3 Under-the-Radar Ways to Play Goldman's $1 Trillion AI Spending 2027 Forecast

Goldman Sachs sees AI infrastructure spending rising to $920 billion-$1.4 trillion next year from more than $700 billion this year, supporting a bullish case for Alphabet, TSMC, and ASML. Alphabet is highlighted as both a major AI capex spender and beneficiary via lower-cost TPUs and direct orders with Broadcom, while TSMC benefits from widening chip demand and potential 15% price increases on 3nm chips. ASML stands to gain from robust demand for its EUV machines, which are essential for advanced chip production and HBM.

Analysis

The cleanest second-order read is that AI capex is no longer a single-narrative trade in NVDA alone; it is becoming a capacity-scarcity trade across the stack. Alphabet’s willingness to self-fund more infrastructure matters because its TPU economics lower the hurdle rate for incremental AI deployment, which should accelerate internal workload migration and preserve margin even as reported capex rises. That creates a subtle competitive pressure on hyperscalers still dependent on merchant GPU supply: they must spend more dollars per unit of inference capacity to keep pace, which can compress returns on capital before revenue fully scales.

TSMC is the best expression of the widening of AI demand, but the market may be underestimating how much pricing power comes from roadmap control, not just wafer volume. As more of the spend shifts into custom ASICs, advanced CPUs, and memory-adjacent components, TSMC’s role becomes less cyclical and more like a bottleneck toll road. The next leg higher in earnings likely comes from mix, yields, and surcharge discipline rather than unit growth alone, which is why the upside can persist even if headline AI spend growth slows modestly.

ASML is the most levered long-duration beneficiary, but also the most timing-sensitive. The market typically overweights order headlines and underweights the install base constraint: once EUV/DUV capacity is allocated, delivery and ramp issues can create a 12-24 month revenue recognition lag. That means the stock can rerate well before the full capex cycle is visible in financials; conversely, any export-control tightening or memory capex pause would hit sentiment first, then orders with a delay.

The contrarian miss is that this is not just a bullish semiconductor call; it is also a differentiation call against less efficient AI spenders. If model economics keep improving for vertically integrated firms like Alphabet, the winners may be the companies that can internalize AI cost curves, not merely those selling picks and shovels. The biggest risk to the trade is a capex deferral cycle in 2H-2026 if AI monetization disappoints, which would pressure the more hardware-exposed names before software-anchored platforms.

Key Decisions for Investors

• Go long TSM over NVDA for a 6-12 month horizon: TSM captures demand from every chip architecture, while NVDA remains more exposed to pricing normalization and customer self-design; use a 2:1 upside/downside profile with tighter stop if AI capex data rolls over.
• Maintain/add to GOOG on pullbacks for a 12-month view: the market underappreciates that internal TPU economics can compound margin advantage while external TPU commercialization becomes an incremental high-margin option; risk/reward improves if capex-to-revenue fears create a 5-8% drawdown.
• Long ASML on any post-order-cycle weakness, 9-18 months: the stock should benefit from order-book visibility and scarcity pricing, but position size should reflect policy/export-control risk; target a 15-20% upside versus low-teens downside on a 12-month basis.
• Pair long TSM / short INTC for a 6-12 month structural trade: AI capex concentration should keep rewarding the advanced-node bottleneck while legacy foundry turnaround remains execution-heavy; this is a cleaner way to express semiconductor capex dispersion.
• If wanting a hedge against an AI capex air pocket, buy short-dated downside protection on AVGO or NVDA into any parabolic move; they are the most sensitive to near-term demand pauses if hyperscaler spending is deferred into 2027.