Dallas, Texas, United StatesFounded 20250 GPUs available
About EcoHash
EcoHash is the high-performance computing and AI inference division of Cango Inc. (NYSE: CANG), the Dallas-based Bitcoin miner that has been converting its owned power and datacenter footprint into GPU capacity since 2025. Founded in 2025 and trading since April 2026, EcoHash runs a single GPU done thoroughly: the NVIDIA RTX PRO 6000 Blackwell Server Edition with 96 GB of GDDR7, sold as dedicated instances of one, two, four or eight GPUs, each GPU paired with 16 vCPUs, 64 GB of RAM and 400 GiB of local scratch. Instances are billed per second from a prepaid credit wallet with no subscription, and a launch-time duration acts as a commitment: when it elapses the instance terminates and unused credit is refunded. Alongside bare GPUs the company serves speech, image, video, embedding and LLM models through one OpenAI-compatible API, and offers dedicated inference endpoints with a monthly 99% success-rate SLA. Capacity is United States only today: the first commercial site is Cango's 50 MW Georgia facility, converted to containerised GPU modules in July 2026, with further nodes in Texas and on the West Coast. The positioning is inference, LoRA fine-tuning and 3D rendering rather than large-scale training, which is exactly what a 96 GB GDDR7 card is good at.
Why choose EcoHash
Parent company
Cango Inc. (NYSE: CANG)
EcoHash is the HPC and AI inference subsidiary of a listed Bitcoin miner with its own power and sites, which is a different balance sheet from a venture-funded neocloud renting colocation space.
The GPU
RTX PRO 6000 Blackwell Server Edition
One card, 96 GB of GDDR7 with ECC, in 1, 2, 4 or 8 GPU instances. Every GPU comes with 16 vCPUs, 64 GB of RAM and 400 GiB of local scratch, and the price scales linearly with the count.
Billing
Per second, prepaid credits
No subscription and no invoices for self-serve accounts. You set an estimated duration at launch; the instance stops when it runs out and unused credit is returned, so a forgotten box cannot run up a bill.
Inference API
OpenAI-compatible
Whisper and Qwen3 speech-to-text, Kokoro and Qwen3 text-to-speech, FLUX.2 and Qwen-Image image models, LTX and Wan video, plus DeepSeek, Llama, GLM and Kimi LLMs behind one endpoint at api.ecohash.com.
Where it runs
United States only
The commercial site is Cango's 50 MW Georgia facility, converted to containerised GPU modules in July 2026, with test nodes in Texas and on the West Coast. No EU or APAC region yet.
Service level
99% monthly, dedicated endpoints
The SLA covers dedicated inference endpoints only. The serverless API is best effort and GPU instances carry no published uptime commitment, which is worth knowing before you put production on them.
Where workloads run — primary datacenters and active deployment zones.
🇺🇸
GeorgiaPrimary
United States· us-georgia
🇺🇸
Texas
United States· us-texas
🇺🇸
West Coast
United States· us-west
Available GPUs
No GPU pricing data available yet. Check back soon.
Frequently asked questions
Common questions about working with EcoHash.
Which GPUs does EcoHash offer?
One model: the NVIDIA RTX PRO 6000 Blackwell Server Edition with 96 GB of GDDR7 memory. It is sold as dedicated instances of 1, 2, 4 or 8 GPUs. Each GPU comes with 16 vCPUs, 64 GB of system RAM and 400 GiB of local scratch storage, so an 8-GPU instance has 768 GB of VRAM, 128 vCPUs and 512 GB of RAM.
How does EcoHash billing work?
You top up a prepaid credit wallet and GPU instances are billed per second of running time. There is no subscription and no monthly invoice for self-serve accounts. When you launch an instance you give an estimated duration; credit for that duration is held up front, the instance terminates automatically when it elapses, and any unused credit is refunded. Enterprise customers can move to NET-30 invoicing by wire or ACH. New accounts receive a small welcome credit.
Where are EcoHash GPUs located?
In the United States only. EcoHash is the compute arm of Cango Inc., and its first commercial site is Cango's 50 MW facility in Georgia, which was converted from Bitcoin mining to containerised GPU modules in July 2026. Cango has also reported test nodes running in Texas and on the West Coast. There is no European or Asia-Pacific region at present.
Does EcoHash have an inference API?
Yes. api.ecohash.com exposes standard OpenAI-compatible endpoints, so existing OpenAI SDK code works by changing the base URL and key. The catalogue covers speech-to-text (Whisper Large V3 and Turbo, Qwen3-ASR), text-to-speech (Kokoro, Qwen3-TTS, Chatterbox), image generation (FLUX.2 Klein, Qwen-Image, Z-Image-Turbo), video (LTX, Wan, Seedance), embeddings and rerankers, and LLMs including DeepSeek-V4, Llama 3.1 8B, GLM, Kimi and Qwen3. Models are priced per million tokens or per second of video.
What is EcoHash's relationship with Cango?
EcoHash Platform Technology LLC was founded in 2025 as the dedicated high-performance computing and AI inference division of Cango Inc., which trades on the NYSE as CANG. Cango is a Dallas-headquartered Bitcoin miner that entered mining in November 2024 and has since been redirecting part of its owned power and datacenter footprint towards GPU compute. EcoHash began commercial operations in April 2026.
Does EcoHash offer an SLA?
A monthly service level of at least 99% of eligible requests completing successfully applies to dedicated inference endpoints. The serverless inference API is offered on a best-effort basis with no SLA, and no uptime commitment is published for GPU instances. No SOC 2, ISO 27001 or HIPAA attestation is stated on the EcoHash website.
What is EcoHash best suited for?
Workloads that want a lot of memory per dollar rather than HBM bandwidth: dedicated inference serving, LoRA and other parameter-efficient fine-tuning, 3D rendering and visual computing. The RTX PRO 6000 uses GDDR7 rather than HBM and connects over PCIe, so multi-node training of large models is not what this fleet is built for.
Ready to deploy on EcoHash?
Dedicated RTX PRO 6000 Blackwell GPUs and an OpenAI-compatible inference API, billed per second.
