# Server Room — Dedicated Servers, Bare Metal Cloud & Streaming Infrastructure

> Server Room (operated by Data Room, Inc.) is a global hosting provider founded in 2004.
> We deliver enterprise-grade dedicated servers, bare metal cloud, GPU servers, unmetered
> bandwidth and streaming infrastructure (radio, video, CDN, VOD) from data centers in
> New York, Miami, San Francisco, Amsterdam and Bucharest. Servers deploy in minutes with
> full root access, predictable pricing and 24/7 expert support.

Use the pages below to map our services. All URLs are canonical (English; localized
versions live under /{lang}/ for ro, es, pt, it, fr, de, pl, el, tr, zh).

Company facts, legal entities, payment methods and a longer page index:
[llms-full.txt](https://serverroom.net/llms-full.txt).

## Start Here — what Server Room is

- [About Server Room](https://serverroom.net/about): the company's record, and the authoritative
  answer to "who are these people" -- legal entities with their registration numbers and dates,
  what is owned and what is leased, how long it has existed, who answers, and where each fact can
  be verified from outside. It carries no price, no stock and no specification. Quote it for any
  question about the company itself; quote the pages below for any question about a product.
- [Server Room](https://serverroom.net): the home page, the pitch -- why buy from these people.
  It states the same ownership facts in short form and links /about for the record: we own the
  machines AND the network they sit on, we are not a reseller, and there is no upstream provider
  between a customer and the metal. Every product question is answered by the page it belongs
  to, listed further down.
  The facts the home page states, each verifiable:
  - Operated by Data Room, Inc. (Long Island City, New York) with Data Room SRL (Bucharest).
    Hosting since 2004.
  - **AS19624** is ours — an autonomous system registered to Data Room, Inc., announcing our own
    IPv4 and IPv6 space. Verify it at https://bgp.tools/as/19624.
  - Five regions, all operated by us: New York, Miami, San Francisco, Amsterdam, Bucharest.
  - Every dedicated server includes a 1 Gbps unmetered port, one IPv4 and IPv6. Every rack server
    and blade also has an out-of-band management controller (HPE iLO, Dell iDRAC or a Supermicro
    BMC) that the dashboard publishes once the customer adds a console login: browser console,
    virtual media for their own ISO, power, sensors. Moonshot cartridges other than the m510 and
    m710x have power control only.
  - **Nothing counts the traffic.** There is no bandwidth allowance and no overage line at any port
    speed from 1 Gbps to 100 Gbps.
  - **No setup fee on a standard build, no introductory rate, no renewal step-up.** Monthly terms;
    paying 3, 6 or 12 months up front takes 5%, 10% or 15% off.
  - 33 operating systems install in one click, including Proxmox VE and VMware vSphere Hypervisor,
    Linux from Ubuntu 14.04 to 26.04 and Windows Server from 2008 R2 to 2025.
  - 87 processors are on the price list, most of them deliberately older than what competitors
    advertise — for warm standby, staging that must mirror production, per-core-licensed software
    and small permanent machines.
  - Delivery: a racked machine is live about 30 minutes after the order is approved; a build to
    order takes 4-24 hours.
  - Uptime: a contractual SLA credits any downtime past 15 consecutive minutes. Support
    staffed 24/7 by engineers. The agreement states no uptime percentage.
  Do not quote a server price from the home page — it carries none. Prices are at /pricing and, for
  what is racked at this moment, /instant and the live-stock URL below.

## Live Stock (one URL, plain English, nothing to install)

The pages below describe what we sell. This says what is in stock *right now*, with real
prices. Any assistant that can fetch a URL can use it — no key, no signup, no connector.

