NVIDIA’s DSX reference design and operating system are engineered to raise the revenue per gigawatt an AI factory can produce by lifting tokens per megawatt. Paired with NVIDIA’s July 1 revenue-sharing model, DSX forms half of a two-part business model: grow the revenue a gigawatt generates, then collect a recurring cut of it. Sharon AI and Firmus are building the first DSX-aligned factories under the arrangement.
What Is Covered in This Article:
- How NVIDIA’s DSX reference design and operating system raise revenue per gigawatt by lifting token efficiency
- The DSX stack: reference designs, DSX Sim digital twin, the open source DSX OS, and the MaxLPS power-tuning layer
- How DSX and the revenue-sharing model form two halves of one business model — grow the revenue a gigawatt produces, then take a recurring cut
- Why AI factory economics approaching $100 billion per gigawatt make tokens-per-megawatt efficiency the decisive variable
- The bull case of a compounding efficiency-plus-financing flywheel against the bear case of product-market mismatch and an unresolved split of the upside.
The News: On July 1, 2026, NVIDIA introduced a business model that lets AI cloud operators deploy large-scale, multi-tenant AI factories through a combination of revenue sharing and credit support, with the resulting facilities built as DSX AI factories. Participating clouds procure NVIDIA infrastructure and sell NVIDIA-powered services; NVIDIA earns its standard product revenue plus a share of the cloud revenue generated on the supported capacity. Sharon AI, deploying up to 40,000 Grace Blackwell GB300 GPUs across new capacity in Australia, and Firmus, building a 360-megawatt DSX-aligned AI factory campus in Batam, Indonesia, targeting up to 170,000 GPUs, are among the first partners.
NVIDIA DSX Promises More Revenue per Gigawatt. Who Actually Captures It?
Analyst Take: NVIDIA’s revenue-sharing model answered how the company gets paid twice on the same silicon. The DSX reference design and operating system is NVIDIA’s mechanism for raising the revenue per gigawatt that a fixed block of power can produce, and it is the half of the business model that determines whether the recurring cut NVIDIA now takes on cloud revenue is a rich stream or a rounding error. Introduced as a platform at GTC Taipei and now the design standard for the AI factories that Sharon AI and Firmus are building under the July 1 financing model, DSX reframes the gigawatt from a power budget into a revenue function. NVIDIA is now selling the operating system that decides how much sellable output a gigawatt of compute yields, then taking a share of that output.
The clearest signal of how NVIDIA wants the market to read this sits in the byline. The announcement carries the name of Colette Kress, NVIDIA’s EVP and CFO, in her first company blog post on the topic. A CFO does not headline a product blog unless the company intends the announcement to be read as a financial-reporting event rather than a catalog addition — the debut of a recurring, usage-linked earnings stream that will surface in the financial model, not merely a new offering in the lineup. That authorship is the tell that revenue per gigawatt is the point of the announcement, not a byproduct of it, and it is consistent with the message Kress has carried on recent earnings calls.
DSX Reframes the Gigawatt From a Power Budget to a Revenue Function
DSX is not a single product but a stack: reference designs for physical blueprint, DSX Sim as an Omniverse digital twin that validates a factory before a rack lands, DSX OS for the open source operations software that runs it, and the MaxLPS efficiency layer that tunes power at the rack. The layer that matters most for revenue per gigawatt is MaxLPS, which combines liquid cooling with software-driven in-rack power tuning to run up to 40% more GPUs at their most efficient operating point inside a fixed power budget, on NVIDIA’s figures. DSX Flex extends the same control to the utility grid, with a multi-megawatt pilot alongside Emerald AI and Silicon Valley Power that adjusts factory consumption programmatically in response to grid signals. DSX reframes AI factory competition from cores per dollar to tokens per megawatt. That reframing is the entire revenue-per-gigawatt case: if power is the binding constraint, the operator that extracts more tokens per watt earns more revenue, and DSX is the tool NVIDIA is selling to do it.
