AI infrastructure is scaling at breakneck speed, with massive investment fueling new data centers, GPU clusters, and edge systems. That growth also expands the attack surface, raising the risk of costly breaches, operational disruption, and compromised trust for organizations across industries.
“Building the Backbone of AI,” a new report by Brookfield, a global investment firm whose Brookfield Asset Management (BAM) division boasts more than $1T in assets under management (AUM), estimates that more than $7 trillion will be invested in AI infrastructure over the next decade — spanning upgraded power grids, global connectivity from fiber and telecommunications to satellites, and modular hardware designed to keep pace with rapid innovation.
This expansion is actively multiplying the attack surface. Every new node, from hyperscale cloud facilities to portable edge units, introduces fresh opportunities for cyber and physical threats. (Crucially, our SanQtum cybersecurity solution secures nodes as this attack surface expands and infrastructure becomes more distributed.) And as AI gets baked into more software and apps that didn't have it before, the digital attack surface multiplies, too — making every added piece of intelligence an exponential vulnerability multiplier.
For leaders across industries like healthcare, energy, industrial, financial, government, manufacturing, and transportation and logistics, it’s critical to understand the scope of this emerging threat landscape.
Key trends from Brookfield’s report — and their cybersecurity implications
The Brookfield report — which outlines opportunities to invest in the infrastructure that is expected to support the next industrial revolution — sheds detailed light on key trends with deep implications for cybersecurity.
- Inference will dominate AI workloads by 2030. The report forecasts that most compute will soon be spent on inference, not training. AI inference is the ability of a trained AI model to recognize patterns and draw conclusions from information it hasn’t seen before. It underpins many of AI’s most exciting applications, such as generative AI, and allows models to imitate the way people think, reason, and respond to prompts.
While AI training typically occurs in centralized, hyperscale cloud data centers, inference increasingly happens at the edge — on distributed, sometimes portable devices that require near real-time, ultra-low latency access to compute resources.
This shift means that model integrity, runtime protection, and on-device data security will be just as important as securing training pipelines in centralized environments.
- Distributed deployment expands the attack surface. Edge systems, mobile units, and geographically scattered nodes offer inherent security advantages — deployments with limited access points can be more resilient against cyber attacks targeting centralized infrastructure like telecom networks or power grids. But distributed deployments are also often physically exposed, harder to monitor, and attractive targets for theft or tampering.
These mixed advantages and risks require security strategies that go beyond traditional firewalls, demanding comprehensive protection to address firmware, hardware, and physical threats simultaneously.
- Hardware must be built for upgradeability. Brookfield’s report emphasizes modular, upgradeable hardware to keep pace with AI innovation. This imperative for adaptability extends directly to cybersecurity systems, which must be designed with the same flexibility in mind. Security architectures need trust anchors, cryptographic modules, and firmware that can be upgraded or replaced without disrupting operations, while organizations must stay current with emerging technologies, and be prepared to deploy them accordingly.
Implementing a trusted cybersecurity SaaS strategy can give you the power of the most robust technology, while freeing you from having to invest directly in the hardware or constantly monitor for updates. For example, as a managed service, SanQtum and SanQtum AI take care of this for you. We’ve designed our hardware to be modular, so we can swap and upgrade switches, routers, chips, and other components. But you don’t need to worry about that. We handle the headache, just as if we were updating software.
- Cyber-physical convergence increases risk. AI infrastructure relies on integrated systems like IoT cooling sensors, power distribution, and robotics. Each system adds new cyber-entry points and interdependencies. This convergence means that edge security must evolve beyond traditional software patches, including tamper detection, geofencing, robust encryption, and rapid remote wipe capabilities that can respond to threats across both digital and physical domains.
Security priorities for edge AI stakeholders
Whether building AI infrastructure or deploying it, all organizations need zero trust architecture and robust security practices. Beyond these fundamentals, here are priorities by role:
For infrastructure providers:
- Embed hardware roots of trust in AI accelerators and ensure supply chain transparency.
- Harden distributed units with physical protections like locks, GPS tracking, and kill switches.
- Design failover capabilities so operation continues even if core networks are compromised.
For enterprise AI deployers:
- Implement quantum-resilient encryption to protect against current and future threats.
- Deploy continuous threat monitoring that can isolate emerging threats in real time.
Why now is the time to defend your organization
Brookfield’s report on the trillions pouring into AI infrastructure investment highlights how massive and distributed growth will be — and with that scale comes a wider range of potential attack surfaces. Many organizations are rushing to deploy AI infrastructure and models without prioritizing security from the outset, waiting until deployment risks unpatchable vulnerabilities and exposes critical models, data, and physical systems to attack.
Now is the time to build in zero trust cybersecurity. Unsecured AI is an inexcusable, massive risk. AI infrastructure, model training, and inference need to grow hand-in-glove with cutting-edge cyber protections to stay ahead of risks that have never been greater.
Security needs to grow with AI, not after it. One of the easiest and fastest ways to do so is via an as-a-service model, such as SanQtum and SanQtum AI. Contact our team to learn more.
Image: Unsplash | Paul Hanaoka