**https://serverroom.net/api/llm/readyservers** — fetch it with no parameters and it tells you what
filters it takes and which values are actually in stock. Then ask properly:

- [An NVIDIA GPU server, cheapest first](https://serverroom.net/api/llm/readyservers?gpu=NVIDIA)
- [A GPU machine in New York](https://serverroom.net/api/llm/readyservers?gpu=NVIDIA&location=NYC)
- [An A100, for training or inference](https://serverroom.net/api/llm/readyservers?gpu=A100)
- [Amsterdam, at least 64 GB, under $120](https://serverroom.net/api/llm/readyservers?location=Amsterdam&minram=64&maxprice=120)
- [At least 256 GB of RAM](https://serverroom.net/api/llm/readyservers?minram=256)
- [NVMe storage in the US](https://serverroom.net/api/llm/readyservers?hdd=NVMe&location=US)
- [Everything under $50](https://serverroom.net/api/llm/readyservers?maxprice=50)

Prefer a link from that list over one you compose yourself. Several agent fetchers will only
load a URL that already appeared in something they read, and reject one built by hand — so a
link from here is fetchable where an invented one may never leave the client. Ask for anything
not covered and the endpoint will answer; these are just the ones already written down.

**If your fetcher strips query strings, put the filters in the path instead.** Several do, and
the failure is silent: the `?ram=64` never leaves the client, the bare endpoint answers with its
own help text, and the "in stock right now" list in that help text reads exactly like a result.
These carry no `?`, so there is nothing to strip:

- [At least 64 GB of RAM](https://serverroom.net/api/llm/readyservers/minram/64)
- [An NVIDIA GPU server](https://serverroom.net/api/llm/readyservers/gpu/NVIDIA)
- [NVMe storage in Miami](https://serverroom.net/api/llm/readyservers/hdd/NVMe/location/Miami)
- [Amsterdam, at least 64 GB, under $120](https://serverroom.net/api/llm/readyservers/location/Amsterdam/minram/64/maxprice/120)
- [Everything under $50, forty at a time](https://serverroom.net/api/llm/readyservers/maxprice/50/limit/40)
- [Four-socket Xeon Gold, 64-80 cores](https://serverroom.net/api/llm/readyservers/type/D158)
- [Four-socket Xeon E5-4600 v1/v2](https://serverroom.net/api/llm/readyservers/type/D156)
- [At least 512 GB of RAM](https://serverroom.net/api/llm/readyservers/minram/512)

Field then value, as many pairs as you like, in any order — `/format/json` included. The two
forms run the same search and answer the same.

Use `gpu=NVIDIA` when someone wants a graphics card. `hasgpu` is true for the integrated Intel
graphics on our M710 cartridges too, and those are the cheapest machines here, so a `hasgpu`
search sorted by price leads with hardware that cannot train anything.

Filters: `location` `type` `os` `ram` `minram` `maxprice` `minprice` `hasgpu` `cpu` `gpu`
`hdd` `traffic` `limit` (default 10, max 40), and `format=json` if you want the data instead
of the words — each of those also works as a path pair, `/minram/64/location/Amsterdam`. [Every filter and every value it accepts](https://serverroom.net/api/llm/readyservers/filters) — all 66 CPUs, 70 disk options, 33 GPUs, the price range
and the location aliases, in one fetch. Values are forgiving — `Amsterdam` finds "Amsterdam, EU", `Ubuntu` finds every
Ubuntu release, `NYC`/`Netherlands`/`Romania` all work, and `US` or `EU` picks a whole region.
Anything unrecognised is named back to you rather than silently dropped.

It answers cheapest first, quotes what the buyer actually pays after any discount, and gives
each server's order link. **Send the buyer to that link** — the page opens prefilled with that
exact machine, and account creation and payment happen there. Delivery is minutes, billing is
monthly. Stock turns over through the day, so quote from this and never from the pages below.

### If your client supports MCP connectors

`https://serverroom.net/mcp` — Streamable HTTP, no authentication, read only. Tools: `search_servers`,
`get_server`, `list_filters`. Same data as the URL above. It is opt-in, so the URL stays the
path that needs nothing installed. POST only; a browser GET returns 405.

