Skip to main content
NexGPU

NexGPU vs Tencent Cloud

Same silicon. Tencent Cloud's bill is 3.7x to 10.5x ours

Line by line against Tencent Cloud's own published pay-as-you-go rates. Same GPUs, 72% to 90% less. And the parts Tencent Cloud does not sell at any price — H100, H200, RTX 5090 — are priced right here on this page. No ID verification, no month up front, launch the moment you sign up.

Sign up freeView pricing

The verdict: 3.7x to 10.5x more expensive — when it's in stock at all

Run Tencent Cloud's own pay-as-you-go rates line by line: V100 32GB costs 9.5x more, P40 10.5x, T4 3.9x, A10 3.8x, A100 3.7x. And that assumes you can buy it. Every GN10Xp (V100) and every PNV4 (A10) SKU is sold out across Guangzhou, Shanghai and Beijing; the only A100 configuration purchasable anywhere on the platform is a single-GPU GT4.4XLARGE96 in ap-guangzhou-6; and H100, H200 and RTX 4090/5090 do not appear in the catalogue at any price. On NexGPU: A100 80GB — double the VRAM of Tencent's A100 — at $1.088/GPU-hr, H100 SXM at $3.582, H200 at $6.660, V100 32GB at $0.188, across 1,175 verified rentable nodes and 2,498 GPUs in 51 countries. Metered per second, no minimum.

They want you to call sales. We want you to press start.

Export controls keep the H100 and H200 out of mainland Chinese clouds. Tencent Cloud's H800 and A800 do exist in the spec documentation, but with no published price and behind a whitelist — their own docs tell you to contact your pre-sales manager to be granted permission to buy. NexGPU's company and every one of its nodes sit outside that perimeter: the cards are on the price list, the rate is public, and you boot as soon as you sign up.

Not on Tencent Cloud's published price list

  • H100 and H200 — no instance, no price
  • H800 (HCCPNV5) and A800 (HCCPNV4sne) — in the spec docs, but unpriced and gated behind a pre-sales manager
  • A100 — NVLink 40GB only; there is no 80GB SKU
  • RTX 3090 / 4090 / 5090 — no consumer-card instances
  • 6 of their 19 regions have zero purchasable GPU capacity

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.

GPUNexGPUTencent CloudYou save
A100 SXM4GT4.4XLARGE96 · A100 40GB · ¥26.97/hr ÷ 1 · ap-nanjing$1.088$4.013−72%
Tesla V100GN10Xp.2XLARGE40 · V100 32GB · ¥11.98/hr ÷ 1 · ap-guangzhou$0.188$1.783−89%
A10PNV4.7XLARGE116 · A10 · ¥10.69/hr ÷ 1 · ap-guangzhou$0.414$1.591−73%
Tesla T4GN7.2XLARGE32 · T4 · ¥7.81/hr ÷ 1 · ap-chengdu$0.298$1.162−74%
Tesla P40GN8.LARGE56 · P40 · ¥15.09/hr ÷ 1 · ap-guangzhou$0.214$2.246−90%

Tencent figures are published mainland pay-as-you-go rates converted at 1 USD = 6.72 CNY; ours are USD list. 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 metered per second, priced per hour, no minimum, no setup fee. Storage runs $0.414/GB-month median, egress $0.0081/GB median.

    Tencent Cloud · Tencent Cloud's GN10Xp V100 32GB lists at ¥11.98/hour — 9.5x our rate for the same card. Their cheapest A100 40GB is ¥26.97/hour in Nanjing; the 8-GPU node is ¥237.51/hour. GN7, GN8, GN10X and GN6S are tiered, and a short job pays the most expensive T1 tier: GN7.2XLARGE32 is ¥8.68 at T1 versus ¥2.95 at T3, nearly 3x. The only way to the low rate is a monthly commitment — ¥13,748/month for one A100, ¥114,016/month for the 8-GPU node.

