Why computing power, as much as platform design, determines armoured vehicle lethality

Modern AI classification is the bedrock for the latest mission support tools and second order effects that are set to transform battlefield effectiveness and lethality – capabilities Stewart, Head of Digital Strategy at Thales, referenced in his article last year ahead of IAVC 2025. These algorithms are proven and deployable today.

Yet, most armoured platforms cannot run them effectively – not because their sensors are not capable, not because the algorithms are not mature, but because the computer architecture sitting between them was designed for a different technological era. “The limiting factor for deploying AI at scale isn't the platform itself, but the processing hardware sitting inside it,” says Stewart, Head of Digital Strategy at Thales.

Why processing hardware matters now

Achieving the Chief of the General Staff's goal to triple Army lethality by 2030 requires us to think differently about how we deploy capability to the frontline.

DigitalCrew capabilities are already deployed and delivering operational value. Utilising traditional mathematical algorithms for detection, tracking, and image fusion, they are running perfectly well on current processing hardware. These building blocks are proven on platforms today, helping crews detect and track potential threats across complex battlespaces. This is not the case for AI classification and the second-order effects that flow from it.

What we are talking about is not incremental improvement – it is unlocking entirely new classes of capability from sensors already installed on platforms. These second-order effects transform raw data into a step-change in tactical advantage. But delivering that kind of capability leap demands significant increases in processing power.

The structural challenge: procurement vs technology evolution

Defence procurement, in its current form, locks in technical specifications years before fielding. Platform development cycles can span 15 years or more to move from a design to frontline service. This timeline works perfectly well for components with multi-decade service lives – such as hull armour, powertrains, and optical systems. These remain capable throughout a platform's operational lifetime.

But GPU evolution follows a completely different clock. New GPU architectures emerge every three to five years, and some manufacturers will cease support on similar timelines, making it increasingly difficult to develop algorithms for older hardware. The result is an unavoidable gap between processing hardware specification during the design phase and algorithm capability at fielding.

Commercial sectors have adapted to this reality, operating on three-to-five-year hardware refresh cycles to keep pace with technology evolution. Defence procurement needs to adopt similar iteration loops for processing hardware, while maintaining longer lifecycles for the platforms themselves.

The consequences of not doing so are already visible. Consider a platform where computer hardware was specified in the early 2010s, now fielding in the mid-2020s. The sensors remain front-line ready and will last the lifetime of the platform, yet the GPU is already struggling to run today's latest algorithms. Running classification algorithms, it manages just 5-6 frames per second versus the required 30-60 fps. The consequence is jerky, unusable displays for human operators – viable only for machine-to-machine data passing rather than real-time crew decision-making.

The real-world consequence: locked opportunities

The impact of this mismatch is profound. Modern armoured fighting vehicles can have dozens of cameras providing 360-degree awareness around the platform, yet human crews can actively monitor perhaps two feeds at most. AI classification could monitor all feeds simultaneously, alerting crews when threats appear. The algorithms exist. The sensors exist. Yet the processing hardware cannot connect them at operational tempo.

Without it, this locks away those second-order effects that would provide genuine tactical advantage. Passive ranging, for instance, would allow crews to determine target distance without laser detection, avoiding the risk of revealing their position. Threat prioritisation algorithms could assess multiple targets based on type, range, and behaviour, then recommend which to engage first. Classification enables vehicle configuration analysis, determining gun orientation, and anomaly detection – all capabilities that help crews make faster, better-informed decisions under pressure.

Perhaps most importantly though, AI classification reduces the cognitive burden. If, instead of requiring crews to watch everything, AI can handle the monitoring tasks, then crews are freed up to focus on decision-making and engagement. The opportunity cost of not having these capabilities is significant: crews operating without advantages that could be unlocked through hardware refresh rather than platform replacement.

The path forward: treating processing hardware as consumable

“We need to recognise that GPU hardware requires planned three-to-five-year replacement cycles, independent of platform lifecycle. One practical way to deliver this is through a hardware as a service model for military platforms, where onboard computing processing is provided, maintained, and refreshed on a contracted cycle to sustain AI performance over the life of the vehicle,” says Stewart.

Delivering this at scale requires a common, GPU-enabled processing architecture across platforms and domains. Standardised interfaces would make DigitalCrew building blocks truly portable – develop once, deploy anywhere without re-engineering the software wrapper for each new system.
 

The benefits are clear: 

• Agility: planned GPU replacement every three-to-five-years keeps processing capability aligned with algorithm evolution, without waiting for platform replacement.
• Portability: standardised architecture means developing once and deploying anywhere, with no re-engineering for each platform.
• Cost: reduced integration costs and predictable refresh budgeting over platform lifetime.
• Immediate impact: unlocking AI capabilities on existing fleets without vehicle replacement.

This is not about choosing between new platforms and new processors – it is about recognising they operate on different timescales. New platforms are essential, but their processing architecture must be designed for regular hardware refresh from day one.

Where the next leap comes from

Sensors and vehicle platforms remain capable for decades. The factor limiting their lethality is the computing hardware sitting between the sensors and the shooter. By treating that hardware as consumable, requiring periodic refresh, forces can unlock new waves of AI-enabled performance without rebuilding fleets.

The next major leap in armoured vehicle capability won't come from a new turret design, a new engine, or a new hull. It will come from upgrading the GPU inside – not because the platform needs replacing, but because the technology evolution cycle demands it. The armies that understand this will field AI-enabled advantages on existing platforms while others wait for next-generation procurement cycles to deliver similar capabilities.

The question isn't whether to modernise processing hardware. It's whether to do it proactively, as part of a planned refresh strategy, or reactively when the capability gap becomes a tactical liability.