Big Tech's combined AI spending is on track to top $700 billion this year. Google alone plans to double its 2025 capital expenditure - $72 billion - to between $125 billion and $145 billion in 2026. Across the four hyperscalers, the number runs $650 billion to $700 billion, roughly the peak of the late-1990s telecom investment cycle adjusted for inflation.

The question isn't whether the money is flowing. It's whether the companies collecting it still have the factor stack to justify their multiples. Growth without valuation discipline is just a faster path to disappointment. Let's run the comparison.

NVIDIA's Q1 fiscal 2027 revenue hit $82 billion, up 85% year over year, with data center revenue of $75 billion climbing 92%. That kind of growth is still exceptional - earnings are expected up 54.5% in fiscal 2026 and another 52.4% in fiscal 2027. The stock trades at roughly 41x trailing earnings.

Here's the detail that matters: the stock fell 5% after Q4 fiscal 2026 results, despite beating expectations in 18 of the last 20 quarters. The market is no longer rewarding beat-and-raise. At 41x, NVIDIA is being graded on the slope, not the level - and the slope, while still positive, is flattening from its 2025 trajectory. In our book, when the biggest winner in a cycle starts trading flat on beats, you widen your position criteria, not your conviction.

Portfolio role: Core growth sleeve anchor. You hold NVIDIA as long as data center revenue growth stays above 40% and the P/E doesn't stretch past 50x on forward earnings. The trigger to reduce is a quarter where data center growth drops below 30% with no product cycle offset.

AMD (AMD)

AMD's data center segment revenue was $5.8 billion in Q1 2026, up 57% year over year. The stock is up roughly 144% year-to-date in 2026. Full-year 2025 revenue rose 34%, with free cash flow reaching approximately $6.7 billion. Forward valuation sits near 30x earnings.

The contrast with NVIDIA tells the story. AMD's growth rate is strong but from a fraction of NVIDIA's base - $5.8 billion in data center revenue versus NVIDIA's $75 billion. That gap is real. The 30x forward multiple is cheaper than NVIDIA's 41x trailing, but you're buying optionality on NVIDIA's mistakes, not a standalone thesis. AMD's position as a pure secondary supplier means the factor stack is momentum-driven: the stock runs when the AI trade runs and lags when it pauses.

Portfolio role: Growth satellite - a smaller allocation that captures upside if AMD gains data center share. The trigger to add is consecutive quarters of data center growth above 60% with visible design wins at multiple hyperscalers. The trigger to reduce is a contraction in data center growth below 30%, which would signal the tailwind is weakening before the stock adjusts.

Broadcom (AVGO)

Broadcom is where the factor stack looks most asymmetric. The trailing P/E sits around 72x, which looks rich on the surface. But the forward story flips: the stock trades at roughly 29x fiscal 2026 earnings and closer to 19x fiscal 2027. That gives Broadcom a forward PEG (price-to-earnings-growth) ratio of about 0.52 - meaning you're paying half a dollar for each dollar of expected earnings growth. Multi-year AI chip supply deals with Google's TPU program lock in revenue visibility through 2028.

The 115% surge from its 52-week low matters less than the PEG compression. What Broadcom has that the narrative often misses is a dual-engine model: custom AI silicon that hyperscalers can't build in-house at scale, plus VMware software cash flows that provide a floor. LTM revenue growth of 25.2% on a forward multiple that prices in 40%+ earnings growth is the kind of setup that defines GARP - growth at a reasonable price.

Portfolio role: Core GARP holding. The factor stack - forward valuation discount to growth rate, revenue visibility from multi-year contracts, diversified cash flow - makes this the strongest quantitative case in the set. Hold through volatility as long as the forward PEG stays below 1.0 and the custom silicon pipeline continues expanding.

Tier 2: Power and Cooling

Vertiv (VRT)

Vertiv delivers the power infrastructure and cooling systems that keep AI clusters from melting. Q1 2026 revenue hit $2.65 billion, up 30.1% year over year, with Americas growth of 53%. Adjusted earnings grew 83% year over year. The company's project backlog more than doubled. The stock trades at a forward P/E near 53x and is up roughly 42% in 2026.

The mechanism is straightforward: AI-optimized servers are the most power-intensive workload in computing history. AI servers will account for 44% of total data center power by 2030, up from 21% today. 190 gigawatts of hyperscale data center capacity has been announced as of early 2026. Vertiv sits between capex announcement and revenue recognition - when a hyperscaler signs a data center deal, Vertiv's power and cooling backlog moves first.

The 53x forward P/E is expensive by historical standards for an industrial company. The justification requires that backlog conversion stays above 25% and that AI power demand doesn't plateau mid-cycle. That's the risk: 190 GW announced is not 190 GW built. If the cycle stalls, a 53x multiple has a long way to fall.

Portfolio role: Cyclical growth in the infrastructure sleeve. This is a momentum trade on the durability of capex spend. Hold while quarterly revenue growth stays above 25% and backlog expansion continues. Reduce if any quarter shows sequential revenue deceleration - the first sign that announced capacity isn't converting to construction.

Tier 3: Data Center Real Estate

Equinix (EQIX)

Equinix is the landlord layer - it owns the physical data centers where hyperscalers place their AI workloads. Q1 earnings delivered EPS of $10.79 versus a $4.30 estimate, a massive beat that sent the stock higher and prompted raised full-year guidance. The company plans to roughly double its data center capacity by 2029. The stock trades at a trailing P/E near 74x, though the forward P/E compresses into the 56x range on expected earnings growth.

The Q1 earnings beat of this magnitude - more than 2.5x consensus - signals either that estimates were stale or that AI-driven demand for premium data center space is accelerating faster than analysts modeled. The raised guidance supports the latter. Equinix's $5.16 quarterly dividend (payable mid-June) adds an income component that neither NVIDIA nor Vertiv offers.

But 56x forward earnings for a real estate investment trust is not cheap. The valuation assumes sustained double-digit same-store revenue growth from AI tenants and no significant vacancy risk as capacity comes online. If hyperscalers delay or self-build more aggressively, the landlord multiple compresses.

Portfolio role: Dividend growth plus AI exposure - the barbell piece. You use Equinix to get AI infra upside with a dividend floor. The combination of yield and growth exposure is useful in a rising-uncertainty environment where you want participation without naked volatility. Watch for same-store revenue growth dropping below 8%, which would signal the AI rental premium is fading.

What the stack says to do

Five stocks, three tiers, and a $700 billion spending backdrop that's real but not equally distributed. Broadcom has the strongest factor stack - forward valuation priced below its growth trajectory, contract visibility, and diversified cash flow. NVIDIA's growth is still exceptional but the market is no longer rewarding it the way it did at 100% growth; at 41x with a flattening slope, it's a hold with defined triggers rather than a momentum buy. AMD is the optionality play - cheap relative to NVIDIA but dependent on NVIDIA's stumbles. Vertiv is the infrastructure call, high-growth but high-multiple, best held while the conversion story holds. Equinix is the dividend-growth barbell piece for when you want AI exposure with income.

The risk to all five is the same: the $700 billion spending number stops growing faster than it did in the prior year. Capex cycles don't decelerate smoothly. When they break, the multiple-compression hits the highest-P/E names first. That's why triggers matter more than conviction. Set them. Follow them. The process works when you let it.

What would change my view: a quarter from any of the four hyperscalers where AI-related capex guidance drops below 30% year-over-year growth. That's the single data point that shifts from build cycle to pause cycle, and it would compress every multiple on this page regardless of current earnings quality.