From Cheap Labor to Cheap Compute: Management’s New Quest
The margin game just moved from wages to watts and tokens.
For decades, management teams have lived with a familiar pressure: protect the margin. When revenue growth slowed or wages climbed, the answer often started with labor. Where can the work be done more cheaply? Which functions can move? Which teams can be resized, relocated, outsourced, or reorganized? Corporate strategy has spent a generation staring at payroll the way a dog stares at a steak.
That old playbook was labor arbitrage. Call centers moved overseas. Back-office departments migrated to lower-cost cities. Software support, billing, processing, customer service, and routine administrative work followed the same logic. The work went wherever the labor was qualified and inexpensive enough. It was a geographic answer to an economic problem, and for many companies it became one of the most reliable ways to defend profitability.
AI introduces a quieter shift. The focus begins moving from managing labor costs to managing compute costs. Once a task can be turned into a prompt, workflow, software agent, or repeatable AI process, management starts asking a different set of questions. How much intelligence does this task require? Which model is good enough? How efficiently can the work be routed? How much power, data, cooling, and token usage are needed to produce the same outcome at a lower cost?
That is the economic heart of this story. The margin game is moving from wages to watts and tokens.
According to Stanford’s 2025 AI Index, the cost of GPT-3.5-level intelligence has fallen roughly 280 times since late 2022. Payroll budgets do not behave like that. Nobody is walking into an annual review and politely volunteering for a 280-fold pay cut. Even corporate America has limits, though it occasionally does its best to hide them behind a PowerPoint deck and the word “optimization.”
This is why the investment question reaches well beyond the frontier model companies. OpenAI, Anthropic, Google, Meta, and others matter, of course. But the broader question is who helps make intelligence cheaper to produce. That lens pulls in chips, networking, power, data centers, cooling, water treatment, energy infrastructure, and the software layers that optimize all of it. In an AI economy, the plumbing becomes strategically important. Sometimes the steadiest money is made by the companies selling the pipes, pumps, valves, and electricity to everyone chasing the miracle.
There is another labor-cost story coming, and it deserves a brief mention: robotics. The next wave of automation will push this same search for efficiency into the physical world. Warehouses, manufacturing floors, logistics networks, restaurants, farms, hospitals, and eventually households will all see more machines doing work that has historically required people. That trend matters. It may become enormous. For this discussion, though, the immediate and investable shift is the one happening inside the data center, where the cost of intelligence itself is being compressed, optimized, and priced in real time.
Today, we hear plenty about massive token bills and runaway compute budgets. That phase is real, and it should get more efficient over time. The history of technology is a long story of exotic miracles becoming cheap utilities. The first version amazes people. The profitable version gets optimized, compressed, streamlined, and eventually taken for granted. Somewhere along the way, the miracle gets a monthly invoice, a procurement department, and three people arguing about it in a conference room.
That is why compute arbitrage could become the next great management cycle. Work can move to a cheaper model, a more efficient workflow, or a better-optimized data center. The old outsourcing boom required geography. The new one may require better routing, cheaper inference, lower power costs, more efficient chips, smarter software layers, and infrastructure built for scale.
For businesses, this could meaningfully change margin structures. Customer service, compliance, reporting, coding, data entry, research, legal intake, marketing, scheduling, and administration are all in play. Smaller teams using AI workflows may be able to do work that once required far larger headcounts. Some companies will use AI to widen margins and improve service. Others will discover that their “platform” was mostly expensive human time wrapped in a software subscription.
That is where the investment story gets interesting. The AI question has moved beyond basic usefulness. The sharper questions are economic. Where do the profits settle? Who captures the productivity gains? Who supplies the critical infrastructure? Who protects margins? Which companies gain operating leverage, and which companies spend the next decade explaining why customers should keep paying yesterday’s price for yesterday’s workflow?
AI is often framed as a technology story, but markets eventually reduce most stories to margins, cash flow, and return on capital. If intelligence keeps getting cheaper, the companies that harness it well should enjoy powerful operating leverage. Infrastructure providers may ride a major capital cycle. Legacy models built around high-cost human effort will face pressure to justify old pricing.
This shift is still early. There will be hype, overbuilding, disappointments, and plenty of investor decks with “AI” added so aggressively you can almost hear the copy machine sweating. Beneath the noise, the economic realignment is real. Management’s new quest is the production of cheaper intelligence. When a major margin story emerges early enough, it can become a compelling investment story.
If you would like to discuss how we are approaching the AI trade, the supporting infrastructure, and the portfolio implications of this shift, you can email me at Jeremiah.Bauman@LPL.com or call me at 661-302-4531. These setups are worth examining before they become obvious to everyone else.