Bank of America resets Nvidia stock price target for 2026

Bank of America raised Nvidia’s price target to $320 from $300, implying about 45% upside, while keeping a Buy rating. The bank lifted its 2030 AI data center TAM estimate to $1.7 trillion from $1.4 trillion and increased FY2028/FY2029 sales and EPS forecasts by roughly 7% each, reflecting stronger AI demand expectations. The note is supported by Nvidia’s Blackwell volume production, expected Vera Rubin launch in 2H26, and near-term catalysts including Q1 FY2027 earnings on May 20.

Analysis

The key takeaway is not the higher target; it is the re-anchoring of the whole AI demand stack to a much larger capex supercycle. If the market starts underwriting a $1.7T AI data-center buildout by 2030, the beneficiaries are not just the accelerator vendor but also the picks-and-shovels layer: memory, networking, optical interconnect, and power/thermal management. That means the most convex exposure may sit in suppliers with pricing power and underappreciated operating leverage, while the prime vendor faces more margin compression than the headline TAM implies.

The second-order effect is that bigger TAM does not necessarily translate into linear share gains for the incumbent. Hyperscalers are explicitly diversifying into custom silicon to manage cost and bargaining power, which can cap the long-run mix of external accelerator spend even as total AI capex rises. In other words, the pie gets bigger, but a larger slice may accrue to AVGO-style ASIC enablers, MRVL networking, and memory vendors rather than flowing entirely to NVDA.

Near term, the setup is more about catalyst risk than valuation. Earnings over the next 1-2 quarters can confirm whether demand is accelerating fast enough to offset rising HBM costs and whether gross margin erosion remains orderly; any guide-down on mix or China exposure would hit the multiple quickly. Conversely, a favorable export policy outcome and a new product/CPU announcement could force another leg higher, but that upside is probably already partially embedded after the move to a $5.5T market cap.

The contrarian angle is that the market may be extrapolating uninterrupted infrastructure spending while underestimating the pacing risk from customer ROI scrutiny. If token economics improve too quickly, hyperscalers may shift from brute-force GPU buying to efficiency-driven architecture changes, which could reduce unit demand growth even as inference expands. That argues for owning the ecosystem into the buildout, but being more selective on beta-sensitive single-name NVDA exposure at this valuation.

Key Decisions for Investors

• Long AVGO / long MRVL as a 3-6 month relative-value expression of continued AI buildout; these names should capture incremental ASIC, networking, and custom-infrastructure spend even if NVDA share is capped. Favor on any post-earnings NVDA strength as a hedge against direct multiple compression.
• Buy NVDA downside protection into earnings: use 2-4 week put spreads around the May 20 print to hedge event risk while preserving upside. Risk/reward improves if implied volatility remains below the realized-move range typical for mega-cap AI prints.
• Pair trade: long SMH basket / short a smaller-cap AI hardware cohort if you want cleaner exposure to the infrastructure cycle without single-name concentration risk. Best held over 1-2 quarters as the market distinguishes durable winners from narrative-driven names.
• If China export headlines turn positive, trade NVDA tactically with calls rather than stock; the policy catalyst is binary and may fade fast after the summit. Treat as a 1-2 week momentum trade, not a core add, given already-extended valuation.
• Use any broad AI pullback to accumulate AVGO or MRVL on 5-8% drawdowns, with a 6-12 month horizon and tighter risk than NVDA. The thesis is that custom silicon and networking should compound even if the accelerator share shifts over time.