The four companies building AI - Microsoft, Google, Amazon, and Meta - plan to spend close to $700 billion this year on infrastructure. That's up from roughly $400 billion in 2025, and Goldman Sachs projects the number could reach $1.6 trillion annually by 2031. The headline number is staggering, but it doesn't tell you where to invest. It tells you where the money is moving. The question isn't whether the buildout is real. It's whether the stocks selling into the pipe have already priced in everything the pipe delivers.
The way to think about this is not as a single AI trade but as a factor stack across the companies that actually get paid when hyperscalers spend. You need the GPU, the networking, the cooling, the custom silicon, and the power. Each layer has a different growth rate, a different margin profile, and a different valuation starting point. I'll walk through five names at the center of this stack: Nvidia, Broadcom, Arista Networks, Vertiv, and a wildcard. The goal isn't to crown a winner but to show which factor grades are still supporting the price and which are deteriorating.
1. Nvidia (NVDA) - The anchor
Nvidia reported fiscal year 2026 revenue of $215.9 billion, with Q4 data center revenue alone at $62.3 billion and non-GAAP gross margins of 75%. For the current fiscal quarter (Q1 FY2027, ended April 2026), revenue grew 85% year over year. The consensus forward P/E sits around 31x. That is not cheap for a semiconductor company, but it's also not the absurdity it sounds like when the denominator is growing at 85%. For context, 31x earnings at 85% growth means the price-to-sales multiple compresses toward reason in fewer than two years if growth holds.
The constraint on Nvidia isn't demand. It's supply - specifically TSM's ability to manufacture advanced-node GPUs fast enough to keep pace. The question the market is watching now is whether Nvidia's Q2 results will show the ramp holding or beginning to flatten. If growth drops below 50% while the stock still carries a 31x multiple, the factor stack cools. If it stays above 60%, the rating holds despite warm valuation. This is your growth sleeve anchor: the name that benefits from every dollar of the $700 billion, with margins that let it absorb competition.
Role: Growth sleeve core. Buy the dips, trim if quarterly growth falls below 50% while PE stays above 30x.
2. Broadcom (AVGO) - The momentum fracture
Broadcom just reported Q2 FY2026 results: total revenue of $22.2 billion, up 29% year over year, with AI semiconductor revenue of $10.8 billion - up 143%. That's a record quarter. The stock fell 13.4% on the report. When a company reports 143% AI revenue growth and the stock drops double digits, the message is clear: expectations had outrun even Broadcom's numbers.
The Q3 guidance miss is the critical detail. Broadcom told investors to expect roughly $22 billion in Q3, and the market read that as below the internal bar it had set over the past three months of analyst conversations. For a company whose custom AI chip business is the fastest-growing segment, a guidance disappointment after triple-digit revenue growth signals a market that priced perfection into the shares. The stock is up 28% in 2026, which is solid, but the PHLX Semiconductor Index is up 74% - Broadcom is underperforming its own sector despite the AI tailwind.
The trajectory here has shifted. Broadcom moved from "hidden AI compounder" to "everyone knows Broadcom does AI now" in roughly six months. The valuation still carries room if custom silicon revenue continues climbing, but the factor grade on revisions has turned negative after this quarter. That matters because revision momentum is usually the timing signal on high-PE names. It's turned off.
Role: Hold if already owned, wait for the revision cycle to restabilize before adding. The catalyst that changes this is a Q3 beat that restores the upgrade cadence.
3. Arista Networks (ANET) - The networking squeeze
Arista builds the Ethernet switches and networking infrastructure that keep AI data centers functioning. In Q1 2026, revenue was $2.71 billion - up 35.1% year over year and above the $2.6 billion guidance mark. The company raised its full-year 2026 revenue growth target to 27.7% from 25%. On the surface, that's a strong report. The stock still sold off.
Why? Because the guidance increase was modest relative to what the stock had already run in. Arista trades at a TTM P/E of roughly 58x, up from a 12-month average of around 50x. The valuation has worsened even as growth continues. In our book, a P/E expansion of 17% above the trailing average is a signal that the stock has absorbed the next two quarters of AI networking growth into the share price. AI center revenues are being guided to $3.25 billion, which is meaningful, but the denominator is large.
