Explained — Second Order by TKN

Explained

What Is an Interposer? The Silicon Sheet Under AI Chips

An interposer is the thin layer, usually silicon, that sits under an AI processor and its memory and wires them together. Every CoWoS package needs one, it is made in a wafer fab, and new supply from GlobalFoundries does not arrive until 2028.

Data center cooling towers with overlay 0.27 vs 2.1 L/kWh, Microsoft fleet WUE versus Singapore PUB data-centre median

Explained

How Much Water Does an AI Data Center Use?

An AI data center uses water mainly to cool servers. Microsoft’s fleet WUE was 0.27 L/kWh in FY25; Singapore’s PUB median is 2.1 m³/MWh. Who pays, and where wet cooling sticks.

Photoreal CoWoS full silicon interposer glowing amber beside a dark EMIB organic substrate — the full sheet is the bind

Explained

CoWoS vs EMIB: Two Ways to Join an AI GPU — and Who Gets Paid

CoWoS puts an AI GPU and HBM on a full interposer. EMIB embeds small silicon bridges in the substrate. The stuck step is still the CoWoS slot, not the second recipe.

Explained

GPU Financing Explained: Loans That Draw as the GPUs Go Live

GPU financing lends against GPU servers and the customer contract that pays for them. The loan is drawn as servers go live and repaid from the customer's cash, so the stuck step is the day the customer accepts the site.

Explained

What Is a GPU Cluster? Why Signed Racks Are Not a Computer

A GPU cluster is racks, fabric and power sold as one computer. NVIDIA SuperPOD builds it in eight-rack units. The stuck step is signed megawatts that are not yet live.

Explained

Data Center Leases Explained: Rent Starts When the Power Does

A data center lease rents powered capacity, priced per kilowatt a month, not per square foot. The money arrives late: rent starts when the hall is powered and handed over, so the gap between signed and billing is the stuck step.

Explained

What Is Rack-Scale AI? Why the Unit of Compute Is Now the Whole Rack

Rack-scale AI is a whole rack sold as one computer. NVIDIA’s NVL72 ties 72 GPUs on one NVLink domain. The stuck step is the last missing tray, switch or watt.

What is AI Capex — AI CAPEX, cash leaves before hours bill: night data-hall construction and power bay with amber on the cash-leaving bind

Explained

What Is AI Capex? Hyperscaler Spend Before AI Hours Bill

AI capex is capital spent on GPUs, servers, networking and data-centre build so AI workloads can run — cash that leaves before those hours bill as cloud or ads revenue. Who gets paid first: chips, landlords, power. The stuck step is the lag between spend and billed utilisation.

Hyperscale vs colocation data center: caged colocation suites with yellow fibre on the left, a vast uniform hyperscale hall on the right, with the words $280 vs $157 per kW, colocation vs hyp

Explained

Hyperscale vs Colocation Data Centers: Same Power, Different Rent

A hyperscale data center is a huge hall for one cloud or AI giant; a colocation data center rents cages, racks and connections to many tenants. Per kilowatt, colocation pays far more. Per lease, hyperscale lasts twice as long, and power scarcity is now repricing the big leases.

Who makes advanced packaging — OSAT COMPANIES, packaging slots are the stuck step: a dark semiconductor clean room with one packaging bay lit amber

Explained

Who Makes Advanced Packaging? OSAT Companies and TSMC CoWoS

OSAT companies are paid to package and test chips they did not make. For AI accelerators the scarce finish is advanced packaging — ASE, Amkor and JCET on the outsourced side, TSMC CoWoS on the captive side. Packaging slots are the stuck step.

Explained

Which Hyperscaler Capex Lines to Watch This Earnings Season? Who Gets Paid Down the Chain

Watch six lines, not one capex headline. Cash pays chip, memory and server vendors first; leases pay landlords later; rented capacity pays neoclouds as a cost.

Explained

What Is AI Training? Why the Run Needs the Whole Cluster

AI training is the long, planned job that teaches a model on a reserved GPU cluster. The stuck step is the last missing piece: the run does not start until every GPU is live.