make the AI indispensable to the actual operation of the business, not just a fancy way to write a memo.

Check out our Interactive Charting Tool.