Josh Jiwoon Inn
Founder & Editor, Compute Current
Josh Jiwoon Inn is the founder and editor of Compute Current. He works on the business side of data center infrastructure, on site screening, power and capacity planning, and colocation and hyperscaler deal work, and he writes The Current from that seat: where the real constraints in AI infrastructure sit, who pays for them, and what to watch next. He writes under his own name. Essays are drafted with AI-assisted tools from source-linked evidence, edited and published under his direction, and every number is attributed to a named source.
How this column works
The Current is written by Josh Jiwoon Inn, Compute Current's founder and editor, with AI assistance in drafting. He does not claim interviews, site visits, or first-hand reporting in these essays; every number is attributed to a named source publication.
Essays are drafted with AI assistance from source-linked evidence, must pass the site’s quality gates before publication, and are corrected after the fact by the author when a reader or a source shows an error. Full detail lives in the AI-assisted disclosure, the editorial policy, and the terms of use.
Standing positions
- Test whether grid access is the binding constraint for this specific project; do not presume it dominates every chip, software or financing story.
- Distinguish announced, contracted, commissioned and operating capacity using the milestones actually evidenced.
- Test capex downside through disclosed utilization, tenant credit, contract terms and residual-value assumptions; overshoot is a scenario, not an established fact.
- Compare training and inference economics using the actual workload, customer retention, utilization and hardware life. Neither label guarantees durable margins.
- Identify the measured constraint for this story. Do not assume a universal sequence from chips to power to cooling to capital.
- Distinguish a proposed integration strategy from funded assets and operating capability; bargaining effects need evidence.
- Test heat rejection, liquid fraction, redundancy and site conditions before inferring usable capacity from a component rating.
- Prefer relevant primary filings and technical evidence to unsupported promotional claims; do not repeat this preference as a slogan.
- Assess whether policy, customer demand or infrastructure readiness drives this project; do not assume subsidy dependence.
- Separate funding availability from execution capability and contract economics; neither cheap capital nor good execution proves a return.
Recent columns
All columns →Europe's AI dividend arrives as equipment exports while its own build costs rise
The Commission credits AI investment abroad for lifting EU exports while cutting investment growth and flagging dearer energy and credit. Europe is selling to the AI buildout more…

I read OpenAI's monthly-doubling coding-agent curve as demand, not as a cost line
Epoch AI finds OpenAI researchers doubling coding-agent use roughly every month, and the list-price dollars attached to that curve measure usage, not what OpenAI pays.

Up to 3× pod density is a benchmark, not a RAM budget cut
The Kubernetes Blog reports density gains of up to 3× from Local SSD swap across three workloads. I read that as a pilot case for idle-heavy agent sandboxes, not a cue to shrink…

Malaysia’s server-trade gap turns permit enforcement into a capacity-risk signal
Epoch AI found a $3.8 billion versus $0.6 billion mirror-trade gap that narrowed after Malaysia tightened transit rules. The pattern is consistent with diversion, but it does not…

EU data center ratings could make heat reuse a commercial test
The Commission's proposed ratings could change procurement and operating choices. Treating them as a new heat-reuse permit condition goes beyond the source evidence.

ERCOT's six-week demand record turns Texas interconnection into a capacity gate
Six weeks of ERCOT load above the 2023 record, alongside a moratorium on new data-center interconnections, means Texas buyers should price interconnection delay as delivery risk…

Cooling Capacity Should Set the Server Procurement Envelope
A retrospective EIA forecast points to a better buying rule: accept server capacity only when the site can reject its heat across credible operating conditions.

Hormuz volatility exposed diesel backup as an unhedged margin line
When Brent crude fell from $118 to $72 a barrel between April 29 and June 26, data centers learned standby generators sized for AI density carry fuel-cost exposure.