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 constructive outlook for Alphabet, TSMC, and ASML. Alphabet is highlighted as both a major AI capex spender and a beneficiary through its TPU cost advantage and Google Cloud growth, while TSMC benefits from broader demand for GPUs, ASICs, and CPUs plus pricing power on 3nm chips. ASML stands to gain from demand for its EUV and DUV lithography machines, which are essential to advanced AI chip production.

Analysis

The market is still underestimating how AI capex redistributes profit pools away from model developers and toward the picks-and-shovels layer. The key second-order effect is that a broader chip mix is now being deployed in data centers, which reduces the risk that any one GPU vendor captures all incremental spend and instead increases the value of the constrained manufacturing stack. That is structurally bullish for the foundry/lithography duopoly, but the marginal benefit is likely larger for the equipment bottleneck than for the chip designers if capex stays elevated for multiple years.

Alphabet’s edge is not just lower training cost; it is capital efficiency that compounds into higher cloud margins and more aggressive product iteration. If TPU adoption widens, Broadcom gains a design-win annuity, but the bigger implication is that custom silicon can compress the economic moat of off-the-shelf accelerator vendors in inference, where unit economics matter more than peak performance. That creates a subtle headwind for the most exposed GPU names over a 12-24 month horizon if customers continue to internalize more workloads.

TSMC’s risk/reward is attractive because capacity scarcity persists even if AI demand normalizes from a surge to a plateau. The bigger risk is not demand loss but eventual supply diversion: if non-TSMC fabs close the process gap faster than expected, pricing power fades sooner than consensus assumes. ASML remains the cleanest expression of the cycle because every new node transition and memory upgrade path still requires tool spend before revenue can scale, but the stock is also most sensitive to any pause in fab ordering, which would show up first in bookings before earnings.

The contrarian miss is that the trade is crowded at the theme level but not at the bottleneck level. Investors are paying up for AI beneficiaries with obvious revenue leverage, while the true scarcity rents may accrue to the least visible names with the longest lead times. That favors a barbell of high-quality infrastructure winners, while fading any rally in companies whose AI story depends on pricing power rather than hard capacity constraints.

Key Decisions for Investors

• Long ASML on a 6-12 month horizon; best risk/reward versus the AI stack because tool demand is the earliest and most durable bottleneck. Add on any post-earnings weakness tied to order timing rather than demand destruction.
• Overweight TSM vs NVDA over the next 12 months; AI spend broadening should support foundry utilization even if GPU pricing moderates. Use NVDA strength to rebalance toward TSM if delivery times remain extended.
• Long GOOGL, but structure it as a quality compounder trade rather than a pure AI beta expression. The convexity comes from capex efficiency and inference cost leverage, so the trade works best if cloud margins expand over the next 2-4 quarters.
• Consider a pair trade: long TSM / short a basket of hyperscaler-adjacent AI hardware names with the most exposed valuation multiples. If capex stays high but hardware ASPs normalize, the bottleneck beneficiaries should outperform over 6-9 months.
• Avoid chasing AVGO purely as an AI infrastructure proxy; the article implies a customer-specific design-win benefit rather than broad operating leverage. Use only as a tactical trade on TPU adoption headlines, with tighter stops.