Why Tech Giants Are Investing Billions in AI Agents

Amazon is planning roughly $200 billion in capital expenditure this year. Microsoft’s fiscal year 2025 capital spending reached $88 billion — a 58% jump year-over-year — with more growth projected. Google, Meta, and the major cloud providers combined are expected to pour somewhere between $530 billion and $710 billion into AI infrastructure in 2026 alone. Goldman Sachs has repeatedly noted that analysts keep underestimating these figures, only for the actual numbers to come in higher.

The obvious question is: why? What could possibly justify investment at this scale?

The answer isn’t just “AI is exciting.” It’s something more structural — and understanding it explains not just the spending, but the broader transformation these companies are betting on.

The Shift from AI That Talks to AI That Works

For most of the past few years, AI’s commercial value proposition has been relatively modest: better search results, smarter autocomplete, a chatbot that could draft an email faster than you could type it. Useful, but not transformational.

AI agents change that proposition fundamentally. An agent doesn’t just respond to a question — it takes a goal, breaks it into steps, uses tools, calls systems, and executes until the task is actually done. The difference, from a business perspective, is enormous. Chatbots reduce how long something takes. Agents can eliminate the need for a human to do it at all.

That distinction is what’s driving the investment. Tech giants aren’t spending half a trillion dollars because AI is a better search engine. They’re spending it because they believe AI agents represent a genuine shift in how work gets done — and they want to own the infrastructure and platforms that make it possible.

Reason 1: AI Agents Are the New Enterprise Operating Layer

The most lucrative market in technology for the past two decades has been enterprise software — the sprawling, subscription-based stack of tools (CRMs, ERPs, productivity suites, HR platforms) that large organizations depend on to run their operations. That market is now being disrupted from multiple directions at once, and the companies with the resources to respond are moving fast.

The core disruption is architectural. Traditional enterprise software is organized around applications: a user logs into Salesforce, clicks through a workflow, inputs data, and logs out. AI agents invert that model. Instead of users navigating applications, agents navigate applications on behalf of users — reading data, making decisions, taking actions, and reporting back. The software becomes infrastructure the agent operates rather than an interface a human touches.

Gartner projects that by year-end 2026, 40% of enterprise applications will have embedded task-specific AI agents, up from less than 5% in 2025. Deloitte has suggested that up to half of organizations will direct more than 50% of their digital transformation budgets toward AI automation in 2026. The companies building the platforms agents run on — Microsoft, Google, Salesforce, Amazon — are racing to become the foundational layer of this new architecture before it solidifies around a competitor.

Salesforce’s response is illustrative: the company has explicitly declared itself the “operating system for the Agentic Enterprise” and tied executive compensation directly to Agentforce adoption metrics. Its Agentforce product reached $540 million in annual recurring revenue and 18,500 enterprise customers in early 2026 — which the company’s CEO called its fastest-growing product ever, though it still represents a modest slice of overall revenue. The point isn’t the current number; it’s the trajectory and the strategic positioning.

Reason 2: The Per-Seat Model Is Breaking — and the Replacement Is More Lucrative

For decades, enterprise software has been priced on a per-seat basis: you pay per user per month. That model is under serious stress, and the companies most threatened by its collapse are the ones most aggressively investing in agents — because agents offer a far more powerful pricing model.

Here’s the tension: AI agents make individual workers dramatically more productive. A single person equipped with well-deployed agents can handle workflows that previously required a team. That’s great for the companies using AI, but it means they need fewer software seats. The use-of-multi-agent systems spiked by 327% over just four months in 2026, according to a Databricks survey — and companies using them are reducing headcount on specific tasks, which directly reduces the seat count they’re willing to pay for. SaaS stock valuations have felt this acutely, with what some analysts have called a “SaaSpocalypse” wiping significant value from the sector.

The response from major platforms is to shift to outcome-based and consumption-based pricing — charging not for how many people use the software, but for how much work the agents complete. Microsoft has introduced consumption-based pricing alongside per-user fees for Copilot Studio. Salesforce is pioneering an “Agentic Enterprise License Agreement” that bundles all-you-can-eat agent access at a fixed price. ServiceNow is moving toward value-based pricing for its agent offerings.

This shift matters enormously for revenue potential. A company paying $50/month per seat for a team of 100 pays $5,000 a month. An agent platform that automates $500,000 worth of labor per month can command a very different price. The economics of outcome-based pricing, if it takes hold, are far more favorable for platform owners than anything the per-seat model offered — which is precisely why the biggest players are reorganizing around it.

Reason 3: Cloud Revenue Follows Compute Demand

AI agents don’t just run on platforms — they run on infrastructure. Every agent call, every tool execution, every multi-step workflow consumes compute: GPU cycles, memory, storage, and network bandwidth. For the three major cloud providers — Amazon Web Services, Microsoft Azure, and Google Cloud — this is one of the most direct financial arguments for agent adoption.