The factor grade on valuation is degrading while the growth grade stays solid. That mismatch is the exact setup where a stock transitions from Buy to Hold: the business is doing the right things, but the price demands more right things than the business can deliver before the number becomes self-correcting. Arista belongs in the networking-and-switching sleeve of an AI infrastructure portfolio, but at 58x, you're paying for 30%+ growth that needs to compound, not just sustain.
Role: Hold on paper, buy on weakness below the 50x P/E range where the growth-to-valuation ratio restabilizes.
4. Vertiv (VRT) - The power-and-cooling premium
Vertiv builds the cooling and power management systems that keep GPU clusters from literally melting down. Q1 2026: revenue of $2.65 billion, up 30% year over year; operating profit up 51%; diluted EPS growth of 136%. The company raised full-year guidance. It's been one of the most explosive runners in the AI infrastructure space - up roughly 550% over the past two years.
The P/E ratio tells you where the excitement has landed. Vertiv trades at roughly 83x trailing earnings (TTM). That is a valuation that requires continued triple-digit earnings growth to justify. Q1 delivered 136% EPS growth, which covers the multiple, but the next quarter needs to stay in that range or the P/E begins looking backward rather than forward. At 83x, you're pricing in two years of 50%+ EPS growth from a $2.65 billion quarterly revenue base.
Vertiv's factor stack is interesting because the growth and profitability grades are still strong, but the valuation grade has moved from "justified by hypergrowth" to "fragile if growth normalizes." A company like Vertiv doesn't need to slow down catastrophically - it just needs to slow down from 136% to something like 40-50% for the multiple to look stretched. The portfolio implication is clear: this is a stock you hold through the hypergrowth phase and reduce when consecutive quarters show EPS growth in the 40-60% range. The momentum is still there, but the margin for error is thin.
Role: Growth sleeve satellite with a defined exit trigger: two consecutive quarters of EPS growth below 50% = trim position.
5. The Wildcard: Semiconductor Equipment (ASML or AMAT)
I'll flag semiconductor equipment as the underpriced layer. SEMI forecasts global semiconductor manufacturing equipment sales reaching $133 billion in 2025, $145 billion in 2026, and $156 billion in 2027. The equipment builders get paid before the chip designers. When hyperscalers announce $700 billion in capex, that money eventually flows into more fabs, more lithography machines, more deposition tools. Companies like ASML and Applied Materials don't headline the AI conversation, but they collect on every expansion cycle.
The structural point is that equipment revenue is lumpy and forward-looking in a different way than chip revenue. Orders spike ahead of fabrication, so a $700 billion spending cycle shows up in equipment backlogs months before the GPUs ship. The valuation multiples on equipment names tend to run lower than GPU designers, which means the growth-to-valuation ratio often favors the equipment layer. I don't have a specific pick to lock in here - it depends on the specific order backlog and regional exposure at each name - but the equipment layer is where you look when the pure-play AI names start looking like they've front-ended the cycle.
Role: Barbell counterweight. When Nvidia trades at 31x and Vertiv at 83x, equipment names at lower multiples offer a structural hedge on the same spending cycle.
What the factor stack actually says
Put together, the $700 billion AI spending number creates a hierarchy, not a uniform bid across the sector:
- Nvidia still has the strongest growth-to-valuation ratio at 85% revenue growth and a 31x forward P/E. It's the anchor. The trigger that changes this is a quarterly growth rate below 50% while the multiple stays elevated.
- Broadcom has the clearest deterioration signal: record 143% AI revenue growth followed by a 13% stock drop on guidance. The revision momentum has flipped. Watch for the Q3 print to determine whether this is a one-quarter blip or a trend.
- Arista is caught between solid execution and a P/E that has stretched past its 12-month average. Growth is the friend, valuation is the constraint.
- Vertiv is the highest-conviction growth story wrapped in the most fragile valuation. 83x P/E with 136% EPS growth is sustainable only if the EPS growth stays near 100%. The second it normalizes to 50%, the P/E demands correction.
- Semiconductor equipment is the unpriced layer - lower visibility, lower multiples, same spending cycle flowing through it.
The $700 billion number isn't the story. The story is which layer of the stack has the best factor alignment right now - and which has already absorbed the cycle into the price. Nvidia still has room. Broadcom needs a revision reset. Vertiv needs continued hypergrowth or it gets repriced. The equipment layer needs nothing from the market but patience. That's the actual portfolio map this spending cycle produces.