Read the funding figures with their scope attached
Review update, September 2026: The Department's FY2027 budget overview requests $17.9 billion for Golden Dome. It describes investment in space-based and terrestrial sensors, command and control, kinetic interceptors and non-kinetic capabilities. These are requested resources, not proof of appropriation, delivery or operational coverage.
The Congressional Budget Office's May 2026 analysis estimates roughly $1.2 trillion over 20 years for a notional national missile-defense system broadly consistent with the executive order. That is an illustrative architecture, not a cost estimate of a fully disclosed government design. Comparing it directly with a one-year request or a differently scoped administration estimate produces a misleading impression of precision.
The figures nevertheless show why architecture matters. Sensor coverage, defensive capacity, replenishment and operating costs all depend on what the system is expected to protect against. A headline total cannot substitute for those assumptions.
Command and control is already part of the program
Battle management is not an undiscovered requirement. The MDA FY2026 justification explicitly included Golden Dome integration within C2BMC work. The issue is whether the integration is sufficiently mature, tested and supportable for each intended increment.
A system connecting space, air, land and maritime capabilities must reconcile different update rates, data definitions, security boundaries and operational responsibilities. A sensor can report accurately within its own design limits while another application misinterprets its confidence or treats delayed information as current. Such failures occur at interfaces, where a component-level demonstration may not expose them.
Hardware remains central. Software cannot create missing sensor coverage or replenish inventory. Equally, capable hardware cannot compensate for information that arrives in an unusable form. Golden Dome needs both, with integration evidence that explains the limits of each fielded configuration.
Apply the CJADC2 lesson carefully
The organizational challenge resembles the broader effort to connect military command-and-control systems. GAO's CJADC2 review identified the need for a comprehensive framework to guide progress. Common interfaces and demonstrations are useful, but they do not independently settle priorities, responsibilities or what counts as an acceptable operational result.
Golden Dome involves homeland-defense responsibilities and coordination across organizations. Requirements, test plans and operating procedures should identify the responsible mission owners, their decision authorities and the process for resolving legal and civilian-safety questions across organizational boundaries.
For a program review, the practical questions are concrete:
- Who owns each shared data product and resolves a disagreement about its meaning?
- Which organization approves a changed interface or software release?
- Who can identify an unusable input and communicate its effect to downstream users?
- Which mission owner accepts the residual risk when an integration is incomplete?
Software integration supports those decisions. It does not make them on behalf of the responsible organizations.
Assess AI where it is actually used
Fast automated processing is not necessarily machine learning. A battle-management architecture may combine deterministic software, statistical methods and learned models. Its assessment should identify the method used for each function rather than describe an entire defensive system as end-to-end AI.
Where AI is proposed, speed must be evaluated alongside reliability and operational suitability. A recommendation delivered quickly but grounded in stale or inconsistent information may create more risk than a slower, well-bounded capability. Performance also has to be assessed under the intended workload and hardware constraints, not only in a prepared demonstration.
An assurance plan should address:
- Representative evidence: test data and scenarios that cover intended use and meaningful exceptions.
- Uncertainty: clear indications when inputs or outputs do not support a confident conclusion.
- Adversarial robustness: independent assessment of misleading inputs and system vulnerabilities within an authorized test environment.
- Human interaction: evaluation of whether operators understand the system's limits and retain the authority required by policy.
- Auditability: protected records of versions, inputs, decisions and relevant overrides.
- Change control: reassessment when models, sensors, interfaces or mission assumptions change.
Timing constraints depend on the intended scenario and approved design. The responsible program should connect each timing requirement to its evidence about detection, information exchange and decision processes, including uncertainty when conditions differ from those tested.
Make each increment earn its expansion
The most useful delivery sequence is one in which each increment has a defined purpose, a controlled configuration and evidence about its performance. Integration work should make it possible to identify which dependencies are mature, which remain experimental and what happens when one is unavailable.
For industry, that creates work in data integration, test infrastructure, cybersecurity, operational evaluation and sustainment as well as hardware production. Reusable tools can reduce repeated effort, but reuse needs an explicit account of what remains valid in the new mission context. A successful test is evidence for the conditions tested, not a universal guarantee.
Sources and further reading
- FY2027 Department budget overview
- CBO: Potential costs of a national missile-defense system
- MDA FY2026 RDT&E justification
- GAO: CJADC2 framework and progress
Spartan X brings AI, cybersecurity and systems engineering together around that integration problem: making the evidence behind a complex capability clear enough to support a responsible delivery decision.