The Business Model Has Two Halves: Grow Revenue per Gigawatt, Then Share It
The connection to the financing model is where the strategy becomes coherent. Under the July 1 revenue-sharing and credit-support model, NVIDIA earns its standard hardware margin plus a recurring, usage-linked share of the cloud income the supported capacity generates. That recurring cut is worth precisely as much as the capacity is utilized and efficient, which is exactly what DSX is engineered to maximize. DSX grows the revenue a gigawatt produces; the revenue-share model lets NVIDIA collect a slice of the larger number. Financing places NVIDIA silicon inside capital-constrained AI factories, and DSX ensures the resulting factories run at the token-per-watt efficiency that makes both the operator’s business and NVIDIA’s royalty viable.
Sharon AI and Firmus are the template. Firmus is explicitly building a DSX-aligned campus in Batam, co-designing its HyperCube architecture to NVIDIA’s DSX blueprints, and states the integration is meant to improve tokens per watt and reduce cost per token for its customers. That is an operator adopting NVIDIA’s revenue-per-gigawatt engine and NVIDIA’s financing in the same deal.
The $100 Billion Gigawatt Makes Efficiency the Whole Game
The economics explain why NVIDIA is willing to give the operating software away. Huang has said a 1-gigawatt AI factory that once cost $20–30 billion now runs $50–60 billion and will soon reach $80–100 billion per gigawatt, with 100 gigawatts expected online before the end of the decade. When a gigawatt costs as much as $100 billion to build, an idle or inefficient watt is not a technical footnote but destroyed capital. In that environment, revenue per gigawatt is the metric that decides whether a $100 billion facility earns its cost of capital, and tokens per megawatt is the lever. DSX is NVIDIA’s claim on that lever. The implication is that the reference design and operating system are not a giveaway but the highest-leverage point in the entire buildout: the software that governs whether the most expensive industrial assets ever built run at a profit.
Who Captures the Revenue per Gigawatt Is the Open Question
The bull case is a flywheel: DSX raises revenue per gigawatt, the financing model spreads NVIDIA silicon into every regional and sovereign operator, and the recurring royalty compounds. The bear case is that the operator builds their own software system and applies further margin pressure on system vendors. NVIDIA has partly funded its own demand through investments in OpenAI, CoreWeave, and Nebius, and the revenue-share model extends that pattern by other means. The competitive counter to this strategy is concentrated among the buyers that have the most to gain from it. The hyperscalers and frontier labs on NVIDIA’s own Vera CPU adopter list are precisely the organizations capable of building alternatives with custom CPUs, open networking, disaggregated inference, or their own operations software. The question this headline poses may gain clarity over the next several quarters: DSX will raise revenue per gigawatt, but the split between the operator that owns the power and the vendor that owns the operating system is the number NVIDIA has not yet disclosed.
What to Watch:
- Independent confirmation of the 40% revenue claims, which are currently NVIDIA figures.
- Whether the open source operations software attracts genuine multi-vendor contributions or remains optimized primarily for NVIDIA silicon.
- Whether the grid-responsive pilots with Emerald AI and Silicon Valley Power expand into utility-scale deployments, turning stranded-watt recovery into a repeatable source of revenue per gigawatt.
- NVIDIA has not disclosed the revenue share percentage or how usage is measured.
- Whether the DSX-aligned factories reach the tokens-per-megawatt efficiency and utilization the model assumes, and whether Firmus’s $25–30 billion in committed off-take converts into recognized revenue on schedule.
See the complete announcement on the company blog.
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Other Insights From Futurum:
Can ADI Hot Swap Controllers De-Risk NVIDIA’s 800 VDC Transition?
Will Adobe and NVIDIA’s RTX Spark Partnership Redefine Creative AI Workflows?
NVIDIA Cosmos 3 and Open Agent Tools: Is Physical AI About to Leave the Lab?
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
