Defence goes digital – data centres and AI on the frontline of national security

It took the ‘red team’ just four days. A cyber-attack led by members of the National Security Agency (NSA), using only publicly available software, gained root access to at least 35 US government networks and took control of the computer systems of US military command in the Pacific. They also conducted simulated attacks on power grids and 911 emergency systems in eight US cities. Perhaps unsurprisingly given the ease of the red team’s infiltrations, they discovered that spies from other countries had been there before them. Codenamed Eligible Receiver 97, and taking place from 9–13 June 1997, the security exercise led directly to the creation of US Cyber Command.¹

Since then, the world has become more connected and the tools of digital attack and defence vastly more powerful. In June 2026, 29 years to the day after Eligible Receiver, another red team exercise was revealed to have discovered exploitable vulnerabilities in numerous highly sensitive US government classified systems. Though there has been no official confirmation, it is widely reported that the tool used was Anthropic’s Mythos artificial intelligence (AI) model, and that it was able to discover weak points in a matter of just hours.²

If geopolitical strength in the nineteenth century rested on control of the seas, and in the twentieth century came increasingly from control of the air, today’s quest for dominance is surely over the digital domain. Having embedded itself across business functions and many areas of everyday life, AI is now playing a growing role in defence – and the data centres that serve as its engine room have found themselves on the frontline of national security. The question for governments is – can they achieve digital sovereignty before it’s too late?

Cybersecurity – digital defence in a realm without borders

In March and April 2026, police forces in India made a series of arrests following a shocking discovery. Chinese-made CCTV cameras near military installations and other sensitive sites had been broadcasting live footage – including of troop movements – to Pakistan’s Inter-Services Intelligence (ISI), via data servers in China.³

Though many of the cameras had been installed covertly, others were pre-installed cameras that had been accessed remotely – and most were of the widely available, cheap, Chinese-made variety that Indian officials and security experts have long warned are vulnerable to backdoor access.⁴ India responded with a nationwide audit of CCTV equipment – at the same time, already-planned certification rules came into force that have, in effect, banned the sale of many Chinese-made cameras.

As autonomous AI capabilities improve, efforts to build digital safe zones will become still more difficult

Keeping out the hardware that could contain cyber backdoors is a key step in achieving digital sovereignty, but import controls can’t address the fundamental problem – in the digital world there are few borders. And as autonomous AI capabilities improve, efforts to build digital safe zones will become still more difficult – AI can quickly get behind enemy lines precisely because the lines are so hard to draw. For example, in July 2026, OpenAI disclosed that, during internal testing, multiple AI agents had collaborated secretly to escape the sealed environment in which they were being tested, gain access to the internet, and break into the production systems of a third-party organisation.⁵

However, just as AI can be used to hack, so it can act as a digital sheriff, identifying and eliminating threats long before a human could respond. Demonstrating this, at the final of the DARPA (US Defense Advanced Research Projects Agency) AI Cyber Challenge in August 2025, AI systems evaluating the open-source code underpinning hospitals, power plants, and water systems found 86% of the planted vulnerabilities and autonomously patched 68% – more than doubling their success rate from a year earlier. They also turned up 18 genuine flaws that had not been identified previously.⁶

Defending against cyber intruders is, of course, as much a corporate story as a government one. At Lombard Odier, we believe the sector is seeing structural growth, unfolding as organisations are forced to rethink how they protect their operations. Industry analysts expect worldwide spending on information security to reach USD 240 billion in 2026, up 12.5% in a year⁷. The market for AI in cybersecurity specifically is growing faster still – forecast to rise from USD 31.5 billion in 2025 to nearly USD 183 billion by 2033, at nearly 25% a year.⁸

AI and digital technology – the new national security paradigm

If cybersecurity is one side of the digital sovereignty coin, the other is defence. Here, too, AI and data centres are playing a growing role. Since 2020, for example, the US’s Project Maven has helped commanders in the field to identify targets at speed by analysing surveillance images and other data to spot patterns.⁹ NATO adopted a related system in 2025 – both depend heavily on the private sector, from the vision models that enable computers to interpret images, to the commercial cloud on which they run.¹⁰

While AI-enhanced weapons targeting systems are, understandably, controversial, AI’s wider uses in the defence sector are often in more prosaic high-volume repetitive and administrative tasks. One of its most valuable uses is in intelligence, sifting a flood of satellite imagery, drone footage, and intercepted signals, and flagging the fraction worth a closer (human) look.¹¹ It is also transforming the unglamorous but decisive business of logistics, anticipating where supplies will be needed and which equipment will fail before it does.¹²

The world’s major powers are investing heavily in what has become an AI arms race

Though the figures are largely classified and opaque, there’s little doubt that the world’s major powers are investing heavily in what has become an AI arms race. Unusually, significant sums are flowing to the commercial firms that build consumer AI, rather than to traditional defence contractors. In the US last year, the Pentagon awarded contracts worth up to USD 200 million each to Anthropic, Google, OpenAI, and xAI,¹³ instead of commissioning bespoke military systems.

