Who Actually Makes Money When Tech Giants Spend $200B on AI Infrastructure

Bank of America's top stock strategist just told clients to focus on what he calls "capex takers" -- the companies actually receiving the massive infrastructure spending as tech giants pour capital into AI buildouts. Business Insider reported his note this week, and the core idea is pretty straightforward: when Microsoft, Google, and Meta are writing checks for $200 billion in combined annual spending, somebody's cashing those checks.

That's the setup. The question for traders is figuring out which stocks are positioned on the receiving end and what that tells us about sector rotation over the next 12-18 months.

Here's the shift that matters. For two decades, tech companies operated as asset-light businesses. They didn't build factories. They didn't own power plants. They rented cloud capacity, licensed software, and kept capital expenses minimal. That model is dead now, at least for the hyperscalers building AI infrastructure.

Microsoft, Google, Amazon, and Meta have collectively shifted into capital-intensive mode. They're buying chips, building data centers, signing long-term power contracts, and commissioning custom silicon. This isn't temporary R&D spending. It's structural change in how these companies operate, and it redistributes where the money flows in the supply chain.

The companies writing the checks aren't necessarily the ones seeing stock price appreciation. The companies receiving the checks often are, which is BofA's whole point.

So who are the "capex takers" in this equation? Think about what you need to build a hyperscale AI training facility:

Semiconductor suppliers. NVIDIA is the obvious one, but there's also AMD, Broadcom, and custom chip designers getting direct orders from hyperscalers. When Meta announces a $40 billion capex budget for 2026, a meaningful chunk of that flows to chip suppliers within 90 days.

Data center builders and operators. Equinix, Digital Realty, CoreWeave. The companies that own the physical infrastructure or build it on contract. AI workloads need floor space, cooling, and redundancy. Hyperscalers are leasing capacity and signing multi-year deals at scale.

Power infrastructure. AI training burns electricity at rates traditional cloud workloads never approached. That means contracts with utilities, investments in renewable energy projects, and buildouts of on-site generation. Companies positioned in industrial power distribution and grid capacity are seeing demand they haven't seen in decades.

Networking equipment. Cisco, Arista, Juniper. When you're moving training data between GPU clusters at petabyte scale, you need switches, routers, and fiber backbone upgrades. The networking layer gets less attention than chips, but it's a real bottleneck and a real spending category.

These are the names traders should be mapping out if they're trying to play second-order effects from AI capex instead of betting directly on the hyperscalers themselves.

The bullish case for capex takers is solid if spending continues at current levels. But there's a big "if" buried in there, and it's worth being honest about what could derail this.

Spending could plateau faster than expected. AI infrastructure buildouts don't go on forever. At some point, Google and Microsoft reach sufficient capacity for current models, and capex drops back to maintenance levels. If that happens in 2027 instead of 2029, the revenue visibility for suppliers compresses fast.

Margins might not hold. High demand usually means pricing power, but if every hyperscaler is building at the same time, suppliers face pressure to compete on price instead of just fulfilling orders. NVIDIA's gross margins are insane right now, but competitors are coming and hyperscalers are designing their own chips to reduce dependency.

Geopolitical and regulatory risk. Chip export controls, data sovereignty rules, and grid capacity constraints in key markets all introduce variables that can slow buildouts or shift where spending flows geographically. A policy change in the U.S. or Europe could redirect billions overnight.

And then there's the bigger structural question: what happens if AI model development hits a wall and the incremental returns from throwing more compute at training diminish faster than expected? The entire capex thesis depends on hyperscalers believing the next model will justify the next $50 billion in spending. If that belief cracks, so does the trade.

The capex taker thesis is essentially a bet on sector rotation away from software and consumer-facing tech toward industrial suppliers and infrastructure plays. That's a real shift in where money flows within the tech ecosystem, and it shows up in how different baskets of stocks move relative to each other.

If you're tracking this with something like XLK (the tech sector ETF), you'll notice the performance spread between software names and semiconductor equipment names has widened significantly over the past 18 months. That gap is the capex rotation playing out in real time. The companies building tools are outperforming the companies using them, at least for now.

What traders need to watch is whether that rotation continues or reverses. If hyperscaler stocks start outperforming their suppliers, it signals the market expects capex to plateau and profit margins to expand. If suppliers keep leading, it means the market still believes in the buildout thesis and expects more checks to get written.

There's no crystal ball here, but the behavioral gap between analysis and execution matters a lot in sector rotation trades. Most traders see the thesis, agree with it, and then do nothing because the names aren't the ones they're used to trading. Sticking with familiar large-cap software names while semiconductor equipment stocks are where the actual money flow is happening costs opportunity, even if the familiar names aren't going down.

Looking at the actual price action, semiconductor stocks and infrastructure plays have been in a pretty clean uptrend since late 2024. There's been volatility, obviously, but the higher lows and higher highs are there. Volume patterns confirm institutional accumulation, especially in names directly tied to AI hardware.

That doesn't mean it keeps going. Trends reverse, and when they do in this sector, they tend to reverse hard because these stocks are momentum-driven and institutionally crowded. But as of right now, the structure supports the thesis. Money is still flowing into the capex taker names, and there's no clear distribution pattern forming yet.

What would change that read? Watch for divergence between price and volume. If semiconductor stocks start making new highs on declining volume while hyperscalers bounce on heavy accumulation, that's a rotation signal worth paying attention to. It means the market is repositioning for the next phase, which probably looks like profit-taking in infrastructure and reallocation back into the companies that will monetize the AI models once they're built.

For now, though, the structural read is still bullish on the supply side. The capex is real, the contracts are signed, and the revenue is showing up in earnings.

If you're trying to play the capex taker thesis, here's the practical stuff that matters:

Map out exposure in your portfolio. If you own QQQ or XLK, you already have indirect exposure to both hyperscalers and suppliers, but the weighting matters. Check how much of your tech allocation is in the companies spending versus the companies receiving. Rebalance if the ratio doesn't match your view.

Watch earnings closely. Revenue guidance from semiconductor equipment companies and data center operators tells you more about the forward trajectory of AI spending than any strategist note. If ASML or Arista starts guiding down, the thesis is cracking.

Don't chase. A lot of the capex taker names have already run hard. Buying at all-time highs without a clear entry setup is how you get caught in the first 15% pullback. Wait for structure, wait for a level, and size positions accordingly.

Consider options strategies if you're playing second-order names. Semiconductor equipment and power infrastructure stocks are less liquid than the hyperscalers, which means wider spreads and more gap risk. If you're trading these, think about how to structure positions that limit downside while keeping upside open.

And honestly, if the whole thesis feels too crowded or too obvious, that's a valid signal. Sometimes the best trade is recognizing when a theme is fully priced in and sitting it out until the market gives you a better entry or a different opportunity.

The capex is real. The money is moving. The question is whether the stocks on the receiving end still have room to run or whether the easy money already got made in 2025. That's not something a strategist note can answer. It's something the chart will tell you if you're watching the right levels.