Pricing model v0.2 · stage 0 — data and anchor parties

Armenia's GPU exchange. Volume + time + demand.

T-Exchange is a two-sided exchange for GPU hours and tokens, operated from Yerevan, Armenia. Data centers and AI hardware owners turn spare capacity into revenue; companies get a minimal, predictable AI bill. One price formula: tokens + GPU time + a transparent demand surge k, shown before you confirm.

Configure a rental Try the demoList hardware
On-demand
k per hourly block, rise ≤ 0.3 per block; price shown upfront
Reserved and commitments
1–12 months: k = 1.0, discount 5 / 12 / 20 / 30%
Spot and bids
max(0.35, 0.5·k) × base; 2-min stop notice

Worked example (fictional demo)

Demo Factory North

EU · 2× GPU 80 GB · 24h

GPU compute

2 × 24h × $1.00

$48.00

Allocated disk

100 GB × 24h × $0.0001

$0.24

Ingress / egress

50 GB × $0.01 + 10 GB × $0.02

$0.70

Quote

$48.94

Final statement

after 18h running + 2h stopped

$36.90

Supplier
$29.52
Platform share
$7.38
Returned (simulated)
$12.04
Demo walkthrough: $48.94 → $36.90 → $12.04

Live resource catalogue

0 qualified offers · 2 active rentals

Browse offersDemo catalogue

Only verified, qualified supply is listed — nothing invented. Stopped storage keeps billing while retained; released containers stop all charges.

PowerPoint · 17 · v0.2

Download the pricing presentation

Скачать презентацию · Ներբեռնել ներկայացումը · Download

The T-Exchange three-part pricing model v0.2 in Russian, English and Armenian: 17 slides, no sign-in required.

Pricing model v0.2

Like Uber: volume, time, surge

Peak · H100, 18:00, 3 h: ×1.50 — $4.05/h instead of $2.70 · $12.25 for 3 h (≈ 4,434 AMD; $8.20 without surge).

  • Distance= tokens

    $ per 1M input / output by model class S/M/L/XL; cache 0.2× input

  • Time= GPU-hours

    Class base per GPU-hour, billed per second, 60 s minimum; H100 — $2.70

  • Surge= demand, k

    0.7–2.0 by utilization and queue; computed every minute, published every 5 min

Sellers are the “drivers”: at the same peak an H100 earns $2.97/h instead of $2.16 (+37.5%), $8.91 for 3 h.

Price and k are shown before confirmation; at k ≥ 1.5 there is a separate checkbox, just like Uber surge.

  1. 1 · Configure

    Pick a deployment template, GPU family and count, memory, CPU/RAM, disk, region, window, transfers and a budget cap.

  2. 2 · Compare & review

    Eligible offers ranked by total estimate, with the reasons others are excluded. The accepted quote pins specification, template and rate version.

  3. 3 · Run & reconcile

    Provision, run, stop and release. The statement reconciles metered usage to accepted rates, with the seller payout, the platform share and any unused return.

Sellers

Data centers and AI hardware owners with spare capacity

  • Goal: maximum utilization and revenue.
  • Idle time is the norm: only 7% of companies run > 85% of GPUs at peak (AIIA, 2024).
  • Share: 80% of base + 60% of surge, never below a 65%-of-base reserve.
  • 8 × H100: utilization 40% → 75%, payout $5,046 → $9,035/mo (+79%).

List hardware Open seller workspace

Buyers

Corporations and products built on the token API

  • Goal: a minimal, predictable AI bill.
  • Transparency: k is shown upfront, 2.0 ceiling, budget cap.
  • 8 × H100 for 12 months: $1.89 per GPU-hour vs $5.97 on AWS Capacity Blocks (−68%).
  • Tokens: reliable access, cache 0.2×, batch 0.5× — within compliance.

Configure a rental Open buyer workspace

Yerevan, Armenia

T-Exchange: Armenia's GPU exchange

Where Europe meets Asia: 11:00–19:00 Yerevan time is the working day in Armenia, Moscow, the UAE, Kazakhstan and Uzbekistan; 18:00–24:00 is Europe's evening (peak hypothesis).

10th of 190 countries for Starting a Business · 3 procedures, 4 days (Doing Business 2020)

  • Jurisdiction. Armenian legal entity under Armenian law; settlement in AMD and USD at the CBA rate (361.93 AMD per $1 as of Oct 9, 2026).
  • Easier here. An LLC via e-register is free and instant with a standard charter; IT pays 1% turnover tax (applicability to GPU rental to be verified).
  • Community. ~36k people work in IT (2024), open Armenian LLMs; nearby are Firebird (6,144 Blackwell GPUs) and the OVIO, GNC-Alfa and Ucom data centers.
  • Yerevan and data. Nodes and GPUs are in Armenia; demand hours follow Asia/Yerevan (UTC+4). The law does not require localization: data residency is our promise.

Launch plan · shadow → narrow bounds → full

Step-by-step launch

  1. 0 · Data and anchor parties

    2–4 wks

    k bounds

    no prices

    ≥ 2 weeks of history, anchor sellers signed

  2. 1 · Shadow mode

    2–4 wks

    k bounds

    shadow

    k is logged, not shown to clients

  3. 2 · Show, no impact

    1–2 wks

    k bounds

    ≤ 1.00

    only discounts below 1.0 and the night window

  4. 3 · Limited live

    2–4 wks

    k bounds

    0.9–1.3

    1–2 classes (RTX 4090, H100); conversion drop ≤ 10%

  5. 4 · Expansion

    4 wks

    k bounds

    0.8–1.6

    all classes, spot with bids, commitments

  6. 5 · Full mode

    next

    k bounds

    0.7–2.0

    k_tok and token tiers, M/M/c calibration

Current stage: 0 — data and anchor parties. Rollback at any time: the "k = 1" kill switch in the admin console.

Compliance gate before first clients from Russia, CIS, Asia

  • KYC/KYB, sanctions screening (OFAC, EU, UK, UN, Armenia), declaration of purpose.
  • Eligibility matrix: Russia and Belarus only after a legal opinion; China, Macau and other D:5 — open-model inference only; A100/H100/H200 require KYB.
  • Armenia makes access, language, payments and company set-up easier — it does not lift any party's sanctions or supplier license conditions, and T-Exchange applies them in full.

Talk to the operator

Workload, capacity or pilot questions go to the administration workspace through this form.

Ask