ANVE: the private AI office, on the record. See what that means

ANVE

Private AI, answered · updated 2026-10-16

How much electricity does an AI server use?
Vendor and reviewer measurements, and the arithmetic from watts to rupees.

Short answer. A small AI server draws roughly 7 to 20 W when idle and 140 to 385 W when working flat out, depending on the machine. Apple measures 7 to 9 W idle and 200 to 385 W maximum at the wall for the Mac Studio, and ServeTheHome measured 10 to 20 W idle and 140 to 170 W under load on a Ryzen AI Max+ 395 desktop. An office box idles most of the day: 16 hours idle and 8 hours busy comes to about 38 to 97 kWh a month across those machines. Multiply by the tariff on your own bill.

01

What has been measured

Apple’s support page says its power figures are measured at the wall and include power-supply losses, and that “max” means a compute-intensive test. ServeTheHome ran its system at about 140 W for a day and a half without problems. These are other people’s machines: ANVE has not published power figures for its launch hardware.

MachineIdleHeavy loadBasis
Mac Studio, M5 Max (36 GB)7 W200 WApple, at the wall, maximum stress test
Mac Studio, M5 Ultra (512 GB)9 W385 WApple, at the wall, maximum stress test
Mac Studio, M4 Max (2025)6 W145 WApple, at the wall
Mac Studio, M3 Ultra (2025)9 W270 WApple, at the wall
Framework Desktop, Ryzen AI Max+ 395, 128 GB10 to 20 W140 to 170 WServeTheHome, Windows desktop idle and sustained load
02

Chip and card ratings, for comparison

PartRatingWhat it covers
AMD Ryzen AI Max+ 39555 W default, configurable 45 to 120 WThe chip only
NVIDIA RTX PRO 4000 Blackwell145 W maximumThe card only, not the rest of the computer
NVIDIA DGX Spark140 W for the GB10 chip; 240 W power supplyChip, and the whole machine’s supply
GeForce RTX 5090575 W for the card; 1,000 W system supply listedThe card, and the supply NVIDIA asks for
03

From watts to kWh to rupees

Energy in kWh is watts ÷ 1,000 × hours. The 8 busy hours are an assumption, not a measurement; an office box usually spends far less time generating. Cost is kWh × your tariff. At an example rate of ₹10 per kWh, the first row would cost about ₹384 a month and the third about ₹967. The rate is a round number for the example, so use the one on your electricity bill, which differs by state and by connection type.

Pattern (16 hours idle, 8 hours busy a day)kWh per daykWh per month (30 days)
10 W idle, 140 W busy (Strix Halo desktop, low end)1.2838.4
7 W idle, 200 W busy (Mac Studio M5 Max)1.7151.4
9 W idle, 385 W busy (Mac Studio M5 Ultra)3.2296.7
385 W for all 24 hours (worst case)9.24277
04

What drives the bill

  • Idle power, because the box is on all day. Over a year, a machine that idles at 7 W costs far less than a hypothetical one that idles at 60 W.
  • How many hours it spends generating, which depends on how many people use it and for what.
  • The machine class: a 385 W part running flat out costs nearly twice what a 200 W part does.
  • Cooling and power conversion: air conditioning removes the heat, and an uninterruptible supply adds its own losses.
05

Questions people ask

Does the machine use power when nobody is asking?
Yes: it draws its idle power all day, 7 to 20 W in the measurements above. Ollama unloads a model after 5 minutes of inactivity by default, which frees memory; the measurements above do not separate loaded from unloaded idle.
How much heat does 200 W produce?
About 680 BTU an hour. Apple’s support page lists 682 BTU/hr for the Mac Studio M5 Max at its 200 W maximum.
What size of uninterruptible supply do I need?
Size it to the machine’s maximum draw with a margin, not the idle figure. Vendors list power supplies of 240 W (DGX Spark) to 480 W continuous (Mac Studio), and a card such as the RTX 5090 asks for a 1,000 W system supply. Check the supply’s own rating and runtime.
06

What is not real yet

No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.

Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.

07

Sources