Investors should maintain exposure to AI hardware and semiconductor chipmakers, as rising GPU prices and expanding enterprise inference demand continue to benefit leaders supporting the NVIDIA H100.

Allocate capital toward next-generation endpoint security and AI governance providers, as legacy signature-based cybersecurity tools face rapid disruption from autonomous models.

This enterprise security upgrade represents an immediate growth catalyst, with roughly 50% of enterprise software applications projected to become agentic AI systems by the end of the year.

Look for opportunities in multi-model routing and open-source AI hosting infrastructure, which are benefiting as businesses actively diversify away from proprietary providers like OpenAI and Anthropic.

Finally, favor infrastructure platforms supporting flexible open-weight architectures such as Qwen and GLM-5-2, as enterprises prioritize customizable systems to overcome restrictive vendor guardrails.

AI Hardware & GPU Infrastructure

• The market for AI hardware is entering a major expansion phase as inference shifts from centralized data centers to endpoints.
• Demand for high-performance GPUs remains robust, driven by the ongoing build-out across cloud providers and enterprise infrastructure.
  • GPU prices are reportedly rising again, accompanied by emerging market shortages and capacity constraints.
  • Advanced hardware remains costly; a single NVIDIA H100 GPU costs approximately $250,000 per year to run, creating a barrier to entry for smaller teams seeking to self-host large AI models.

Takeaways

• Hardware and chipmakers powering AI infrastructure continue to see strong demand fundamentals as workloads transition from model training into distributed, enterprise-wide inference.

Agentic AI & Enterprise Cybersecurity

• Enterprise cybersecurity is undergoing a structural disruption as traditional defenses become obsolete.
  • Traditional security tools were designed to combat human attackers and static malware, whereas AI agents are autonomous and lack fixed behavioral signatures.
  • Modern security methods like honeypots and static signatures are generating high false-positive rates because autonomous agents actively discover and misuse embedded credentials.
• Adoption is accelerating rapidly:
  • An estimated 50% of enterprise applications are expected to become agentic by the end of the year, with remaining applications expected to follow shortly.
  • The average enterprise manages between 6,000 and 7,000 unique software applications, creating a massive new attack surface.
• Defensive strategies are shifting toward endpoint security and AI governance tools that monitor the interaction layer between humans and AI models.

Takeaways

• Legacy cybersecurity providers relying purely on signature-based detection face disruption, creating growth opportunities for modern endpoint security and automated defense platforms tailored to agentic workflows.

Frontier AI Models & Open-Source LLMs

• Frontier AI labs (including developers of proprietary models such as OpenAI and Anthropic) are implementing strict safety guardrails to prevent exploitation.
  • A side effect of these guardrails is that defensive cybersecurity teams face high rates of "refusals" and false positives when testing their own systems for vulnerabilities.
  • To bypass vendor lock-in and guardrail limitations, security teams are increasingly adopting flexible, multi-model architectures.
• Enterprises are balancing between closed-source API providers and open-weight/open-source models (such as GLM-5-2, KMEK-3, and Qwen) to ensure operational flexibility and full control over security workflows.

Takeaways

• Companies offering model-routing, open-source AI hosting, and flexible infrastructure layers are well-positioned as enterprises look to avoid vendor lock-in and overcome restrictive model guardrails.