JPMorgan Priced Bitcoin at 10% and Left It Out of Every Portfolio — In a Book Built on $900 Billion of AI Capex

Bitcoin's 10.0% base case sits at the top of the 2027 return ladder, above private equity at 9.8% and well above the 7.5% assumed for US large-cap equities or the 5.5% for gold. The bank published the inputs alongside the number: 55% annualized volatility against roughly 18% for US large caps, and a 10% compound return that is 20.9% in arithmetic terms — a gap the firm itself describes as volatility drag. That combination makes the 10% a discount rate for judging bitcoin rather than an instruction to own it.

Read the ladder as a hurdle rate rather than a forecast. The two asset classes that clear the 60/40's 7.0% — private equity at 9.8% and bitcoin at 10.0% — sit outside the core of a balanced portfolio, but only private equity can be dropped into a diversified mandate without breaking the optimizer, because bitcoin's correlations are too unstable to model. That leaves bitcoin the only asset class in the book with a return forecast and no portfolio role.

On JPMorgan's own numbers, the 10% is not the best risk-adjusted return in the document. Bitcoin returns 10% for 55% of volatility, roughly 0.18 per point of risk; US large-cap equities return 7.5% for about 18%, roughly 0.42. The bank treats bitcoin as a volatility-adjusted match for gold rather than a building block, and the arithmetic backs it up — leaving the headline 10% as the worst risk-adjusted return of the three.

JPMorgan's forecast is a forecast of construction, not of software demand. Global AI capital expenditure runs at $900 billion in 2026 alone and reaches $5.5 trillion by 2030, while US AI capex rises from over $800 billion this year to nearly $1.2 trillion by 2030 before settling at roughly 70% of that peak. The five largest US hyperscalers — Amazon, Alphabet, Meta, Oracle and Microsoft — are on track to spend as much in 2026 as they have cumulatively since ChatGPT launched in November 2022. That makes these assumptions an underwriting case for data centers rather than a call on software.

The most useful line in the document is not $5.5 trillion. It is that equipment — GPUs, servers, memory and networking — falls from just over 40% of AI capex in 2026 to 30% by 2040. The dollar keeps compounding; the chipmakers' slice of it does not. That shifts the durable part of the AI build away from silicon and toward power, land and balance sheets.

The useful thing about a hurdle rate is that it can be tested against a price. JPMorgan has now published the first one a frontier AI lab can be measured against, and it is 10%: the same number it attached to bitcoin, a fifth of a point above the 9.8% it assumes for private equity and three points above the 7.0% it assumes for a balanced stock-bond portfolio. Compounded over ten years, that is 2.59 times, against 1.97 times for the 60/40. That turns a $1.25 trillion private AI mark into a claim on $3.2 trillion a decade out.

The crypto side of the book is shrinking rather than compounding. Coinbase reported $1.216 billion of revenue in the September quarter, 34.9% below the $1.869 billion of a year earlier, and an operating loss of $26 million; $850 million of non-operating gains turned that into $824 million of pre-tax income and $718.7 million of net income. Trailing twelve-month revenue of $5.63 billion now carries a $702 million trailing loss. The print shows fee income evaporating as crypto volumes shrink.

Strategy is the pure leverage expression of the same call. Of the $77.15 billion on its September 30 balance sheet, $70.95 billion sits in mark-to-market non-current assets, with $13.06 billion of preferred stock ahead of common shareholders and a trailing-twelve-month loss of $14.93 billion on $495 million of revenue. A base case that carries bitcoin from $83,000 to roughly $215,000 by 2036 is the bull case for that structure, leaving the preferred dividend waiting on a bitcoin recovery rather than on operating cash flow.