Skip to main content
NexGPU

NexGPU vs Alibaba Cloud ECS GPU

One hour of Alibaba Cloud's A100 buys you more than four hours on NexGPU

The only A100 with a published price on Alibaba Cloud's list is the Beijing bare-metal ecs.ebmgn7.26xlarge: $39.1067/hr ÷ 8 = $4.888 per GPU-hour — and their price list states outright that it excludes system disk and bandwidth. NexGPU's A100 SXM4 80GB is $1.088 per GPU-hour, metered per second, priced per hour, with no minimum, no setup fee, and nothing to prepay.

Sign up freeView pricing

The verdict up front: on demand vs on demand, we're 35% to 77% cheaper

Same terms on both sides — pay as you go, no contract, no prepayment, shut it down whenever. Alibaba Cloud's A100 40GB in Beijing is $4.888/GPU-hr. Our A100 SXM4 80GB is $1.088: double the VRAM for under a quarter of the price. Flagship class is the same story — the cheapest published rate for Alibaba's gn8v (96GB HBM3, 900GB/s NVLink) is $5.5492/GPU-hr in Ulanqab, against $3.582 for our H100 SXM 80GB. And none of that counts what sits outside their price figure: the ¥100 minimum account balance, the per-region per-family GPU quota, the VPC and system disk you build yourself, the Tesla driver you install yourself. On our side it's sign up, pick a GPU, pick a template, boot — 2,000+ prebuilt images, ready to run.

The advanced parts aren't on their price list. They're on ours.

Export controls keep the H100 and H200 class out of mainland Chinese clouds, leaving China-specific variants — and Alibaba Cloud will not even name the chip in its top AI SKU: gn8v is described only as 96GB HBM3, 4TB/s of bandwidth and 900GB/s NVLink, with no model given anywhere. NexGPU's company and every one of its nodes sit outside that perimeter. H100 and H200 are ordinary line items on our price list; sign up and boot one.

Not on Alibaba Cloud's published price list

  • H100 SXM and H200 — not offered on ECS, no published rate anywhere
  • A100 80GB — gn7e / ebmgn7ex / sccgn7ex appear in the spec docs, but carry no published price in any of the 31 regions
  • H800, A800, H20 — no publicly named or priced instance family
  • gn8v, their top part, lists memory and bandwidth but never the GPU model
  • RTX 4090 / 5090 — no consumer-card instances

Rentable on NexGPU right now

  • H100 SXM 80GB · $3.582 per GPU-hour
  • H200 141GB · $6.660 per GPU-hour
  • A100 SXM4 80GB · $1.088 per GPU-hour
  • RTX 5090 $0.723, RTX 4090 $0.540 per GPU-hour
  • 51 countries and regions, 1,175 verified rentable nodes, 2,498 GPUs online

01 —

Same GPU, rate by rate

Every competitor figure is a link to the page it was read from.

GPUNexGPUAlibaba CloudYou save
A100 SXM4ecs.ebmgn7.26xlarge · A100 40GB · $39.1067/hr ÷ 8 · cn-beijing$1.088$4.888−77%

Competitor rates are each vendor's own published on-demand price — click a figure to check the source. Whole-node prices are divided by GPU count.

02 —

Four dimensions, head to head

Price, time to first boot, global footprint, support.

  • Price

    NexGPU · RTX 3090 $0.193, RTX 4090 $0.540, RTX 5090 $0.723, A100 SXM4 80GB $1.088, H100 SXM 80GB $3.582, H200 141GB $6.660 — all per GPU-hour. Metered per second, priced per hour, no minimum and no setup fee. The listed price is the price you pay.

    Alibaba Cloud · Alibaba Cloud on demand: A100 40GB bare metal at $4.888/GPU-hr in Beijing; the flagship gn8v at $5.5492 in Ulanqab, $5.8412 in Beijing, $7.5825 in Frankfurt. Their price list says it plainly — the figures exclude system disks and bandwidth, so every row needs a cloud disk added on top (Hong Kong ESSD PL1 is $0.200592/GiB per 30-day period) plus public bandwidth.

  • Speed to first job

    NexGPU · 2,000+ prebuilt images — PyTorch, vLLM, ComfyUI, Stable Diffusion, AUTOMATIC1111, Whisper ASR, Ubuntu CLI — pick one and the environment is already running, drivers included. Get in by SSH, Jupyter or web terminal, and wire it into your pipeline with the REST API or CLI.

    Alibaba Cloud · Before your first GPU boots you clear the whole chain: real-name verification (individual verification unlocks only part of the catalogue; the full set needs a business licence), a cash-plus-voucher balance of at least ¥100, a GPU card quota scoped per region and per instance family and per billing mode (short? file a Quota Center request or a ticket), a VPC, vSwitch and security group, a system disk — and then you install the Tesla or GRID driver yourself. The region is locked once the instance is created.

  • Global footprint

    NexGPU · 51 countries and regions, 1,175 verified rentable nodes, 2,498 GPUs, 75 GPU models, up to 14 GPUs in a single node and up to 2,152GB of VRAM on one node. Run where you want, on the silicon you actually wanted.

