Big Tech Is Spending $700 Billion on AI. Here's How That Money Reaches Your Business.

Amazon is spending $200 billion on AI infrastructure in 2026. Alphabet is spending $175–185 billion. Microsoft, $145–150 billion. Meta, $115–135 billion. Combined, the four hyperscalers are putting roughly $700 billion into AI — a 60% increase over 2025 levels, funded by $575 billion in combined net cash flow.

That number is large enough to feel abstract. It shouldn't.

The right question isn't what Big Tech is building. It's when that spending becomes your tools — and what you pay for them. Based on a Goldman Sachs report published this April, the answer to both is: now, and less than you think.

From Seats to Tasks: The Goldman Sachs Signal

In April 2026, Goldman Sachs published a research note based on conversations with approximately 40 software and internet companies. The finding: AI software pricing is actively shifting from per-user models to "units of work" — meaning you pay for what the AI does, not how many people have access to it.

This is not a projection. It's already live in the market.

Salesforce has introduced "agentic work units" and agentic enterprise license agreements (AELAs), pricing tied directly to tasks completed by AI agents rather than seats occupied by humans. Workday is selling credits denominated in "units of work." Sam Altman has been explicit about OpenAI's long-term model: AI priced like electricity, metered by the token, consumed like a utility.

The business logic is sound on both sides of the table. For vendors, it decouples revenue from AI running costs and opens new budget lines — IT, operations, finance — that were never in play under seat-based models. For buyers, it removes the penalty for being small. A 20-person company is no longer locked out of enterprise-grade tooling because it can only justify 20 seats. It pays for the work the AI completes, at whatever volume makes sense.

That's a structural advantage for the $1M–$50M company. Enterprise pricing was built for enterprise headcount. Task-based pricing isn't.

The Trickle-Down Is Real This Time

The standard objection to "this technology will reach you" stories is timing. The Gartner data this cycle is more specific than usual.

Gartner's April 2026 analysis estimates that 40–45% of hyperscaler capex value is now reaching accessible SaaS layers — tools you can buy through existing platforms without a custom implementation. The total AI software market sits at $452 billion, representing 18% of $2.52 trillion in total AI economic activity. AI services add another $588 billion.

The relevant number for most SMBs is narrower. IDC's 2026 SMB digital landscape report identifies augmented customer service as the primary entry point — a $35.9 billion market — with tools increasingly embedded in platforms businesses already operate. The sophisticated automation that cost seven figures to implement three years ago is now available for $20–30 per month inside existing CRM, communication, and operations stacks.

Usage data confirms the landing. Between 82% and 89% of SMBs are now using at least one AI tool daily. That's not aspirational — it's operational. The infrastructure has arrived. The question is whether you're using it strategically or incidentally.

What This Means for Your Budget

Per-seat pricing was always a headcount tax. As your team grows, your software costs scale linearly — regardless of how much work those seats actually produce. AI tools under the old model inherited this logic, which meant small teams paid full price for partial leverage.

Per-task pricing inverts the relationship. You pay proportionally to output, not to employment. For a 15-person company competing against a 150-person competitor, that's not a marginal improvement — it's a different game.

The budget math is already showing up in productivity data. Fortune and Goldman Sachs both cite a consistent figure from AI-adopting companies: workers save 40–60 minutes per day. At a fully-loaded cost of $60/hour, that's $40–$60 in value per employee per day. For a team of 15, that's $600–$900 per day, or roughly $150,000 annually, at scale.

That's the ROI case. The pricing shift is what makes it achievable without a dedicated AI budget line.

One more signal worth watching: Forrester's 2026 AI Reckoning report finds that 25% of enterprise AI spend is currently being deferred for ROI validation. Vendors are aware of this. The competitive pressure to demonstrate value — and to offer pricing structures that lower adoption risk — is real and active right now. That's leverage for buyers. It won't last indefinitely.

The Vendor Lock-In Window Is Open — for Now

Here's the part most operators are missing: the pricing competition among AI software vendors is a temporary condition.

Right now, Salesforce, Workday, Microsoft, ServiceNow, and a dozen SaaS platforms are competing for AI market share. They want to be your AI infrastructure — your default layer for automated work. That competition is why task-based pricing exists at all: it's easier to get a customer to try usage-based pricing than to commit to a large seat expansion upfront.

Once these vendors establish dominance in their categories, the pricing flexibility contracts. You've seen this before — cloud infrastructure in 2014 was fiercely competitive and offered significant discounting. By 2019, pricing power had consolidated. AI software is earlier in that cycle, but the trajectory is the same.

The SMBs that move in this window — locking in enterprise-grade capabilities under adoption-friendly pricing structures — will have a meaningful cost advantage over those that wait for "the right time." In technology markets, the right time is usually visible only in retrospect.

The Playbook: Three Moves for the Next 90 Days

1. Audit your per-seat AI costs. Pull every subscription with an AI component. Flag which are charging per seat vs. per use. For seat-based tools, find out whether the vendor has a usage-based tier — most do now, even if it's not prominently marketed.

2. Negotiate agentic pricing. If you're a Salesforce or Workday customer and you haven't been offered units-of-work pricing, ask specifically for it. Use the Goldman Sachs report as context — your account manager knows this is the direction their company is moving. The negotiation window exists precisely because vendors want early adopters locked in before the market matures.

3. Budget for outcomes, not tools. Start tracking AI spend against tasks completed and time saved — not licenses purchased. This reframes AI from a cost center to a productivity input, makes ROI visible, and gives you the data to negotiate better terms at renewal.

---

Bottom Line: $700 billion in hyperscaler spending is a tech story. The shift from per-seat to per-task pricing is a business story — specifically your business story. For the first time, the economics of AI structurally favor smaller operators. The window where vendors compete for your adoption, on terms that benefit you, is open. It won't be open indefinitely.

If you want help auditing your current AI spend or identifying where task-based pricing applies to your stack, we do this work with clients directly.