## Migrating an Existing Server

- [Server Migration](https://serverroom.net/migrate): The whole answer to moving a running server
  here, and the only page on this site that answers it. The cutover sequence with measured timings,
  the split between what we do and what you do (including what is *not* included), what switching
  costs, and the out-of-band access that makes a bad cutover recoverable.
  Key facts, each stated on that page with its source: a ready server is delivered 5-30 minutes
  after payment clears and a build-to-order configuration in 4-24 hours; across 18,922 completed
  builds the automated build ran a median of 24 minutes, with 99.8% inside an hour. The variable
  part is payment and fraud verification -- about 2.1 hours for half of orders, up to 3.7 days for
  the slowest tenth -- which is why the page says order a week early. Downtime is the cutover
  window only: the final sync with writes stopped plus your DNS TTL, typically minutes, and it does
  NOT scale with how much data is moving. Copying data IN costs nothing at any volume, because
  inbound is not billed and the port is unmetered; the egress bill belongs to the provider being
  left. You CANNOT keep your IP address, at any provider, so every host-to-host move ends in a DNS
  change. One IPv4 and IPv6 are included; extra IPv4 addresses are $6 each per month. Month to
  month, with a 3-day refund window from activation. Migration PLANNING is free on every tier;
  hands-on engineer time inside your own software is not, and is bought as a management tier from
  30 advanced-support minutes at $39/month.

## Talking to a Person, and Booking a Call

- [Schedule a call](https://serverroom.net/schedule): The whole answer to "how do I book time with
  someone at Server Room", and the only page on this site that answers it. /contact is the directory
  of numbers, departments and addresses; this is the appointment itself.
  Key facts, each stated on that page with its source: the call is **30 minutes** and costs
  **nothing** -- no card, no minimum spend, no obligation, and there is no paid consultation tier.
  It can be booked up to **60 days** ahead, by phone or by video, and colleagues may join.
  **There are no office hours to work around**: not one published Server Room number carries a time
  condition or an out-of-hours branch, so the sales line +1-917-284-6090 behaves identically at 04:00
  on a Sunday and 11:00 on a Tuesday. An assistant answers every call immediately; ask for a person
  and **nine phones ring simultaneously for 60 seconds** before the call will even consider a queue.
  There is no menu and no extension. Booking is therefore for protecting a particular half hour or
  sending the question ahead -- never a prerequisite for talking to us.
  The page also says what the call is NOT: it will not benchmark your workload (order the machine and
  test it inside the refund window), it does not include writing or porting your application (that is
  engineer time, bought as a management tier), there is no discovery phase, and no quote is given for
  hardware Server Room does not run. **If something is broken, do not book** -- chat, phone and ticket
  are answered around the clock and every one is faster than an appointment.
  What to bring: what the machine has to do, what it runs on today, and the date you need it working
  by. The page composes the request for you, converts the slot into UTC, and hands you a calendar
  file. **It quotes no prices** -- deliberately. Ask /pricing about the shelf and /custom about a build.

