AI Infrastructure, Power & Semiconductors
Archive current through 4 August 2026
AI is usually described in models and tokens. The build-out arrives in megawatts, cooling water, transformers, fabs, leases and grid queues.
This hub gathers Long Arc News analysis of the physical and financial infrastructure behind compute. It follows electricity pricing into data-centre siting, power contracts into curtailment rights, tenant credit into project finance, and semiconductor capacity into industrial policy. It also keeps the institutional question in view: who grants access, who carries the obligation, and who can stop the machine when the system is tight?
On 4 August 2026, When the Electron Learned to Tell Time followed China’s quarter-hour power pricing into data-centre geography. Other essays move through leases, machine evaluation, semiconductor concentration and the values embedded in automated systems. Each entry retains its original title. Its description names the entity and mechanism beneath the literary surface.
The point is not to reduce AI to electricity. It is to read software at the altitude where it becomes land, water, wire and credit. When compute demand rises faster than the infrastructure around it, which contract becomes the real model?
Selected analysis
When the Electron Learned to Tell Time
4 August 2026
China’s spot-power reform makes the quarter-hour cost of electricity a direct operating and siting signal for AI data centres.
The Tenant of Record
28 July 2026
AI data-centre debt ultimately rests on tenant identity, lease enforceability and what stands behind the rent.
The Right to Stop
24 July 2026
Curtailment rights let AI data centres monetize electricity by stopping compute when grid power is more valuable than model output.
I Lied to China’s New AI. It Caught Me Twice.
17 July 2026
Kimi K3’s lower token price and fact-checking performance challenge the scarcity premium embedded in Western AI laboratories.
The Libraries That Burned Twice
5 April 2026
Nalanda and Baghdad frame advanced-chip concentration as permanent-impairment risk when knowledge, fabrication and talent compound in too few places.
The Friday Deadline
26 February 2026
Pentagon pressure on Anthropic’s weapons guardrails shows how procurement power can compress the moral space between building AI and deploying it.
When India Said No to Bill Gates
22 February 2026
India’s AI-summit response argues that technological acceleration must remain subordinate to human dignity, cultural identity and national ethical agency.
Rock, Water, and Wire
15 February 2026
AI’s scaling constraint lies in mines, water, power, transformers and grid permission rather than software alone.
Your Children’s Teachers Will Be Machines. Who’s Writing Their Values?
6 February 2026
Because AI training data encodes cultural priors, societies that do not participate risk outsourcing the values machines teach their children.
Continue through Long Arc News: Browse the complete Library · Energy & Chokepoints · AI Infrastructure · Capital & Industrial Policy · Insurance & Supply Chains · About Long Arc News