Graphics cards and Macs
RTX 4090 24 GB: $5,900 in July 2026; on 30 September 2026 the seller in the source no longer has it. RTX 5090 32 GB: $7,700–$7,900; in January 2026 about $3,600, in March 2025 from $4,200. RTX PRO 6000 96 GB: $13,000 at DNS on 2 February 2026, with no newer price in the sources. A100 40 GB: $6,600; A100 80 GB: $13,000. H100 SXM5 80 GB: $28,000. MacBook Pro 14 M5 Max 128 GB 2 TB: $8,100. Mac Studio M3 Ultra 96 GB: $7,800; M4 Max 128 GB: $11,000; M3 Ultra 512 GB 4 TB: $38,000. A workstation platform costs $3,600–$4,200, and a server platform $3,600–$4,800 by our estimate.
Ready configurations
One computer for 1–3 users: $7,800–$13,000 with the platform, about $300–$800 a month in full cost by our calculation. A team server with two 96 GB cards: about $30,000, about $2,100 a month with hosting and an engineer by our calculation. A server with four H100 cards: from $120,000 with the platform, about $6,100 a month by our calculation. A server for 1,000 simultaneous users, by an industry estimate: about $660,000.
Electricity, hosting, people
The single-rate household tariff in Moscow from 1 January to 30 June 2026 is about $0.10 per kWh; companies pay by contract. One RTX 5090 running around the clock: about $39 a month. Data center hosting: from $90 a month for a server of up to 300 W, plus $24 for each extra 100 W. A DevOps engineer: a median of $2,700–$3,000, up to $3,800 for a middle engineer in Moscow, $4,400–$5,900 for a senior; a team server takes 10–20% of one salary by our estimate. Maintenance: 10% of the hardware price per year by a rule of thumb.
GPU rental in Russia
According to the GPU Index aggregator on 30 September 2026, hourly rates with hourly billing: RTX 4090 from about $0.50, RTX 5090 from about $0.80, RTX PRO 6000 from $2, H100 from $3 to $7, H200 from $5; rentals of a month or more are cheaper: RTX 4090 from about $0.40, RTX PRO 6000 from $2. Per month: RTX 4090 from $300, RTX PRO 6000 about $1,800, A100 about $1,800, H100 from $1,700. Discounts of 6–20% for rentals of three months or more at a provider with a site in the Samara region.
Cloud APIs, US dollars per million tokens
Low-cost: GPT-6 Luna, about $0.10 input and about $0.50 output; Gemini 3.1 Flash-Lite, about $0.25 and $2; the light model of a large Russian bank, about $0.80. Mid-range: Gemini 3.8 Flash, about $0.75 and $4; GPT-5.4 mini, about $0.75 and $5; Haiku 4.5, $1 and $5; Sonnet 5.5 and GPT-6 Sol, $2 and $10; the Russian bank’s Pro model, $6, and its Max model, $8; Gemini 3.1 Pro, $2 and $12. Flagships: the Pro model of a Russian search company, $14; Opus 5.5, $4 and $20; GPT-5.5, $5 and $31; Fable 5.1 and GPT-6 Astra, $10 and $51. At 60 million tokens a month (45 million input and 15 million output): from $12 to $1,200.
Rollout with us
A pilot AI agent on a local model: 14 days, from $1,499. An AI department of 3–5 agents: 4–6 weeks, from $4,999. A large company: 2–3 months, from $18,000. Support from $480 a month, an AI director from $999 a month. You buy the hardware to our specification, and we take no commission from sellers.