Nebius Startup Deals & Discounts

Nebius runs an application-based startup program that bundles introductory credits, engineering time and investor introductions. Our listing says up to $100k in credits for new customers or free-plan accounts. Nebius's own startup page is narrower than that, and anyone planning around the $100k should read its terms first:

• Funding: at least $5M raised, from one of the vendor's approved venture partners.
• Age and product: incorporated fewer than 5 years ago, building or scaling an AI-enabled product or model.
• Excluded: consultancies and outsourced dev shops, resellers, cloud providers, crypto-only companies and public companies.
• Speed: no response time is promised, and applicants can land on a waitlist.

The page gives no duration and no tier amounts, so the $100k ceiling comes from our listing alone.

Against other programs in our catalog. Together AI lists up to $50,000 in credits, DigitalOcean for AI lists up to $100,000, and Google Cloud runs from $2,000 up to $350,000 depending on funding history. Nebius ties with DigitalOcean for AI on the $100k ceiling. The entry bar is what separates them: Nebius's page asks for a venture-backed company with $5M raised, while Google Cloud's low tier starts at $2,000 for startups that have raised nothing.

Fits: a venture-backed AI lab fine-tuning or pre-training its own models. Large H100 or H200 runs are the case the credit was priced for, and Soperator gives a Slurm-style scheduler without a licence bill.

Fits: an inference-heavy product with steady traffic. Reserved capacity brings the rate down by up to 35%, and egress traffic costs nothing, which matters when you serve many tokens to many users.

Skip: a pre-seed team calling a hosted model API. The program's entry bar is $5M raised, and an app that only sends prompts to someone else's model has no use for GPU hours. Hosted-credit programs such as OpenAI's or Together AI's fit that team better.

Skip: an agency or consultancy building AI for clients. Outsourced development firms are on the excluded list.

Skip: a founder who wants a predictable start date. With no promised response time and a waitlist, a launch that depends on the credits arriving by a fixed week is a risk.

Nebius's price page lists these on-demand rates per GPU-hour: B300 $9.50, B200 $8.50, H200 $5.40, H100 $4.50, RTX PRO 6000 $1.80, and L40S from $1.55. Preemptible H100 and H200 capacity starts at $0.79. Committing to large clusters for months takes up to 35% off on-demand.

What up to $100,000 of credit covers at those rates:

Hardware	Rate per GPU-hour	GPU-hours bought
H100, on-demand	$4.50	22,222
H200, on-demand	$5.40	18,518
B200, on-demand	$8.50	11,764
L40S, Intel host	$1.55	64,516
H100, preemptible	$0.79	126,582

An eight-GPU H100 server costs $36 per hour, so the full credit runs one for 2,777 hours, about 115 days of nonstop use. A day of that server costs $864 at list. Preemptible prices can change as often as every 15 minutes, so the last row is a floor, not a plan.