T2C

A Deep Understanding of the DSX Ecosystem and IREN’s Position Within It Conclusion: DSX is NVIDIA’s system for productizing the “AI factory.” It combines Apple-like system control with Wintel-like externalization of assets. IREN occupies a position of generational exclusivity: the inaugural co-designed operator and a primary source of site-level and operational engineering standards. I. Platform Architectures Across Three Eras The PC era: Wintel — two standards and a thin-margin integration layer. Microsoft defined the software interfaces, Intel defined the processor and platform reference designs, and OEMs such as Dell and Compaq handled integration and shipment. Profits concentrated in the two standards-setting layers, while the integration layer faced intense competition and thin margins. OEMs bore the inventory and capacity risks. The mobile internet era: iPhone plus apps — a closed-loop hardware platform and a software ecosystem monetized through revenue sharing. Apple controlled the chips, devices, operating system, distribution, and payments, while also owning the end-user relationship. The more active the developer ecosystem became, the greater the platform’s returns. Apple compounded its earnings through a share of distribution revenue rather than by developing the apps itself. The AI inference era: the DSX ecosystem — system-level control without end-user ownership. The aspects resembling Apple: NVIDIA vertically defines the entire system, from compute, networking, power supply, and liquid cooling to digital twins and control software. It uses certification (Exemplar) and flagship demonstrations to select operators. The aspects resembling Wintel: Productization and integration are carried out by multiple partners. NVIDIA neither owns nor operates the factories, externalizing the heavy-asset and depreciation risks to its partners and operators’ balance sheets. Where DSX differs from both: it does not own the end-user relationship. Customers remain in the hands of operators, so NVIDIA can exert influence only through certification, reference designs, and demonstration factories. II. IREN’s Unique Position and Irreplaceable Role Once the DSX ecosystem begins operating, it will enter a path of continuous iteration and upgrades, much like Wintel and Apple. IREN has already become the flagship deployment site for DSX AI factories. Combined with its 5.8 GW portfolio of sites backed by PPAs, this positions IREN to be among the leading co-development partners in every DSX generation, conducting joint commissioning with unreleased next-generation hardware and operating it over the long term at real, gigawatt-scale sites. This position could bring IREN enormous commercial benefits. The specific mechanisms and channels through which those benefits will materialize remain an open question, and we should expect to see multiple ways of realizing them. IREN can occupy this position because it possesses a combination of attributes that can only be accumulated over time: Gigawatt-scale power capacity already connected to the grid, with approximately 5.8 GW secured, including 2 GW at Sweetwater. Sweetwater, the world’s largest single site to be energized first, remains open for joint design and serves as a co-development sandbox. A neutral position: no proprietary chip development, no token sales, no status as a hyperscaler, and no geopolitical baggage. An end-to-end chain of delivery responsibility, from site selection through optimization, backed by a unique execution record: Horizon 1 was delivered and accepted by Microsoft on August 13, 2026, and received GB300 NVL72 Exemplar Cloud status on the same day. Infrastructure software control capabilities through Mirantis. NVIDIA’s own core R&D workloads running on IREN’s infrastructure, under a $3.4 billion cloud services agreement. Advanced liquid-cooling expertise, including talent such as John Gross. Vertical integration and the capabilities of an engineering general contractor. At present, IREN is the only candidate that combines all these attributes. CoreWeave lacks owned power infrastructure and neutrality. Nebius lacks grid-connected power capacity and carries geopolitical baggage; its Token Factory is a core commercial offering, with multiple areas of overlapping commercial interests with NVIDIA. Crusoe lacks sufficient depth in behind-the-meter power and integration. Hyperscalers have a site-density profile that does not fit the model, while their proprietary chip development creates conflicts of interest with NVIDIA. DSX standards operate across three layers: the architecture layer, primarily controlled by NVIDIA; the modules and equipment layer, involving multiple partners such as Dell, Vertiv, and Schneider; and the site and operations layer. IREN is a primary contributor at the third layer and a key source of as-built evidence for the entire system. III. IREN’s Potential Commercial Benefits Given the long-term nature of the DSX ecosystem and its continuous iteration and upgrades, IREN’s position within these three layers of standards could create immense potential value, provided it continues to keep pace with the system’s evolution. That continued participation is far more significant than it might initially appear. IREN operates primarily at the site and operations layers, both of which offer substantial value to develop. The sites themselves have value. The operations layer, however, serves as the point of contact between the DSX ecosystem and its users. For IREN, this layer plays an even more important role by bridging the upstream system and downstream demand. Its potential customer base may extend beyond a diverse range of compute users to include multiple customers seeking to build AI compute infrastructure rapidly. Given the broad spectrum of AI users, IREN’s decision to start building its brand now and sponsoring major sporting events demonstrates considerable strategic foresight. For IREN, NVIDIA is both a customer and an essential routing channel for demand. NVIDIA can direct customers requiring gigawatt-scale, sovereign, or enterprise-grade deployments to IREN, a model that aligns with NVIDIA’s own structural interests. Option value: A neutral control plane combined with reserved capacity could allow IREN to increase revenue per unit of capacity by serving sovereign, enterprise, and multi-tenant customers requiring capacity partitioning. It could also serve as a natural physical anchor for standardized compute futures. This is only a very rough overview. There are likely to be multiple additional channels through which IREN can benefit, and I welcome everyone’s contributions.

NVIDIA will establish the overall standards, but IREN can develop engineering-level implementation standards. These practical engineering standards will gradually take shape as the DSX ecosystem evolves and matures. If companies want to deploy products based on the DSX ecosystem in the future, the ability to rapidly build the supporting infrastructure will be crucial, as it will enable computing capacity to come online quickly. IREN’s ability to bring computing capacity online rapidly opens up an entirely new dimension of opportunity—not just for itself, but with the potential to extend its impact across the world.