Better Buy: Bitcoin or AI?

Nvidia is up ~1,266% over five years versus Bitcoin's ~28%, while Bitcoin miners pivoting to AI (TeraWulf +390% and Cipher Digital +365% over 12 months) have materially outperformed BTC. Bitcoin trades around $70,000 versus an estimated mining cost of ~$87,000, pressuring mining economics and prompting pivots into AI compute. The article notes a potential longer-term convergence if AI agents adopt Bitcoin for micro-payments (per Cathie Wood) but flags Goldman Sachs' concern over an AI revenue–capex gap and the possibility of an AI bubble. Overall, near-term sector flows favor AI/miners, but the author remains bullish on Bitcoin long-term.

Analysis

A material, underappreciated second‑order is that crypto miners already own three scarce inputs for AI compute: long‑term power contracts, purpose‑built DC shells, and operational expertise running 24/7 high‑density racks. If even a subset of these operators repurposes capacity to AI training or inference marketplaces, they can undercut hyperscale unit economics by ~20–40% (lower colocation/O&M per GPU hour) and force a faster commoditization of GPU pricing than current sell‑side models assume.

That competitive shift has asymmetric implications up and down the stack. Near term it props revenue for GPU suppliers and services (benefitting NVDA sentiment), but over 12–36 months it increases the bargaining power of AI compute resellers and raises the probability of margin pressure for companies that rely on sustained GPU ASPs. Simultaneously, the pivot accelerates demand for alternative silicon (Intel, Habana, custom ASICs) and for used/server‑refurb ecosystems — a structural headwind for ASIC‑only miners and a tailwind for chip vendors able to supply whole data‑center solutions.

Key risks and catalysts to watch are non‑linear: (1) GPU supply normalization or new fab capacity that eases ASPs within 6–18 months; (2) a pullback in enterprise model training budgets if ROI on generative AI pilots disappoints (a 20–30% cut in model spend would flip profitability for marginal custom compute players); and (3) regulatory/frictional barriers to autonomous AI agent payments — that use‑case requires custodial, throughput and privacy plumbing that is unlikely to produce material on‑chain revenues in under 2–3 years. These create clear windows where market leadership can reverse rapidly, so execution and timing matter more than narrative alignment.

Key Decisions for Investors

• Buy NVDA 6–9 month call spread (bull call) sized to 1–1.5% NAV: entry on a <5% pullback in NVDA. Rationale: capture continued AI services momentum while limiting capital at risk; target asymmetric payout (2–3x) if model/service spend sustains, max loss = premium.
• Long WULF equity 6–12 months (small starter position 0.5–1% NAV) paired with protective 6–9 month put: thesis is re‑rating if the AI pivot monetizes excess power/data center assets, but execution/regulatory risk is high — hedge to cap downside while keeping upside optionality.
• Long INTC 12–24 month LEAP calls (or stock) as a convex, defensive play on alternative silicon and data‑center integration: if GPU ASPs compress, Intel’s scale and silicon roadmap can recapture share — reward multi‑turns on a multi‑quarter cadence, risk is slow product ramp.
• Event hedge: if you hold any positions long AI hardware, buy short‑dated tail protection (quarterly OTM puts) across the basket before major earnings windows (NVDA/INTC). Rationale: earnings or macro headlines can compress frothy multiples quickly; protection cost is justified given asymmetric downside.