I'll interpret "out of 100" as asking for a percentage share breakdown of data center capital expenditure across major segment/product categories. Let me note this is an estimate-based industry view, not a single company's disclosed figure.

One-line explanations

• Servers & compute — the bulk of spend, driven by high-cost GPU/accelerator systems for AI and general compute.
• Networking — high-speed switches, optical transceivers, and NICs that connect nodes at scale.
• Storage — flash and disk arrays plus controllers for hot and cold data tiers.
• Power & cooling — UPS, distribution, and increasingly liquid cooling for dense racks.
• Racks, cabling & facilities — the physical shell, racking, and structured cabling to house gear.
• Software & services — management/orchestration licenses, plus installation and integration labor.

These are approximate industry ranges and shift meaningfully with AI-heavy builds (compute and cooling skew higher) versus traditional cloud/enterprise data centers.

If you want this tied to a specific company (e.g., Microsoft, Amazon, Meta, Nvidia, or a hardware vendor), I can pull the actual disclosed capex or segment figures.

Here's a further split for an AI inference data center (per 100 units of capex).

One-line explanations

• GPU/accelerators — dominant cost in inference builds, dominated by Nvidia (with AMD emerging).
• CPU — host processors, split between Intel and AMD.
• Server OEM — system assembly and boards from Dell, Supermicro, Quanta, Foxconn.
• Switches — high-speed fabric from Arista, Cisco, Broadcom, and Nvidia (Mellanox).
• Optics — pluggable transceivers from Coherent, Lumentum, Innolight, Eoptolink.
• NICs/DPUs — SmartNICs and DPUs from Broadcom, Nvidia, Intel.
• SSDs — flash storage from Samsung, SK hynix, Micron, Kioxia.
• HDDs — bulk/cold storage from Seagate and Western Digital.
• Power — UPS and distribution from Vertiv, Eaton, Schneider Electric.
• Cooling — increasingly liquid cooling from Vertiv, Delta, CoolIT, Boyd.
• Cabinets & cabling — Rittal, Panduit, CommScope for physical infrastructure.
• Construction — Turner, DPR, and general contractors for the shell/mechanical.
• Management software — VMware, Red Hat, HashiCorp, Palo Alto for orchestration/security.
• Integration — IBM, Accenture, HPE for deployment and integration services.

Caveat: these are directional estimates for an AI-inference-focused build — inference skews slightly higher on accelerators and networking versus training, and liquid cooling share is rising fast.