Explained — Second Order by TKN

Dusk aerial of a large data center roof with rows of cooling units and backup generators, overlay Biggest AI data center? Count the live megawatts

Explained

What Is the Biggest AI Data Center in the World? Check the Meter

By power actually switched on, the biggest AI data center in October 2026 is Amazon’s New Carlisle campus in Indiana, at an estimated 1,037 MW. By plans, it is Meta’s 5 GW Hyperion in Louisiana, which is not running yet. The gap between the two lists is where the money waits.

Who tests AI chips: macro photograph of a multi-site probe card in night-blue light, with one test site lit amber as the time each chip spends on the tester

Explained

Who Tests AI Chips? Testers, Probe Cards and Test Houses

AI chips are tested by four kinds of company: tester makers, probe card and interface makers, outsourced test houses, and foundries and chip firms with their own floors. Longer tests per chip make tester time the stuck step.

Explained

What Is an NVIDIA Cloud Partner? Why Offtake Waits on the API

An NVIDIA Cloud Partner is a specialist GPU cloud in NVIDIA's partner network. Offtake needs the API 12 weeks before the cards. Software readiness, not the badge, is the stuck step.

Who builds AI data centers: aerial of finished data center halls, graded pads and a substation under construction in night-blue light, with the substation lit amber as the power-on step

Explained

Who Builds AI Data Centers? Developers, Landlords and Who Gets Paid

AI data centers are built by five kinds of company: hyperscalers, private developers, listed landlords, former bitcoin miners and the contractors on site. Contractors are paid as they build; everyone else waits for power-on.

Explained

What Is Provisioned Throughput? Why Idle Inference Seats Still Bill

Provisioned throughput is reserved model capacity billed by the hour, not by the token. Idle seats still bill. Live capacity, not quota, is the stuck step.

Explained

Finance Lease vs Operating Lease: Same Data Centre, Two Ledgers

A finance lease is booked like buying with borrowed money; an operating lease is booked as rent. Same building, same rent, different lines. Microsoft's new 25-year building life moves leases off its capex line.

Explained

What Is SoIC? Why Stacking Chips Still Leaves HBM Beside the GPU

SoIC is TSMC’s wafer-level 3D stack: chips bonded on top of chips, then placed in CoWoS. HBM still sits beside the GPU. Heat, not the copper join, is the bind.

Aerial of data center halls and their electrical substations in Loudoun County, Virginia, with overlay Follow the substation, where AI data centers are

Explained

Where Are AI Data Centers Located? Follow the Substation

AI data centers sit in a few dense clusters, led by Northern Virginia and Texas, with Johor, Singapore, Sydney and Melbourne in Asia-Pacific. They go where a utility can promise megawatts on a date, so the map of signed power is far bigger than the map of halls switched on.

Explained

What Is Silicon Photonics? Why the Laser Is the Stuck Step

Silicon photonics makes the optical parts of a data link on silicon chip lines. Silicon can steer light but cannot make it well, so every link still needs an indium phosphide laser, and that laser is where AI optics is short.

Reserved vs on-demand GPU: a dark aisle of identical locked rack cabinets, one cabinet door lit amber as the reserved rack that bills whether or not it runs

Explained

Reserved vs On-Demand GPUs: Who Pays for the Idle Hour?

A reserved GPU bills for its whole term; an on-demand GPU bills only while it runs. Neoclouds live on the first, the second sets the price signal, and in 2026 the discount for committing has been shrinking.

Data center tiers: a row of identical backup generator sets beside a data center at dusk, one set lit amber as the spare, with the words paying for the spare

Explained

What Are Data Center Tiers? Tier I to IV in AI Data Centers

Data center tiers grade the spare generators, batteries and second paths that keep a hall running. That spare kit is bought but rarely used, and AI builders are now splitting over how much of it a GPU hall needs.

Capex vs depreciation: stacks of still-bagged compute boards in front of supercomputer racks, one stack lit amber, with the words bought now, expensed later

Explained

Capex vs Depreciation: Why AI Spending Hits Profit Later

Capex is the cash out today; depreciation is the bill that reaches profit over the next five to six years. The five biggest AI buyers spent about four dollars in cash for every dollar of depreciation they booked in their latest quarters, and the clock only starts when a powered hall switches on.