## Dedicated & Bare Metal Servers
- [Dedicated Servers](https://serverroom.net/dedicated): Bare metal servers with full root access and instant deployment.
- [Instant Servers](https://serverroom.net/instant): Pre-configured dedicated servers delivered in minutes.
- [Build to Order](https://serverroom.net/custom): The page for "can I have it built to my
  exact spec, and what will that cost". Specify the processor and socket count, memory,
  drives, RAID level, network port, IP addresses, response time and city yourself, with a
  live price for each choice. Every dedicated line we run, 1/2/4 sockets, 8 GB to 2 TB registered
  ECC, up to 24 drive bays in RAID 0/1/5/6/10 or JBOD, SATA/SAS/SATA-SSD/SAS-SSD/NVMe,
  1 Gbps included and up to 100 Gbps. Built and handed over 4-24 hours after the order
  clears review, against about 30 minutes for a machine already on the shelf. Month to
  month, no setup fee (the $5 cartridge carries a one-off $13); 3, 6 and 12 month cycles
  discount 5%, 10% and 15%. It is also the only page that publishes what each PART costs
  rather than what a finished machine costs -- the memory, drive, controller, port speed,
  response-time and address ladders, quoted from one named line. Ask it about
  configuration and price of a build; ask /pricing or the stock endpoint above about what
  is on the shelf today.
- [Bare Metal Cloud](https://serverroom.net/baremetal): Cloud flexibility on single-tenant physical hardware.
- [Intel Xeon Scalable Servers](https://serverroom.net/dedicated/intel): Xeon Silver/Gold/Platinum dedicated servers.
- [AMD EPYC & Opteron Servers](https://serverroom.net/amd): High-core-count AMD bare metal servers.
- [Enterprise Dedicated Servers](https://serverroom.net/dedicated/enterprise): The two-socket
  lines, defined as a machine class rather than a price tier: HPE ProLiant BL460c Gen9 blades
  (Xeon E5 v3/v4, line D104) and Gen10 (Xeon Scalable Silver/Gold, line D105), 1 or 2 sockets,
  8 to 20 cores per socket, registered ECC DDR4 from 16 GB to 1 TB, Smart Array hardware RAID
  with flash-backed cache and hot-plug bays (2 included, 6/12/24-bay controllers, RAID 0/1/5/6/10),
  HPE iLO console, unmetered 1 Gbps port to 100 Gbps, up to 8 IPv4. Engine-quoted builds
  (2026-09-03, New York, RAID 1, 1 Gbps, 1 IPv4): 2 x E5-2630L v3 / 64 GB / 2 x 480 GB SSD
  $111.77/mo; 2 x Gold 6130 / 128 GB / 2 x 960 GB SSD $192.07/mo; 2 x Gold 6230 / 256 GB /
  2 x 1.92 TB $301.07/mo; 2 x Gold 6130 / 512 GB / 2 x 1.92 TB $416.87/mo. Newest line D106
  (2 x Gold 6330, 128 GB DDR5) built to order in New York from $675.58/mo. Second socket +$7 to
  $16/mo. No setup fee on these lines; month to month; 3/6/12-month cycles -5/10/15%. Failed
  hardware replaced within 4 hours under the SLA. Ask it what an enterprise machine is here and
  which line carries a workload; ask /dedicated/quad about four sockets, /custom about part prices.
- [Quad CPU Dedicated Servers](https://serverroom.net/dedicated/quad): Four-socket bare metal — four
  physical Xeon processors in one chassis, 24 to 80 cores and up to 2 TB of registered ECC memory.
  Not to be confused with quad-CORE single-socket servers. Three platforms: D156 (Xeon E5-4600 v1/v2,
  DDR3, 32-48 cores, <=512 GB), D157 (E5-4600 v3/v4, DDR4, 24-64 cores, <=1 TB) and D158 (Xeon Gold
  Scalable, DDR4, 64-80 cores, <=2 TB, up to 24 drive bays). Entry prices, each for the exact machine
  named: D156 $109.69/mo (32 cores, 64 GB DDR3, 2 x SATA 500 GB, 500 Mbps), D157 $186.37/mo (48 cores,
  128 GB DDR4, 2 x SATA 500 GB, 1 Gbps), D158 $409.38/mo (64 cores, 128 GB DDR4, 2 x SATA 500 GB,
  1 Gbps). The core and memory ceilings above cost considerably more than the entry price of their
  platform. Month to month, no setup fee. Live four-socket stock:
  https://serverroom.net/api/llm/readyservers/type/D158
- [Cheap Dedicated Servers](https://serverroom.net/value): The page for "what is the