    Alibaba Cloud · Alibaba's public price list covers 31 regions, but GPUs are not spread across them. Seven regions list no GPU SKU at all — Manila, Hohhot, São Paulo, Mexico, London, Paris, Dubai. Johor lists exactly one. Seoul lists eight rows. Silicon Valley lists 17, and only L20, T4 and the P100-era gn5. The HBM-class gn8v appears in mainland China and Frankfurt only; Hong Kong and Singapore can't buy an HBM training card at all.

  • Support

    NexGPU · Bilingual Chinese and English, around the clock, straight through to Telegram at t.me/nextipkf. Ask and get an answer — no ticket queue, no support tier to purchase.

    Alibaba Cloud · Ticket-based. Everything above the free Basic tier is a paid subscription: Developer at $19.99/month, Business from $100/month, Enterprise from $8,000/month — and the last two are priced off your cloud spend. Phone and TAM channels live in the upper tiers; for a small team the default experience is the ticket queue.

03 —

Common questions

What to settle before you move.

Alibaba Cloud GPUs are too expensive — what's the alternative?

Start with the numbers. Alibaba's on-demand A100 40GB bare metal is $4.888/GPU-hr (ecs.ebmgn7.26xlarge, Beijing, $39.1067/hr ÷ 8), and the cheapest published rate for the flagship gn8v is $5.5492/GPU-hr — neither of which includes a system disk or public bandwidth. NexGPU's A100 SXM4 80GB is $1.088/GPU-hr with twice the VRAM; H100 SXM 80GB is $3.582, H200 141GB is $6.660, RTX 4090 is $0.540. Same terms — no contract, no prepayment — and 35% to 77% less.

How much is an Alibaba Cloud A100 per hour?

The only A100 with a published price on Alibaba's list is the 40GB part: the Beijing bare-metal ecs.ebmgn7.26xlarge at $39.1067/hr for 8 GPUs, or $4.888 per GPU-hour on demand, excluding system disk and bandwidth. The 80GB gn7e / ebmgn7ex families appear in Alibaba's spec documentation but carry no price in any of the 31 published regions — there is no hourly rate to look up for an 80GB A100 at all. NexGPU lists the A100 SXM4 80GB at $1.088/GPU-hr, metered per second.

Does an Alibaba Cloud GPU instance require real-name verification and a prepaid balance?

Both. Alibaba's own documentation is explicit: verification must be completed before you buy anything, and without it you cannot purchase, renew or use most cloud services. Individual verification unlocks only part of the catalogue — the full set needs enterprise verification with a business licence. There is also a balance gate before a pay-as-you-go ECS instance can be created: cash balance plus vouchers must total at least ¥100. NexGPU has no ¥100 prefund gate, no setup fee and no minimum spend, and its 1,175 verified nodes are bookable inventory you pick straight off the list.

Does Alibaba Cloud offer H100 or H200?

No instance family on Alibaba Cloud's public ECS price list is named H100, H200, H800, A800 or H20. The closest part is gn8v, documented only as 96GB of HBM3 per GPU, up to 4TB/s of memory bandwidth and 900GB/s NVLink between cards — the model name is never published, so the price list can't tell you which silicon you're renting. NexGPU spells it out: H100 SXM 80GB at $3.582/GPU-hr, H200 141GB at $6.660/GPU-hr. Model, VRAM and rate, all on the page.

Alibaba Cloud spot instances are cheap — why not just use those?

Because it can be taken back at any moment. The spot floor is 10% of the pay-as-you-go rate, but the only guarantee Alibaba gives is that the instance won't be reclaimed within one hour of creation. After that window it can be interrupted at any time, and a straight release takes the vCPUs, GPU, memory and disks with it. Spot pricing also excludes cloud disk and public bandwidth. A training run killed at hour six saves you nothing. NexGPU is per-second on-demand: an A100 at $1.088 runs as long as you want it to, no gamble attached.

My Alibaba Cloud GPU quota won't let me launch 8 cards — what now?

Alibaba's GPU card quota is scoped per region, per instance family and per billing mode, with separate quota IDs for on-demand and spot (q_ecs_enterprise_postpay_c, q_ecs_enterprise_spot_c). New accounts don't get 8-GPU headroom by default — you file a Quota Center request or a support ticket, and a quota granted in Beijing doesn't travel to Singapore. On top of that, Alibaba's docs treat insufficient regional inventory as a normal condition, with the recommended remedy being to change region, change spec, or wait — and the region can't be changed after the instance is created. NexGPU has none of that layer: pick from 1,175 verified rentable nodes, up to 14 GPUs and 2,152GB of VRAM in one node, and boot what you see.

Where can I rent an H100 cheapest?

NexGPU lists the H100 SXM 80GB at $3.582/GPU-hr, metered per second, priced per hour, with no minimum and no setup fee. The cheapest comparable on-demand rate on Alibaba Cloud is $5.5492/GPU-hr (gn8v, Ulanqab, $44.3934/hr ÷ 8), rising to $5.8412 in Beijing and $7.5825 in Frankfurt — and none of those include system disk or bandwidth. That's over 35% less on our side. Want cheaper still? RTX 4090 at $0.540 and RTX 5090 at $0.723, billed the same way. Need more VRAM? H200 141GB at $6.660. Boot one at console.nexgpu.net.

Weighing up the others too?

Related

Start building on NexGPU

Enterprise R&D team or solo developer — either way, your first job can be running in minutes.

Sign up freeTalk to sales

Sign up to browse live network pricing. No payment method required.