The Manhattan Project brought allied scientists to Los Alamos because winning required pooling allied scientific capacity. Today, as Australia invests $3.4 billion in quantum and advanced defense technologies and autonomous capabilities, the United States has built the supercomputing infrastructure those technologies require — then locked its partners out of accessing it.
President Donald Trump’s Genesis Mission invokes that wartime urgency to win the AI race — but the November executive order gives American companies detailed frameworks to access federal supercomputers while offering international security partners one vague sentence about exploring collaboration “to the extent appropriate.”
That isn’t an accident: It reflects policy priorities. The Department of Defense’s January 2026 AI strategy frames allied collaboration as instrumental: “leverage the capabilities and insights of allies and partners to enhance our collective defense capabilities.” The language is open to a generous reading — “collective defense capabilities” could imply shared infrastructure and co-development. But Genesis partnerships tell a different story.
The Department of Energy announced collaboration agreements with 24 organizations in December 2025. Every partner is American, including OpenAI, Anthropic, Google, Microsoft, Amazon Web Services, and NVIDIA. The United States controls 74 percent of global AI compute capacity. Genesis gives American industry structured access through Cooperative Research and Development Agreements, while AUKUS allies — already investing billions in quantum and autonomous systems requiring exactly this infrastructure — received no equivalent mechanism.
The barrier isn’t technical. It’s whether Washington wants allied access when policy objectives center on American companies exploiting American infrastructure to cement American technological leadership.
The administration’s approach may prove insurmountable under current policy priorities — treating allied capacity as something to leverage rather than integrate. But that calculus ignores what the United States loses through fragmentation: Every AI model that Washington forces allies to develop separately is duplicated effort across the partnership. Every Australian quantum processor or British autonomous system trained on domestic infrastructure instead of Genesis represents capability development that the United States either funds unilaterally or foregoes entirely. China now leads in 57 of 64 critical technologies — including quantum, AI, and advanced materials — central to Pillar II of AUKUS, but Beijing lacks allied research ecosystems. The strategic advantage isn’t American dominance of infrastructure. It’s allied innovation capacity integrated with American computing power that no competitor can match. That requires infrastructure access, not just technology transfer agreements.
Unless policymakers act during Genesis implementation, this infrastructure gap will cripple the partnership’s most ambitious technology initiative. The fix is conceptually straightforward, even if politically difficult: extend the Cooperative Research and Development Agreement authorities that Genesis establishes for American industry to AUKUS allies.
The Gap Pillar II Was Never Built to Close
In June 2025, Peter Dean and Alice Nason argued that AUKUS Pillar II was failing in its mission — regulatory reform masquerading as progress while delivering no actual capabilities. They were right. Genesis is not a defense program — it is a Department of Energy initiative, and Pillar II was never designed to reach across that institutional boundary. But the technologies Pillar II prioritizes — quantum computing, autonomous systems, defense AI — have computational requirements that only Department of Energy scale infrastructure can meet. What Dean and Nason couldn’t have anticipated was Washington building exactly the computing infrastructure Pillar II needs in a department that AUKUS was never structured to access.
The gap isn’t theoretical. When Australia funds quantum machine learning processors or successfully trials quantum clocks for satellite navigation in contested environments — demonstrations conducted in Washington as part of AUKUS Pillar II — where do they validate performance at scale? What infrastructure trains the AI models these systems require? The computational power needed exists almost exclusively in Department of Energy laboratories that Genesis now integrates. And allied partners cannot access it.
Training defense AI models isn’t like fine-tuning commercial chatbots. These are foundation models that must process sonar data and provide weapon recommendations in contested battlefield scenarios. Training such AI systems, which must operate autonomously in underwater environments and coordinate across multiple domains, demands computing power measured in exaflops — the scale of the Department of Energy exascale supercomputers that Genesis integrates for large-scale model training, which are resources only a handful of facilities globally possess.
Allied governments face three options, none of which are adequate. They can build domestic Genesis-scale infrastructure — the United Kingdom’s computing roadmap targets a new exascale system by 2027, requiring years of construction and billions in capital that might better fund actual quantum sensors or autonomous vehicles. They can rely on commercial cloud services — Australia is building a dedicated Top Secret cloud with Amazon Web Services scheduled for 2027 operations — but this approach requires bespoke security arrangements negotiated program-by-program rather than shared infrastructure designed for trilateral defense AI research. Or they can hope for ad hoc access to American facilities through arrangements that don’t exist — and if they did, would require negotiating every request without established frameworks for compute allocation, data handling, or intellectual property protection.
None of these options serves the strategic logic the partnership was designed to embody: integrating allied innovation capacity to deliver advanced capabilities at unprecedented pace rather than fragmenting efforts across uncoordinated national programs.
This gap is both unnecessary and fixable. Genesis was purpose-built for the integrated AI platform this research requires.
These are operational requirements, not theoretical research domains demanding advanced computing infrastructure, secure compartmented environments, and integrated protocols that protect both data and intellectual property while enabling collaborative development.
A Fix Hidden in the Executive Order
The solution exists within the executive order itself. Section 5(c) establishes Cooperative Research and Development Agreements as the mechanism for structuring industry partnerships with the Department of Energy. These authorities come with decades of precedent managing intellectual property, controlling technology transfer, and maintaining security. The framework already exists. What’s required is extending it.