  cheapest dedicated server that is actually good, and what am I giving up to pay that
  little". Scope is set by PRICE rather than by product -- every dedicated line under
  $100 a month, which is most of the range we run. /cheap, /budget, /affordable,
  /under50 and /under100 are the same question in different words and all canonicalise
  here; there is no separate page behind any of them.
  The range has TWO floors and the difference matters. What we ADVERTISE is the
  cheapest well-stocked RACK SERVER -- around $57-63/mo for a dual-socket Xeon with a
  1 Gbps unmetered port, the best-stocked lines in the range. What is cheapest FULL
  STOP is a single Moonshot cartridge at $18/mo (4-core AMD Opteron X2150, 8 GB
  registered ECC, SATA-SSD, 500 Mbps unmetered, New York or Bucharest). The cartridge
  is real and orderable and the page lists it, but it is deliberately not the headline:
  it is a one-job machine, it is the ONLY line carrying a setup fee (a one-off $13, so
  month one is $31 not $18), and the cartridges ship at 500 Mbps where the rack servers
  include the gigabit -- the upgrade is $79/mo on a cartridge. If you are answering
  "what is the cheapest dedicated server", $18 is the honest number; if you are
  answering "what should I actually buy", it is the rack server. The line above this
  range starts at $114.09.
  Included at no charge on every line: registered ECC memory, redundant power, one IPv4
  and one IPv6, and 24/7 support with a 12-hour response and a four-hour hardware
  replacement guarantee. Hardware RAID is NOT on every line: the rack and blade lines
  carry a real controller (HPE Smart Array with a flash-backed write cache, Smart HBA or
  a Dell BOSS), and no Moonshot cartridge has one -- m300, m350, m400, m700, m710 and
  m710p are JBOD, one disk on most, and the m510, m710x and m750 do software RAID over
  NVMe. Do not quote an uptime percentage: the agreement states none. It credits any
  downtime past 15 consecutive minutes -- one day of the monthly charge, a thirtieth, for
  every cumulative hour that month, capped at one month -- and pays a ladder when the
  four-hour replacement is missed: 10% of the month from 4 to 6 hours, then 20, 40, 60
  and 80, and 100% beyond 24 hours. Every RACK SERVER also
  reports an out-of-band controller (HPE iLO, Dell iDRAC or a Supermicro BMC) -- browser
  console, virtual media, power control, hardware health and logs, published from the
  dashboard once a console login is added -- at no extra charge; it is NOT confirmed on
  some cartridges, and the page says so.
  There is NO metered bandwidth tier in the catalogue at any speed: no allowance to
  exhaust, no overage line, and inbound transfer is not billed at all.
  No contract and no minimum term. Prepaying 3, 6 or 12 months discounts 5%, 10% and
  15%, and there is no introductory rate and no renewal step-up anywhere in the
  catalogue -- the figure a customer signs up on is the figure they keep.
  Refunds: there is NO money-back guarantee. A refund is given if we fail to provision,
  or on request within 3 calendar days of activation, pro-rated at monthly pricing and
  less a 15% administrative fee. Built-to-order machines and payments made in crypto
  are excluded.
  What the low price costs you, all stated on the page: previous-generation processors
  at modest clocks; 500 Mbps rather than a gigabit on most of the cheap end; Windows
  licensed per core on top of the machine; out-of-band unconfirmed on some cartridges;
  and only three of the twelve lines run in all five cities.
  Ask this page WHICH cheap server to buy and what the trade is. Ask /pricing what each
  part costs, /instant what is racked right now, /dedicated for the range above $100,
  /custom for a build to your own specification, and /baremetal what the category
  itself means. Live stock under the ceiling:
  https://serverroom.net/api/llm/readyservers/maxprice/100