The battle for data centre dominance

AI’s growing importance for defence and national and corporate cybersecurity has pushed data centre capacity to the top of many government agendas.

In the US, for example, what appears to be a drab-looking logistics park dropped, incongruously, into the vast, rolling “Red Plains” of west-central Texas, is in fact the flagship of the world’s largest ever planned data centre project – Project Stargate – with an explicit intention of “protecting the national security of America and its allies.”¹⁴ Unveiled at the White House in January 2025, Stargate has since seen delays and reported disagreements between the project partners – OpenAI, Oracle, and SoftBank. Despite this, the physical buildout continues, with seven sites under construction that are projected to bring online more than 9 gigawatts (GW) of capacity by 2029¹⁵, equivalent to roughly 20% of current US capacity¹⁶.

The European Union has answered with InvestAI, a plan to mobilise EUR 200 billion explicitly to build technological sovereignty, with a call launched in July 2026 aiming to unlock more than EUR 30 billion for up to seven “AI gigafactories”.¹⁷ China is pursuing similar goals, with Beijing reported to be drafting a USD 295 billion plan for a national network of data centres. Significantly, China’s plans will aim for domestic firms to supply at least 80% of the core technology.¹⁸

This focus on digital sovereignty is a sign of the startlingly fragile supply chains that underpin AI and data centre new-builds

Semiconductors and critical minerals – ‘friendshoring’ to secure supply

This focus on digital sovereignty is a sign of the startlingly fragile supply chains that underpin AI and data centre new-builds. The most advanced AI processors are predominantly designed by just one company – US-based Nvidia.¹⁹ Around 90% are manufactured by another single company – Taiwan’s TSMC.²⁰ And everyone relies on the one supplier that makes the chips possible – Dutch firm ASML, the world’s only producer of the extreme-ultraviolet (EUV) lithography systems required to print the finest circuitry.²¹

The vulnerability runs deeper still, however, to the underlying materials themselves. Modern chips and servers depend on rare-earth elements and critical minerals. Here, China holds commanding control, accounting for around 60% of global rare-earth mining and 91% of the world’s processing and refining capacity²², giving Beijing a powerful geopolitical lever. Throughout 2025, China imposed a widening series of export controls on rare earths and critical minerals – extending even to the processing technology itself – sending prices spiking and Western technology manufacturers scrambling to secure supply.²³

Western powers have responded with a coordinated “friend-shoring” push to route around China, including an EU-US critical minerals pact and a landmark EU-Australia trade deal which will remove tariffs for nearly all Australian critical minerals exports.²⁴ For today’s digital Allied forces, freedom from the chain of chokepoints won’t come overnight, however – industry analysts believe it could take a decade for the West to begin to challenge China's rare-earth dominance²⁵.

Data centres – from frontline to final frontier

At Lombard Odier, we analyse today’s changing approaches to defence and critical supply chains through our rethink security investment theme. Enormous national commitments to defence investment – such as NATO’s commitment to spending 5% of GDP on defence²⁶ and the EU’s EUR 800 billion ReArm Europe plan²⁷ – will boost both traditional and cutting-edge defence capabilities.

Investors must not think in silos, however. As tech firms and new AI technologies become increasingly embedded in national security, so broader digital infrastructure becomes key. According to consultants McKinsey, between now and 2040, USD 19 trillion must be spent on digital infrastructure to keep pace with growing demand.²⁸

Data centres will be an important part of this new infrastructure, and plans are ambitious – industry analysts expect such rapid build-out that data centres’ annual power demand will nearly triple between 2025 and 2030²⁹. Over the longer term, even this ambition could be outstripped if the world’s biggest tech names achieve their goal of putting data centres in space³⁰. Combined plans for well over a million ‘data centre’ satellites tout the benefits of uninterrupted solar power and zero land constraints.

However, cooling is difficult beyond Earth’s atmosphere, where the vacuum of space prevents convection or conduction, so whether data centres successfully move at scale from the frontline to the final frontier remains to be seen. What is certain, though, is that as the defence landscape evolves and cyberspace becomes the new contested domain, the data centres that power it will become increasingly attractive targets. For governments and businesses alike, this makes data centres both a key vulnerability and a prized asset in efforts to build long-term security.