Allied governments would operate under bilateral frameworks mirroring the Cooperative Research and Development Agreements model — tiered access with pre-approval for identified defense technology priorities such as quantum computing and autonomous systems, case-by-case review for exploratory research, and congressional notification for allocations exceeding defined thresholds.
This provides oversight without bureaucratic gridlock. Intellectual property rights create no barrier. Models trained with allied datasets would remain allied property under technology transfer controls — but implementation matters. “Allied datasets” means sovereign defense data: Australian sonar signatures from Royal Australian Navy platforms, or British quantum sensor data from British defense systems. “Allied property” means algorithms and models trained on that data.
The framework would need enforceable provisions ensuring Australian AI models trained using classified submarine data on Genesis infrastructure remain under Canberra’s control, transferable only under Australian decisions, with technical safeguards preventing unilateral U.S. access. This addresses longstanding concerns about intellectual property governance and preserves strategic autonomy — if properly implemented and maintained amid policy volatility.
Data sovereignty simply requires extending the Department of Energy’s existing security protocols — authentication systems, audit trails, secure compartments already deployed for classified research — to allied datasets. This is operational adaptation, not architectural redesign.
Technology transfer risks are manageable through existing controls. Genesis-trained models carry the same export restrictions as any jointly developed defense technology, operating under the International Traffic in Arms Regulations that already govern U.S.-U.K. and U.S.-Australian cooperation.
The bilateral frameworks would include end-use monitoring, third-party transfer prohibitions, and termination clauses — standard provisions in current defense agreements. The legal architecture would layer bilateral frameworks for each ally onto shared AUKUS protocols. When trilateral collaboration occurs, the frameworks interlock, preventing conflict while maintaining each nation’s strategic control.
The Technical Cooperation Program — a multilateral defense science forum among the United States, the United Kingdom, Canada, Australia, and New Zealand — has coordinated classified research for decades. Genesis frameworks would build on that foundation, adapted for the scale and speed these emerging technologies demand.
Making this happen faces barriers on both sides. The United States and Australia lack reciprocal security clearance recognition, requiring duplicate vetting for every researcher. Personnel security grows complex when researchers have family or educational ties to countries of concern, complicating clearance timelines and creating friction between programs demanding rapid development and security processes requiring thorough review. More fundamentally, recent U.S. policy decisions raise Australian sovereignty concerns.
Diversity, equality, and inclusion scrutiny affecting international research collaborations, proposed data collection initiatives, and intelligence-sharing disruptions create legitimate questions about whether partnerships would remain insulated from political interference — and whether sovereign datasets and intellectual property actually stay sovereign once trained on U.S. infrastructure. Australian quantum companies developing dual-use technologies face additional structural challenges: segregating defense-funded research from commercial activities while meeting stringent U.S. export control requirements could disadvantage them against purely domestic competitors.
Companies like Q-CTRL and Diraq — which develop quantum processors and software for both financial services and defense applications — already navigate complex dual-funding streams from Australian government programs and U.S. Army Research Office contracts. Adding Genesis infrastructure access would layer additional export compliance requirements onto existing obligations. While AUKUS reforms eased some controls in 2024, industry concerns persist that harsh penalties for minor compliance failures could outweigh collaboration benefits — particularly for startups competing against purely domestic U.S. quantum firms facing no such restrictions.
This isn’t just about Genesis. AI infrastructure isn’t just compute — it’s the “digital value chain” of labeled datasets, cloud environments, and development operations. Genesis addresses the compute piece. If that infrastructure successfully enables allied capability development through considered frameworks, it establishes a template for other strategic resources: data repositories, test ranges, and evaluation facilities.
The Cold War model involved American technological dominance with selective transfers to allies. That worked when U.S. defense research and development spending dwarfed combined allied investments, and technology development timelines stretched across decades. Neither condition holds today. AUKUS reflects recognition that technology competition requires allied innovation ecosystems, not just American breakthroughs with partner adoption. Australia’s quantum technology leadership shows what allied expertise enables. But allied innovation requires infrastructure. If the United States builds Genesis-scale compute platforms while allies lack equivalent resources, the “partnership” defaults to asymmetric dependency regardless of policy rhetoric.
The window for action is narrowing. As the Department of Energy signs industry agreements and deploys security protocols, retrofitting allied access means renegotiating established arrangements — exponentially harder than incorporating partners from the start.
Congress should require the Department of Energy to extend Cooperative Research and Development Agreement authorities to AUKUS partners, or Genesis will become another case study in partnerships undermined by infrastructure gaps the United States had the capacity to close but chose not to.
Eighty years after Los Alamos, Genesis offers the same choice: build infrastructure for allied innovation or watch AUKUS’ most ambitious technology initiative fragment across uncoordinated national programs when the computing power for integration already existed.
Javaid Iqbal Sofi is an AI governance researcher and policy consultant who has advised international organizations on regulatory frameworks. He can be reached on [email protected]
**Please note, as a matter of house style, War on the Rocks will not use a different name for the U.S. Department of Defense until and unless the name is changed by statute by the U.S. Congress.
Image: U.K. Ministry of Defence via X.