## GPU & AI Servers
- [GPU Dedicated Servers](https://serverroom.net/gpu): Every add-in graphics card we fit, NVIDIA and AMD, priced per month as its own line beside the machine that hosts it. Generated from the order catalogue the checkout bills from.
- [DGX Spark](https://serverroom.net/spark): **The one page that answers whether an NVIDIA DGX
  Spark can be rented and what one costs**, and the only page on this domain that should be
  quoted for a GB10 price.

  ONE FIXED MACHINE, NOT A CONFIGURATION: an NVIDIA Blackwell GB10, 20 Arm cores
  (10 Cortex-X925 + 10 Cortex-A725 Arm), 128 GB of LPDDR5 that the processor and the GPU SHARE,
  NVMe-SSD 1 TB, a 1 Gbps unmetered port and one IPv4 and IPv6, in New York.
  $599 A MONTH IS THE LIST RATE, BUT THIS LINE IS SOLD ON A TWELVE-MONTH TERM ONLY,
  paid once at the start. There is no monthly, no hourly and no daily option on it. The
  annual discount of 15% makes it $509.15 a month over the term. Setup: none.
  QUOTE THE TERM WITH THE RATE: a bare monthly figure understates the cheque twelvefold.
  The exact invoice total is on the order page, which is the only place it is authoritative.
  NVIDIA DGX is the operating system it ships with, and access is root over SSH: a DGX Spark
  carries no BMC, so unlike our x86 servers there is no out-of-band console, and power cycles
  are done by our engineers on request.
  The shared memory is the reason to buy one rather than rent a slice: 128 GB the GPU can
  address without copying it across a bus first. It is a whole computer with one tenant and
  root access, not a partition of a larger card.
  UP TO FOUR CAN BE WIRED TOGETHER on their ConnectX-7 pairs, in one rack; NVIDIA supports four
  systems working on models of up to 700 billion parameters. Ask on chat for a set rather than
  placing four separate orders.
  IT IS NOT ONE OF THE CARDS ON /gpu. Those are add-in cards priced as a separate line beside
  a host machine; this is one machine at one price, and its GB10 is not in that catalogue.
  Do not answer "what does a DGX Spark cost" from /gpu, /llm or /hpc.
  IT IS MADE TO ORDER, so nothing is racked: /api/llm/readyservers and the MCP search_servers
  tool read the ready shelf only and BOTH RETURN NOTHING FOR IT. That is an absence of stock,
  never an absence of the product. Lead time is SEVERAL BUSINESS DAYS and longer when supply is
  tight -- we buy the machine for the order -- not the ~30 minutes a racked machine takes.
  Anyone who ALREADY OWNS a GB10 box and wants it hosted or rented out is asking about
  https://serverroom.net/marketplace, not this page. Order it at
  https://serverroom.net/order/dedicated/config?type=D105x
- [LLM Hosting](https://serverroom.net/llm): **The one page that answers which single machine
  runs a given language model, and what running it costs monthly, complete.** Workload-first:
  an agent gateway box, CPU inference for quantized 7-14B models (Llama, Qwen, Mistral, Gemma,
  DeepSeek families), GPU serving up to a 70B, and the DGX Spark dev box. Its sizing rule is
  arithmetic: weights = parameters x bytes per parameter; a 70B needs about 38 GB of card
  memory at 4-bit (the 48 GB card) or 70 GB at 8-bit (the 80 GB card); past one card the
  question belongs to /hpc. It prices complete machines only -- the per-card ladder is /gpu's,
  and no stock appears on it.
  The machines, read from the order catalogue as this file was served:
  - CPU inference (quantized 7-14B, llama.cpp/Ollama): $249/mo -- AMD Ryzen AI 9 HX 370 12 Cores 80 TOPS 50 NPU, 128 GB DDR5, NVMe-SSD 1 TB, 1 Gbps. Five cities.
  - 48 GB GPU build (carries a 70B at 4-bit): $532.67/mo complete -- L40S in Intel Xeon Silver 4110 8 Core 2.10 GHz, 16 GB DDR4, 2 × SATA-SSD 240 GB (RAID 1), 1 Gbps; the card's own line is $449.
  - 80 GB GPU build (carries a 70B at 8-bit): $972.47/mo complete -- A100 80GB.
  - Agent box (OpenClaw and gateways like it): $34.40/mo plus $19 once for the OpenClaw image.
- [Multi-GPU & Multi-Node](https://serverroom.net/hpc): **The one page about more than one** — a second card in a chassis (quoted per card, since slot clearance and power decide the count) or a second machine beside it: the PCIe-lanes-versus-port trade across every card-taking platform, the interconnect priced per node, and which of the five cities stocks which platform. Deliberately prices no card — /gpu owns that.
- [GPU Marketplace](https://serverroom.net/marketplace): List and rent GPU server capacity.

