What We Learned on the MFS Technology Bus Tour: AI’s COGS Shock

In brief

• AI spending forecasts already look extraordinary, but recent MFS fieldwork suggests they may still be too low.
• Bottlenecks in memory, power, equipment and infrastructure are causing higher COGS (Cost of Goods Sold) across many industries. 
• The alpha opportunity is not only identifying durable beneficiaries of AI spending but also avoiding companies whose returns may fall because they lack the pricing power to pass through rising costs.

MFS recently held its annual US technology bus tour, during which 29 MFS investors, including both technology specialists and generalists, visited public and private companies across the AI ecosystem.

The mix of participants matters. AI is no longer just a technology-sector issue — it’s become a cross-sector capital, cost and competition issue. Putting specialists, generalists and portfolio managers in the same room with management teams and engineers helps us pressure-test assumptions, understand supply chains, debate outcomes, and separate durable economics from extrapolated enthusiasm.

These meetings affirmed that demand for AI inputs will be meaningfully outstripping supply into 2027. But the more important takeaway for me, wearing both my portfolio manager and global investment strategist hats, was that AI demand and spending are expanding beyond hyperscalers and neoclouds. Estimates may be too conservative, and the resulting pressure on input costs may be underappreciated in both inflation forecasts and corporate profit assumptions.

One striking example involved compute in space. To be clear, this was not about commercial launch. It was about defense and lunar infrastructure. If geopolitical competition extends beyond Earth, satellites and lunar systems will need data, memory, and inference closer to where decisions are made. That’s a sentence I never expected to write, but it captures the broader point: that AI is creating demand for compute infrastructure in almost every direction. While I would prefer a simpler world, scientific progress does not care what I want, or what I fear.

In sum, the world is not just short chips. It is short entire, massive, physical stacks.

In a piece earlier this year titled “Memory Is the New Oil”, we argued that memory was becoming one of the key bottlenecks in the AI buildout. That proved true faster than almost anyone expected, but the issue is now broader. Memory scarcity is moving from a technology constraint to an economic transmission mechanism.

Similar to how higher oil prices translate into higher prices at the gas pump and force consumers to cut spending elsewhere, memory and other AI bottlenecks will ultimately show up in cost of goods sold, or COGS, and pressure profits for companies far from Silicon Valley. This matters for investors, particularly passive investors, because much of the corporate world was built around the opposite assumption.

For decades, investors became accustomed to semiconductor deflation. Typically, more performance arrived at lower cost. Many business models assumed that technology inputs would become cheaper over time, but the race to build compute is upending that assumption. High-bandwidth memory, server DRAM, enterprise storage, and more are being pulled toward AI infrastructure, while traditional buyers like makers of PCs, smartphones, autos, networking equipment, industrial hardware, and other chip-based consumer devices compete for what remains. What looks like strategic investment for one buyer may become immediate gross-margin pressure for another.

This is why the design of our bus tour was so critical. Technology specialists were able to map the bottlenecks while generalists and portfolio managers asked the next natural question: “whose P&L gets hit?” 

Companies facing rising costs have a few simple choices. They can raise prices, absorb the costs, or attempt some combination of the two. 

Raising prices tests demand elasticity. Consumer staples offered a recent reminder. Many companies raised prices despite limited product innovation, only to discover that consumers can substitute, which weakened volumes and caused stock price multiples to fall. Companies that keep prices fixed may retain customers but are more likely to disappoint investors as margins fall.

I don’t believe the market’s broad AI narrative captures this nuance. A company can use AI to generate efficiencies but still be harmed by AI-driven input inflation. Productivity gains from technology tend to get competed away, and I doubt AI will be any different. A company can benefit from productivity improvements in one area while suffering COGS pressure in another, all while spending heavily to defend market share.

The real differentiator is economic power.

Businesses with differentiated products, mission-critical use cases, or trusted brands may be better positioned to seek to preserve returns. But companies selling more substitutable products into price-sensitive end markets may find that the customer, not spreadsheets, determines how much inflation can be passed through, making return-on-capital assumptions fragile.

Headline inflation may understate this pressure because many affected products carry small index weights, but producer prices, AI spending, and product availability may tell a different story. Often in paradigm shifts, averages conceal the changes that matter most.

Benchmarks and portfolios tethered to the past don’t distinguish between companies that can raise prices and those that can’t. Indices don’t separate businesses strengthened by AI from those whose margins are eroded by AI-related competition and input inflation. Passive exposure owns both sides of AI-induced inflation.

This doesn’t make active management easy, as recent performance reminds us, but it does make underwriting more important.

The conclusion from our tour is that AI is both a demand and a cost shock. It’s raising input  costs across industries while also lowering barriers to entry and enabling new competitors.  We feel that combination should increase dispersion between companies with real economic power and those whose returns may prove less durable than markets assume. This is where, I believe, active management will drive value. A benchmark owns both outcomes, while active managers can underwrite which businesses have the supply access, pricing power, and capital discipline to pursue more durable returns compared with those that are simply carrying yesterday’s margins into a different regime. 

In this environment, avoiding the future victims of AI-induced inflation may be as important as owning the most obvious AI beneficiaries.

Keep in mind that all investments carry a certain amount of risk, including the possible loss of the principal amount invested.

The views expressed are those of the author(s) and are subject to change at any time. These views are for informational purposes only and should not be relied upon as a recommendation to purchase any security or as a solicitation or investment advice. No forecasts can be guaranteed. Past performance is no guarantee of future results.