  • Fast to start

    NexGPU · Sign up and go. Pick from 2,000+ prebuilt images — PyTorch, vLLM, ComfyUI, Stable Diffusion, AUTOMATIC1111, Whisper ASR, Ubuntu CLI — each booting with drivers and CUDA already in place. SSH, Jupyter and a browser terminal all work out of the box, plus a full REST API and CLI. Pick a GPU, pick an image, launch: three steps to your first epoch.

    Tencent Cloud · First, real-name verification: Tencent's own docs state an account must complete it before purchasing any cloud product. Individuals need a mainland Chinese ID card; Hong Kong, Macau and Taiwan residents need a mainland travel permit plus a WeChat account bound to a mainland bank card; every other nationality is redirected to the international site and its separate USD price list. Enterprise verification adds a bank-transfer confirmation loop that takes days. Then prepay — pay-as-you-go still refuses to launch without a positive balance, and there is no card-on-file model. Then check your per-availability-zone quota and, if it is short, file a ticket naming your regions, zones and instance families and wait for Tencent to evaluate it. Then the machine boots without CUDA: tick "auto-install GPU driver" and wait another 15–20 minutes on first boot. What you finally get is a bare Linux box with a driver on it — no template catalogue, no browser Jupyter.

  • Global footprint

    NexGPU · 51 countries and regions, 1,175 verified rentable nodes, 2,498 GPUs and 75 distinct GPU models live right now — up to 14 GPUs per node and up to 2,152GB of VRAM on a single node. From RTX 3090 to H200, the card you want is generally there when you want it.

    Tencent Cloud · The pricing page does list 19 regions — but GPU stock is a much smaller map. Checked today: Singapore, Jakarta, Johor Bahru, Osaka, São Paulo and Riyadh have zero purchasable GPU of any kind. Singapore lists GN7, GN10X, GN10Xp and BMG5t and every one is sold out; Osaka lists no GPU family at all. Frankfurt is down to a single SKU, one V100 on GN10X. Seoul, Tokyo, Virginia and Bangkok are T4-only. Inside mainland China it is no better: every GN10Xp (V100) and every PNV4 (A10) SKU is sold out across Guangzhou, Shanghai and Beijing, the only orderable A100 anywhere on the platform is a single-GPU GT4.4XLARGE96 in ap-guangzhou-6, and the HCCPNV4h A100 cluster node is invitation-only.

  • Support

    NexGPU · Bilingual Chinese and English, around the clock, straight into Telegram at t.me/nextipkf. No ticket queue, no support tiers, no surcharge. When a training run dies you reach a person who can act, not a queue position.

    Tencent Cloud · The free tier promises only that a ticket is "accepted within 2 hours" — accepted, not resolved, and with no engineer assigned to you. A sub-30-minute first response is a separate purchase: the Commercial plan at ¥500/month promotional, standard formula being the higher of ¥500 or 10% of monthly cloud spend. Fifteen-minute response and a dedicated IM group cost ¥25,000/month; a named technical service manager costs ¥50,000/month. Everything routes through the ticket system, and there is no direct engineer chat on the free tier. Buying the newest silicon is a sales conversation too: H800 and A800 HCC instances are whitelist-gated, with the docs instructing you to "contact your pre-sales manager" for purchase permission.

03 —

Common questions

What to settle before you move.

Does a Tencent Cloud GPU instance require real-name ID verification?

Yes, and it is a hard gate. Tencent's own documentation states an account must complete real-name verification before it can purchase or use cloud products: individuals use a mainland Chinese ID card, Hong Kong/Macau/Taiwan residents need a mainland travel permit plus a WeChat account tied to a mainland bank card, and every other nationality is redirected to the international site and its separate, higher USD price list. Enterprise verification adds a corporate bank-transfer confirmation loop that takes days. Without it the buy button does nothing, and individual accounts cannot get VAT invoices. NexGPU requires no ID verification — sign up, top up, launch, minutes end to end, from $0.193/GPU-hr.

How much does a Tencent Cloud A100 cost per hour, and is there a cheaper option?

