News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise Expansion
EveryNeocloud
Firmus

Firmus

Neocloud operator building immersion-cooled AI Factories across Asia-Pacific

Firmus, also known as Sustainable Metal Cloud, designs and runs modular, liquid-cooled AI compute sites it calls AI Factories. The company says it began designing systems for over 1,000 GPUs in 2019 and now has more than 900 MW of contracted capacity, with OpenAI as an anchor customer and expansion into Malaysia.

Share on WhatsApp

  • Claims over 900 MW of contracted capacity, with OpenAI as anchor customer
  • Runs SMC by Firmus in Singapore for sovereign AI in South-East Asia
  • Converted a former basement car park into a 3 MW GPU AI factory
  • Publishes live energy, thermal and compute benchmark data
  • Named a top-three global GPU cloud by SemiAnalysis

What it is

Firmus is a developer and operator of AI computing infrastructure, trading in part as Sustainable Metal Cloud. It builds what it calls AI Factories: modular, vertically integrated compute sites designed for dense GPU work with lower power, water and space needs than conventional halls. The company sells access to GPU capacity and also delivers sovereign AI projects in the countries it serves. Its published hardware listings include Nvidia Blackwell GB300 machines in a 72-GPU configuration and 8-way HGX nodes with 80GB each, connected over InfiniBand with BlueField DPUs, plus S3-compatible WEKA storage. The company claims more than 900 MW of contracted capacity.

How it came about

Firmus says its work began with system design for clusters above 1,000 GPUs in 2019. Early milestones it lists include finishing HyperCube research and development, completing an immersion-cooled data hall running on renewable energy, and delivering an integrated compute platform. SMC by Firmus then went live in Singapore for sovereign AI services in South-East Asia. Later steps include a grid-integration project with Eaton in Tasmania, an award from DCD Asia Pacific, a top-three global GPU cloud ranking from SemiAnalysis, scaling deployments across APAC, and the announcement of Project Southgate, a Tasmanian AI campus.

The compute

The company describes its HyperCube as the unit of its platform: a modular, multi-petascale block with thermal optimisation and high availability. Cooling is liquid-based, using immersion, and the firm says orchestration software is aware of the silicon while buildings are designed around the hardware. Published node specs include 72-way Blackwell GB300 systems at 800 Gbps with XDR, and 8x 80GB HGX nodes with NVSwitch, full rail InfiniBand and BlueField DPUs rated at 3,200 Tbps. Storage is WEKA, S3-compatible. The company says it publishes live energy, thermal and benchmark data.

Where the GPUs are

Deployments named include a retrofit inside an existing building, a 3 MW GPU factory converted from a basement car park, and Australia's flagship AI infrastructure initiative. SMC by Firmus runs in Singapore for South-East Asian sovereign AI. A grid-integration project with Eaton went live in Tasmania, and Project Southgate, a Tasmanian AI campus, has been announced. The company says it has expanded into Malaysia and signed a Trans-Pacific connectivity agreement with SUBCO. It claims over 900 MW of contracted capacity.

Customers and deals

OpenAI is named as an anchor customer alongside a claim of more than 900 MW contracted. Meta and Firmus have announced strategic AI infrastructure agreements across Asia-Pacific. The company says it supports Thinking Machines Lab's frontier AI development. Its infrastructure also powered work on SEA-LION, described as the first open-source large language model family aimed at Southeast Asian languages and contexts, where 100 candidate models were produced and evaluated.

Money and ownership

Firmus lists recognition rather than financial figures: SemiAnalysis ClusterMAX rankings in 2024 and 2025, MLPerf certified benchmark results, Cleantech Group's APAC Top 25 for 2025, and DCD Asia Pacific Data Center Project of the Year 2024.

Its place in AI

The company positions itself around sovereign AI infrastructure for Asia-Pacific, with a Singapore service aimed at the region and Australian projects including a Tasmanian campus. It claims a top-three place in SemiAnalysis's global GPU cloud ranking and industry awards for infrastructure deployment. Its platform supported training and evaluation work behind SEA-LION, an open-source model family for Southeast Asian languages.

Questions and criticism

The company says its designs cut energy and water use through liquid cooling and publish live consumption data. It also states that physical layouts are built to accommodate future GPU generations without retrofit.

Questions people ask about Firmus

What does Firmus actually sell?

The company designs and operates modular AI Factories and rents out GPU compute. Published listings include 72-way Nvidia Blackwell GB300 systems and 8x 80GB HGX nodes with InfiniBand networking and WEKA storage. It also builds sovereign AI projects for the countries it serves.

Where does it operate?

Named sites include Singapore, where SMC by Firmus runs sovereign AI services for South-East Asia, plus Tasmania, where a grid project with Eaton went live and a campus called Project Southgate is planned. The company says it has also expanded into Malaysia.

How is its approach to cooling different?

Firmus uses liquid immersion cooling inside modular HyperCube blocks. It says this lowers power, water and space needs compared with conventional data halls, and that it publishes real-time energy and thermal readings so outsiders can check the claims.

Who are its customers?

OpenAI is described as an anchor customer. Meta has signed strategic AI infrastructure agreements with the company across Asia-Pacific, and Firmus says it supplies compute for Thinking Machines Lab's frontier AI work.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Firmus yet, so the facts on this page come from the sources listed above: firmus.co/about, firmus.co/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    Photo credits on this page

    The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

    The weekly digest

    Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.