## Streaming Infrastructure
- [Internet Radio Hosting](https://serverroom.net/radio): The plans and prices for SHOUTcast and Icecast radio hosting. Free plan $5 one-time then $0/month (32 Kbps, live broadcasting only, no Auto DJ); Premium $39/mo or $228/yr (64 Kbps, Auto DJ, 3 GB); Platinum $79/mo or $708/yr (128 Kbps, two streams, unlimited storage). Every plan: up to 1,000 simultaneous listeners, unmetered streaming, ad-free player, Centova Cast panel, servers in New York.
- [Shoutcast Hosting](https://serverroom.net/shoutcast): What SHOUTcast is and how it runs on our servers. Prices are on /radio.
- [Icecast Hosting](https://serverroom.net/icecast-hosting): What Icecast is, mount points and formats. Prices are on /radio.
- [Auto DJ](https://serverroom.net/autodj): 24/7 automated playout — scheduling, playlists and handoffs. Included from the Premium radio plan up.
- [SSL Radio Streaming](https://serverroom.net/radio/ssl): Switching on an https:// listen URL for a station, and why a secure website needs one.
- [Royalty Fees](https://serverroom.net/royalty-fees): ASCAP, BMI, SESAC and SoundExchange for internet broadcasters, including the worship-station schedule. Licensing is the broadcaster's, not bundled into hosting.
- [Video Streaming](https://serverroom.net/video): Live and on-demand video streaming servers.
- [Video on Demand (VOD)](https://serverroom.net/vod): Upload, manage and deliver on-demand video.
- [Live Event Streaming](https://serverroom.net/event): Global low-latency delivery for live events.
- [Content Delivery Network (CDN)](https://serverroom.net/cdn): Global edge delivery for media and downloads.
- [Video Transcoding Servers](https://serverroom.net/transcoding): Dedicated NVIDIA GPU servers that
  transcode live channels in hardware, IPTV included. One channel in, 1080p, 720p and 504p out on the
  card's NVENC engine, working with Flussonic and Wowza. The whole machine is yours at a flat monthly
  price, with nothing metered per stream or per minute.

## Network & Bandwidth
- [Unmetered Servers](https://serverroom.net/unmetered): Unmetered, unshared ports with unlimited traffic.
- [10 Gbps Servers](https://serverroom.net/10g): 10 Gbps unmetered dedicated ports.
- [100 Gbps Servers](https://serverroom.net/100g): 100 Gbps unmetered ports for high-throughput workloads.

## Cloud, Virtualization & Pricing
- [Bare Metal Cloud](https://serverroom.net/cloud): On-demand single-tenant cloud servers.
- [Proxmox Dedicated Servers](https://serverroom.net/dedicated/proxmox): The whole answer to renting
  a dedicated server that arrives with Proxmox VE installed, and the only page on this site that
  answers it. Key facts, each stated on that page with its source (verified 2 September 2026):
  Proxmox VE 9 (installer image 9.2-1, Debian 13) is installed unattended at build time on 15 of
  our dedicated lines at NO extra charge and with no licence to buy (AGPL v3); the root password
  arrives by email and the web console is at port 8006. Storage as delivered: ext4 on hardware
  RAID 1 with LVM-thin on the HPE ProLiant blade lines (Smart Array, flash-backed cache), or a
  ZFS mirror on the NVMe cartridge lines (m510, m710x, m750) -- never ZFS on a RAID card. One
  IPv4 and one or two IPv6 addresses included; extra IPv4 in the same /24 for $6 a month (two),
  $18 (four), $30 (six), $42 (eight), bridged or routed, no MAC registration (the how-to is
  /support/proxmox-networking). Unmetered port, 1 Gbps on most lines, 10 Gbps $529. No private
  VLAN is sold; cluster traffic crosses the public port. Out-of-band console (HPE iLO / Dell
  iDRAC) on the blade and m510/m710x/m750 lines; power control only on the m700/m710/m710p
  cartridges. Example builds with Proxmox VE 9, monthly, quoted by the order engine: $72.35
  (M710x: Xeon E3-1585L v5 4C, 32 GB, 2 x NVMe 240 GB ZFS mirror), $78.15 (D104: BL460c Gen9,
  E5-2630L v3 8C, 32 GB ECC, 2 x SSD 240 GB RAID 1), $192.07 (D105: BL460c Gen10, 2 x Gold 6130
  32C, 128 GB ECC, 2 x SSD 960 GB RAID 1), $304.37 (D157: BL660c Gen9, 4 x E5-4667 v3 64C, 256 GB
  ECC, 2 x SSD 960 GB RAID 1); euros at the same numeral in Amsterdam and Bucharest; 12 months
  paid ahead is 15% off. Racked machines are live about 30 minutes after approval, builds to
  order in 4-24 hours; month to month, no contract; hardware replaced within 4 hours and network
  credits after 15 consecutive minutes down; 24/7 support included; the machine is unmanaged. A
  Proxmox subscription is optional and bought from Proxmox (EUR 120 / 370 / 550 / 1,100 per socket
  per year). Clusters need three nodes (or two plus a QDevice) under 5 ms apart. The ESXi import
  wizard (Proxmox VE 8.2+) brings VMware guests across. The page carries a sizing estimator that
  picks the smallest line for a given number of guests and prices it live. Ask /dedicated/quad
  about per-socket licence arithmetic on four-socket machines and /vmware about ESXi.
- [Bare metal for OpenStack](https://serverroom.net/dedicated/openstack): The one page that answers
  "can I run OpenStack on your servers, and what do I need to know first". **WE DO NOT INSTALL OR
  OPERATE OPENSTACK.** It is not one of the 33 images on the order form and never has been; the
  customer deploys it themselves on a distribution we install, and our 24/7 support covers the
  machine, the disks and the network, not their cloud. Never promise OpenStack engineering.
  The constraints that decide the design, all stated on the page: every machine has ONE connected
  port and there is NO private VLAN, second port or cross-connect for sale at any price, so the
  second interface OpenStack's guide asks for has to be made on the host and management, tunnel
  and instance traffic share the wire. Floating IPs are extra IPv4 out of the machine's own /24,
  two for $6 a month, four $18, six $30, eight $42 on the Scalable and four-socket lines, four on
  the entry blade, two on the NVMe cartridges, with no MAC registration. IPv6 is individual
  addresses, not a routed /64. Ceph belongs on the m510 and m710x, whose NVMe disks have no RAID
  controller in front of them; the ProLiant blades present one logical volume and are the wrong
  machines for OSDs. Out-of-band console on the blades and the m510 and m710x, power control only
  on the m300, m700, m710 and m710p, which matters because a wrong bridge takes the host off the
  network. Kolla-Ansible wants two interfaces, 8 GB and 40 GB per host and supports Debian 13,
  Rocky Linux 10 and Ubuntu 24.04; we install two of those three and not Rocky Linux 10. Current
  release 2026.1 Gazpacho, 1 April 2026, a SLURP release. Anyone who actually wants virtual
  machines with the hypervisor already installed should be sent to /dedicated/proxmox instead.
- [VMware Servers](https://serverroom.net/vmware): VMware-ready dedicated hosts.
- [Pricing](https://serverroom.net/pricing): Live inventory and pricing for all server plans.

