Barclays' model shows AI spending cycle is far from peak. That means Nvidia shares are too cheap

Barclays says consensus hyperscaler capex is at least $225 billion too low for 2027-28, implying a longer and larger AI infrastructure up-cycle. The bank argues Nvidia is underpriced given this view, noting NVDA trades at ~17.5x CY27E P/E (and ~14.5x on CY28E under a conservative +44%/+11% EPS growth assumption for CY27/CY28). Barclays highlights next-gen Nvidia chips and higher ASPs as upside for AI semis and calls the stock's flat YTD performance a buying opportunity. For context, the iShares Semiconductor ETF is up 13.6% YTD and Micron has rallied ~46%.

Analysis

If hyperscaler compute budgets rebase materially above consensus, the immediate winner is not just Nvidia’s wafer shipments but the entire high-margin attach of HBM memory, advanced packaging and software monetization that sits on top of each GPU sale. That implies incremental EPS leverage for Nvidia that scales faster than linear GPU unit growth — each additional $1 of hyperscaler spend likely translates to >$1.50 in vendor revenue when you include higher ASPs, services, and ecosystem fees, magnifying upside to 2027–28 earnings estimates.

Second-order winners include DRAM/HBM suppliers and foundry/equipment players because lead times for capacity (12–24 months) and willingness to pay premium ASPs compress the usual price declines; conversely, incumbents reliant on legacy on-prem hardware and un-integrated accelerators face margin pressure. For hyperscalers themselves, elevated capex intensifies a short-term cash conversion impact but creates optionality: owning proprietary models and silicon pathways can drive outsized SaaS pricing power for 3–5 years if they win a persistent performance lead.

Key risks that can unwind the bullish view are algorithmic efficiency (models that cut training compute needs), a credible open-source stack that reduces vendor lock-in, or a supply-side shock that brings cheaper non-Nvidia alternatives with competitive software ecosystems. Watchable catalysts in the coming 3–12 months: quarterlies with capex commentary, tapeouts/ramp proof-points for successor silicon families, and public hyperscaler disclosures on model refresh cadence.

Given the convexity in outcomes, the market is likely pricing a ‘peak-and-decline’ capex scenario; if hyperscalers instead extend spend at higher ASPs, valuation rerating should be rapid. We should position to capture asymmetric upside while capping downside from a potential short-term rotation away from megacaps.

Key Decisions for Investors

• Long NVDA equity (size 2–4% AUM) with a 12–24 month horizon; hard stop at -18% and target +60–120% if cloud capex and ASPs re-accelerate. Rationale: direct exposure to ASP and software monetization upside with controlled drawdown.
• Long NVDA Jan-2027 LEAP calls (funded by selling 3–6 month calls) — buy-dated calls to capture multi-cycle capex uplift, sell nearer-term premium to reduce cost basis. Risk/reward: limited premium at risk vs multi-fold upside on a sustained capex surprise.
• Pair trade: long NVDA / short AMD (dollar-neutral, size 1–2% AUM) over 6–18 months to express capture of high-end training GPU ASP upside versus AMD’s inference/commodity exposure. Cut if AMD wins multiple large hyperscaler design-ins (reassess at quarterly cadence).
• Selective long MSFT or GOOGL 12–18 month call spreads (small allocation 1% AUM) to play hyperscalers’ monetization of models rather than raw capex — defined downside, levered upside if cloud ARPU per customer rises materially.