June 2026
The Number
I have funded the account with $1,500 in personal capital, earned through my military service in Korea. I’m starting this project with a low-stakes amount but enough to make my skin in the game and take the portfolio seriously.
The Starting Position
I plan to wait and evaluate the markets for a while before deploying capital too hastily. With my other personal brokerage accounts, I’ve noticed that I tend to get too trigger-happy and end up making investment decisions without enough research. I hope to use Sonsu Research as a platform to make well-informed allocations to make returns over the medium-to-long term time horizon.
The Current State
The macro environment, with the war in Iran likely becoming a persistent conflict, appears to be volatile. Some voices in the investment community see that the AI market is a bubble, citing similarities in exponential growth in share prices and capital commitments that outrun any credible near-term revenue. J.P. Morgan estimates the industry needs roughly $650 billion in new annual revenue to earn a 10% return on the infrastructure currently being built, against AI-attributable revenue somewhere between $50 and $150 billion depending on how much incremental cloud growth one is willing to credit to AI.
The structural concern is not valuation, which is the argument cited most often. High multiples are a normal feature of every technology cycle. The financing is the part that requires further scrutiny. By 2026 estimates, more than $800 billion in deals sit inside arrangements where a chipmaker invests in an AI laboratory, the laboratory spends that capital on the chipmaker’s hardware and on cloud capacity from its partners, and the cloud provider borrows to buy more hardware. Nvidia, OpenAI, Oracle, AMD, and Microsoft appear on multiple sides of the same transactions. GMO has compared the structure to the circular financing of the internet bubble, and the telecom vendor financing of the late 1990s, which collapsed when usage failed to match the revenue booked against it. The difficulty is that revenue generated inside the loop looks identical, on an income statement, to revenue generated by a customer spending money it earned elsewhere.
On the other side of the argument, Alphabet, Microsoft, and Amazon carry real earnings, real free cash flow, and cloud segments compounding at rates the 2000 cohort never approached. The buyers of this infrastructure are solvent in a way that the dot-com buyers were not, and the assets being built have measurable utilization. Both of these things can hold at once.
Watchlist
MU. A pure way to own the memory cycle from a US-listed account.
EWY. The only available Samsung and SK Hynix exposure, diversified in name and roughly half memory in fact. Buying it is a bet on two companies with a currency attached.
GOOGL and TSM. Hyperscaler capex from the spending side and the fabrication side.
SGOV. Pays roughly 3.5 to 3.75% to wait, which makes doing nothing a decision with a return rather than a default.
The account is live. The process begins.