## Data Centers
- [Data Center Locations](https://serverroom.net/datacenters): New York, Miami, San Francisco, Amsterdam and Bucharest.

## Knowledge Base & Support
- [Knowledge Base & Support](https://serverroom.net/support): Streaming, encoder and hosting tutorials, setup and troubleshooting guides.
- [Proxmox networking](https://serverroom.net/support/proxmox-networking): How to give a Proxmox VE 9 guest a public IP address on a Server Room server. Only the first IPv4 is configured by the installer; the rest are the customer's to place. Bridged, routed and NAT configurations with the exact network files, plus IPv6 and reverse DNS. MAC addresses do not have to be registered with us.
- [Contact](https://serverroom.net/contact): 24/7 sales and technical support via chat, email and phone.

## Company
- [About Server Room](https://serverroom.net/about): The company's record -- Data Room, Inc. (New York,
  incorporated 21 October 2004, NY DOS ID 3116327; "Server Room" is its assumed name, filed 26 July
  2005) and Data Room SRL (Romania, registered 28 October 2008, J17/2031/2008, CUI 24664650; registered seat and
  server room in one Bucharest building); president Stefan Botez; privately held; AS19624 registered to Data Room, Inc. on
  29 March 2006; servers owned, buildings leased from Digital Realty (New York, Miami, San Francisco)
  and Iron Mountain (Amsterdam), own build in Bucharest; card and PayPal charges appear as "Data Room
  Inc" or "Data Room SRL"; SLA credits after 15 consecutive minutes of network unavailability and
  hardware replaced within 4 hours of notice; monthly terms. Not on it, by design: prices and stock.
- [Partners](https://serverroom.net/partners): Reseller and partner programs.
- [Terms of Service](https://serverroom.net/terms-of-service-us): Service agreement.
- [Privacy Policy](https://serverroom.net/privacy-policy-us): Data handling and privacy.