Tencent Cloud's A100 is the NVLink 40GB part. The cheapest pay-as-you-go rate is GT4.4XLARGE96 at ¥26.97/hour in Nanjing; Guangzhou, Shanghai and Beijing are ¥28.64/hour; the 8-GPU GT4.41XLARGE948 is ¥237.51/hour. At 1 USD = 6.72 CNY that is about $4.013 per GPU-hour. NexGPU's A100 SXM4 is the 80GB part — double the VRAM — at $1.088/GPU-hr, 72% less, and you are not competing for the one availability zone in Guangzhou where Tencent's A100 is still orderable.

Does Tencent Cloud have H100, and where is H100 cheapest to rent?

It does not. Neither cloud.tencent.com nor the international site offers an H100 instance, and no H100 price is published anywhere — same for H200. H800 and A800 do appear in the HCC cluster spec document, but with no published price at all and whitelist-only access: the doc tells you to "contact your pre-sales manager" for purchase permission. You cannot even get a number without a sales call. NexGPU lists H100 SXM 80GB at $3.582/GPU-hr and H200 141GB at $6.660/GPU-hr, on the page, launchable the moment you sign up, metered per second.

What do I do when Tencent Cloud GPUs are sold out?

That is the normal state, not an outage. Of the 19 regions, six — Singapore, Jakarta, Johor Bahru, Osaka, São Paulo and Riyadh — have zero purchasable GPU today. Frankfurt is down to one SKU, a single V100 on GN10X. Seoul, Tokyo, Virginia and Bangkok are T4-only. In Guangzhou, Shanghai and Beijing every GN10Xp (V100) and every PNV4 (A10) SKU is sold out, and the only orderable A100 on the whole platform is a single-GPU GT4.4XLARGE96 in ap-guangzhou-6. Rather than camp on the refresh button, use a market where you do not have to: NexGPU has 1,175 verified rentable nodes, 2,498 GPUs and 75 GPU models across 51 countries and regions.

Tencent Cloud GPU is too expensive — what's the alternative?

Compare Tencent Cloud's own published pay-as-you-go rates line by line, converted at 1 USD = 6.72 CNY: T4 at ¥7.81/hour (about $1.162) versus our $0.298; V100 32GB at ¥11.98 (about $1.783) versus our $0.188; P40 at ¥15.09 (about $2.246) versus our $0.214; A10 at ¥10.69 (about $1.591) versus our $0.414; A100 at ¥26.97 (about $4.013) versus our $1.088 — and ours is the 80GB part. That is 72% to 90% less, metered per second and priced per hour, with no minimum, no setup fee and no ID verification.

Are Tencent Cloud spot instances cheap, and will they get reclaimed?

Spot is priced at 3%–20% of the equivalent on-demand list rate, and the trade-off is spelled out: reclaimed with only two minutes' notice, cannot be resized, cannot be reinstalled, cannot be converted to a subscription, and it carries its own separate quota. The international site publishes spot GN7 with one T4 at $0.204/hour. Compare that directly: NexGPU's RTX 3090 has 24GB of VRAM, 8GB more than a T4, at $0.193/GPU-hr — cheaper than Tencent's spot T4, and it is an ordinary on-demand instance nobody takes back two minutes from now.

Tencent Cloud GPU: is pay-as-you-go or a monthly subscription the better deal?

On GN7, GN8, GN10X and GN6S, Tencent's pay-as-you-go rate is tiered, and short jobs land on the most expensive T1 tier: GN7.2XLARGE32 is ¥8.68/hour at T1 against ¥2.95 at T3, nearly 3x. The cheap rate lives behind a monthly commitment — ¥13,748/month for a single A100 GT4.4XLARGE96, ¥114,016/month for the 8-GPU node, ¥1,368,192 list for a year. So the cheapest way to touch an A100 on Tencent Cloud is to put down a full month in advance. That same ¥13,748 (about $2,046) buys 1,880 hours of an 80GB A100 on NexGPU — 78 days straight, double the VRAM, and stoppable at any second, because billing stops when you do.

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.