Bank of America Analyst: AI Is Becoming “Deeply Embedded in Enterprise Workflows” as AI Spending Could Reach $1.5 Trillion

Bank of America’s Vivek Arya says AI infrastructure demand is still early, pointing to the Philadelphia Semiconductor Index up nearly 80% YTD through early July 2026 and hyperscaler cloud capex rising from a $600–700B expectation to nearly ~$1T in 2026. He projects cloud capex could grow another 40–50% next year to ~$1.4–$1.5T, supporting continued multi-vendor AI chip outperformance. He highlights NVIDIA’s Q1 FY27 revenue of $81.615B (+85% YoY) and Q2 FY27 revenue guidance of $91.0B, alongside Broadcom (+143% AI revenue), TSMC (+30.1% May 2026 revenue), AMD (data center rising to 57% of revenue) and Micron’s Q3 FY26 revenue of $41.456B (+345.7%).

Analysis

The cleanest read-through is not “AI is strong,” but that the bottleneck has shifted from model demand to physical capacity, which extends pricing power for the infrastructure layer. That favors NVDA and AVGO first, then TSM as the toll collector on leading-edge wafers and packaging, with MU as the highest-beta beneficiary if memory remains undersupplied. The second-order effect is that every incremental dollar of hyperscaler capex becomes more committed and less reversible, so supplier order books should stay tight even if end-customer monetization is still evolving.

The underappreciated loser is the AI spenders themselves: GOOGL and META can still win strategically, but rising capex is a near-term drag on free cash flow and can cap multiple expansion if revenue conversion lags. In the next 1-3 months, the market will care less about AI narrative and more about whether monthly revenue and capex guidance keep ratcheting higher; any pause in hyperscaler budgets would hit the most levered names first, especially AMD and MU, where sentiment is already priced for continued acceleration.

Contrarian risk: consensus may be underestimating how quickly this becomes a crowded duration trade. If lead times normalize or export constraints tighten, the market can reprice the whole basket even while demand stays healthy, because the marginal upside is already embedded in expectations. The thesis is falsified if the next round of hyperscaler capex updates stops surprising upward or if TSM/NVDA guide below the current implied run-rate; that would argue for taking profits on the highest-multiple names rather than averaging down.

Key Decisions for Investors

• Go long NVDA / AVGO / TSM on a pullback over the next 1-3 weeks; best risk/reward is in the suppliers with recurring order visibility rather than the spenders. Use 3-6 month horizon, with thesis failure if NVDA or TSM guide below implied run-rate.
• Pair trade: long NVDA or AVGO vs short GOOGL or META for 1-3 months to isolate capex beneficiaries from capex funders. This expresses margin transfer from hyperscalers to chip vendors; exit if hyperscaler capex growth decelerates less than feared or ad/cloud revenue inflects higher.
• Long MU only on weakness, preferably via call spreads 3-6 months out. MU has the most torque if memory tightness persists, but it is the first name to get hit if inventory builds or if capex pauses; keep risk sized smaller than NVDA/AVGO.
• If already long the AI basket, trim into strength rather than chase: the trade is crowded and any guidance miss can trigger a sharp de-rating over days, even if the 6-18 month structure remains intact.
• Watch for a reversal signal in monthly TSM revenue and hyperscaler capex commentary; if either stops accelerating for one reporting cycle, shift from net long semis to a market-neutral posture.