Ark Invest Says AI Spending Could Triple: Here's the Stock to Buy for 2026

Ark Invest's Cathie Wood projects data-center capex rising from roughly $500 billion in 2025 to $1.4 trillion in 2030, underpinning strong demand for custom AI ASICs. Broadcom, a leader in ASIC IP and with close foundry ties to TSMC, has supported Google TPUs, holds a $21 billion Anthropic TPU order for delivery this year, and estimates a $60–90 billion opportunity from its three furthest-along customers in fiscal 2027; the company reported just under $64 billion in fiscal 2025 revenue. Citigroup projects Broadcom AI revenue could grow from about $20 billion last fiscal year to $100 billion in fiscal 2027, and Apple and OpenAI are also cited as customers, positioning Broadcom as a major beneficiary of rising AI infrastructure spending.

Analysis

Market structure: The shift to bespoke AI ASICs materially reallocates TAM from general-purpose GPUs to ASIC/IP/foundry stacks. Using Ark’s scenario ($500B 2025 → $1.4T 2030) implies incremental ~$900B annual data‑center capex by 2030; Broadcom (AVGO) stands to capture high‑margin design/IP and integration revenue while TSMC (TSM) captures wafer share. Nvidia (NVDA) faces pricing pressure on inference workloads and potential share loss in customers seeking vendor diversification.

Risk assessment: Key tail risks are regulatory export curbs on advanced nodes, Broadcom execution/customer concentration (three customers = $60–90B opportunity), and software/stack lock‑in (CUDA dominance). Timeline: sentiment moves immediately (days), order book recognition in quarters (2–8 quarters), structural revenue shifts by FY‑2027. Hidden dependency: foundry capacity and process node availability (TSMC) are binding constraints; a TSMC shortage or yield issues would throttle ramp.

Trade implications: Direct trades favor AVGO and TSM exposure; use LEAPs for convexity. Implement relative trades that monetize de‑risking away from NVDA dominance (long AVGO, hedged short NVDA) and overweight cloud infra owners (GOOGL) that internalize ASICs. Monitor quarterly capex guidance and Anthropic/GCP TPU shipment cadence as 30–90 day catalysts.

Contrarian angles: Consensus underestimates integration, software migration and customer pullback risks — hyperscalers may internalize design but externalize manufacturing bargaining power, compressing vendor margins. Historical parallel: server ASIC cycles where vendor incumbents lost pricing power (e.g., custom ASICs vs CPUs); if NVDA aggressively cuts inference pricing or bundles software, AVGO upside could be muted.

Key Decisions for Investors

• Establish a 2–3% long position in Broadcom (AVGO) sized to portfolio risk now; add incremental 1% on pullbacks of 5–10%. Target +30–40% total return by end of CY2027; place tactical stop‑loss at 12% from entry to limit execution risk tied to customer concentration.
• Purchase AVGO 12–18 month LEAP calls (size 0.5–1% of portfolio premium; target delta ~0.5–0.7) to capture upside from multi‑year ASIC ramps; roll or take profits after confirmed Anthropic/TPU revenue recognition or a 25% rally.
• Implement a pair trade: long AVGO (1.5%) and short NVDA (0.75%) via notional balancing or buy NVDA Dec 2026 puts sized to 50% of the AVGO long. Rationale: capture asymmetric upside from ASIC adoption while hedging GPU deflation risk; trim the short if NVDA reports >10% QoQ revenue growth tied to new product demand.
• Overweight TSM (1–2% overweight) and GOOGL (1% overweight) for foundry capacity and cloud ASIC monetization; underweight pure GPU plays by reducing NVDA exposure by 20–30% if current allocation is overweight. Reassess after next 2 quarterly capex disclosures—if Broadcom/TMSC guidance misses, cut exposure by at least half.
• Set monitoring triggers: if Broadcom announces inability to meet Anthropic/TPU deliveries or TSMC signals node capacity shortfalls within 60 days, close options positions and reduce AVGO net exposure to <1% immediately to avoid execution and supply‑chain tail risk.