As agents become standard parts of enterprise workflows, the amount of AI compute consumed by businesses rises accordingly. That compute runs in data centers these companies own and are spending hundreds of billions to expand. The capital expenditure and the commercial rationale are directly linked: build the infrastructure, attract agents onto the platform, and collect the compute revenue every time those agents run.

Amazon’s partnership with OpenAI, announced in 2026, captures this dynamic clearly. AWS became the exclusive third-party cloud distribution provider for OpenAI’s enterprise agentic platform, which enables organizations to build and deploy teams of AI agents. Amazon committed $50 billion to OpenAI. OpenAI agreed to consume substantial capacity through AWS infrastructure. Both parties get what they need: OpenAI gets compute at scale, Amazon gets the workloads that fill its data centers.

Microsoft’s $88 billion in capital spending in fiscal year 2025 — a number the company expects to keep growing — is the infrastructure bet that every Copilot agent interaction eventually monetizes. Google’s commitment to its partner ecosystem includes $750 million specifically to accelerate joint customers’ transformations with agentic AI. The infrastructure spend and the agent strategy are not separate lines on a balance sheet; they’re the same bet.

Reason 4: Agents Create Platform Lock-In That Dwarfs Historical Software Moats

Software platforms have always pursued lock-in: the more deeply a product integrates into a company’s operations, the harder it is to rip out and replace. AI agents amplify this dynamic to an almost unprecedented degree.

An agent platform that has learned a company’s policies, workflows, customer data, and internal processes over months or years isn’t just software — it’s institutional knowledge embedded in a system. Switching away from it means losing that accumulated context and rebuilding it somewhere else. The switching costs are genuinely higher than anything traditional SaaS created, because the value isn’t just in the software’s features; it’s in what the agent has learned about the specific organization.

This is why the platforms investing most aggressively in agents aren’t just chasing near-term revenue — they’re building structural advantages designed to compound over time. Salesforce’s integration of Agentforce across the full customer lifecycle (Sales Cloud, Service Cloud, Marketing Cloud) isn’t coincidental. The more of a company’s operations touch those agents, the harder it becomes to replace the underlying platform. Kore.ai’s multi-agent orchestration engine operates across workplace productivity, customer service, and process automation simultaneously for the same reason.

For organizations evaluating whether to build or buy, the decision increasingly carries long-term implications that go beyond the immediate purchase price.

Reason 5: The Competitive Cost of Losing Is Existential

The final reason behind the investment levels is the one that doesn’t get said as directly in earnings calls: fear.

In previous technology cycles, losing a platform war meant ceding some market share and pivoting. The stakes of the AI agent race appear different because agents are being positioned not just as a product category but as the interface through which all other software is accessed. If one company’s agent platform becomes the dominant operating layer for enterprise work, the companies whose software isn’t natively accessible through that layer lose relevance — not because their software is bad, but because nobody navigates it manually anymore.

That’s the threat model that’s producing $710 billion in combined investment from four companies. It’s not just optimism about AI’s potential — it’s a rational response to the possibility that the company that wins the agent layer wins distribution over everything that runs on top of it.

Goldman Sachs has consistently noted that capex estimates for AI hyperscalers have been too low for two consecutive years. The implied message from the companies doing the spending is that they don’t think this race has a comfortable middle outcome: the winners will define enterprise computing for the next decade, and the losers will operate in a world they didn’t build.

The Legitimate Skepticism

None of this means the bet is certain to pay off. Investors have become more selective about AI spending, asking harder questions about the timeline between capital expenditure and return on investment. Microsoft, Amazon, and Alphabet all saw stock reactions when they announced spending increases in 2026 — markets expressing concern about the gap between infrastructure cost and realized revenue.

The honest answers to those concerns are:

• Many agents still work best within narrow, well-defined tasks; the broader “virtual employee” vision is real but still ahead of current reliability.
• Governance, security, and compliance remain friction points that slow enterprise adoption.
• The shift from per-seat to outcome-based pricing is a transition, not a switch — it will take years to fully reprice the market.
• IBM reported $3.5 billion in savings from enterprise-wide agent deployments, with a 50% improvement in employee productivity — real numbers, but in an organization with the resources to deploy agents at that scale.

The trajectory is clear. The exact speed at which it plays out is genuinely uncertain.

The Bottom Line

Tech giants are investing billions in AI agents because they believe — with good reason — that agents represent the next platform layer in enterprise technology. The companies that own that layer will collect the compute revenue every time an agent runs, benefit from pricing models more favorable than anything per-seat software ever offered, and accumulate switching costs that make the resulting lock-in durable.

The investment isn’t irrational optimism. It’s a calculated race to own the infrastructure and platforms that will sit beneath the next generation of how work gets done — at a moment when the outcome still isn’t settled, and the cost of being late is high.