Independent research firm SemiAnalysis argues that the computing power required for AI safety and cybersecurity may sustain demand for Nvidia Corp. (NASDAQ: NVDA) GPUs, even as frontier model training slows.

Analyst Max Kan stated on a Tuesday podcast that market participants may be overlooking two critical factors supporting chip demand. First, AI companies continue to seek more computing capacity than can physically come online. Second, the process of making models safer itself requires significant compute resources.

Safety Work Eats Compute

Kan emphasized that AI labs increasingly use models to monitor other systems, test safeguards, and check reinforcement-learning environments. He noted that investors often underestimate the compute required for safety, alignment, and monitoring.

Interpretability research, which aims to understand how models reach answers, was also cited as a compute-intensive workload. Anthropic CEO Dario Amodei has echoed this view, proposing that the industry pace frontier AI by devoting more resources to alignment, interpretability, testing, and operational safeguards rather than halting training entirely.

Cybersecurity Cuts Both Ways

SemiAnalysis suggested that advanced AI could reshape cybersecurity by forcing companies to fight AI with AI. The panel referenced Greg Brockman’s description of a continuous security system at OpenAI that uses the latest models to constantly find and patch vulnerabilities.

As attackers deploy autonomous agents at scale, defenders may need equally capable AI systems running continuously to detect and stop them. One speaker described this as potentially "the most impactful shift in cybersecurity," arguing that firms delivering security outcomes rather than just tools could capture enormous value.

The hosts warned that major cybersecurity firms could become prime targets for increasingly capable AI attacks. They named:

• Okta Inc. (NASDAQ: OKTA)
• CrowdStrike Holdings Inc. (NASDAQ: CRWD)
• Palo Alto Networks Inc. (NASDAQ: PANW)
• Zscaler Inc. (NASDAQ: ZS)

Near-Term Pause Still Bites

OpenAI paused frontier reinforcement-learning training following its Hugging Face incident. Its largest planned frontier run remains on hold while smaller training and evaluations continue.

President Donald Trump called fears of runaway AI a "hoax" on Monday, questioning why an industry would seek rules that could bankrupt it. Polymarket traders assign a 16% chance to a U.S. AI safety law passing before 2027, with about $123,000 traded on the contract. The contract covers measures including restrictions on AI training or use and requirements for human oversight.

For Nvidia, the central question is whether slower frontier training cuts more GPU demand than safety, monitoring, and cybersecurity add back. Shares recovered 0.6% on Tuesday.

What the Numbers Show

The divergence between OpenAI’s paused large-scale training and SemiAnalysis’ argument for sustained demand hinges on the volume of compute allocated to non-training tasks. While the source does not quantify the exact ratio of safety compute to training compute, it highlights that safety, alignment, and cybersecurity are becoming distinct, resource-heavy workloads. This suggests that total GPU utilization may remain high even if traditional model training cycles slow, shifting the demand driver from pure scale expansion to operational safety and defense.

Nokia Corp (NYSE: NOK) and Nvidia Corp (NASDAQ: NVDA) are advancing AI-native radio access network trials with eight global telecom operators. The partnership aims to integrate accelerated computing into wireless infrastructure.

AI Moves Into RAN

Nokia announced Wednesday that A1 Group, Chunghwa Telecom, du, e&, Mobily, stc, TPG Telecom, and Zain Saudi are conducting proofs of concept and live field trials. These trials utilize Nokia’s AI-native radio access network platform powered by Nvidia’s Aerial RAN Computer across Europe, Asia-Pacific, and the Middle East.

The technology uses AI to improve wireless network efficiency. Nokia reports its AI-RAN platform has demonstrated more than 20% spectral-efficiency gains. Advanced AI models target more than 2x efficiency over time. Higher spectral efficiency allows operators to move more data through existing spectrum without adding hardware.

Nvidia CEO Jensen Huang described the radio access network as "the next AI infrastructure." He framed the shift as moving from using AI to optimize networks toward making the network itself an AI computing platform.

Nokia Wants AI at the Edge

Nokia expects its AI-native RAN platform pilots to begin later this year. The company targets commercial availability in 2027. Nokia views this technology as a pathway from 5G and 5G-Advanced toward AI-native 6G.

The strategy turns Nokia’s existing network footprint into a software-driven growth opportunity by combining anyRAN software with Nvidia accelerated computing to run AI models at radio-network timescales.

What the Numbers Show

The disclosed efficiency metrics highlight a divergence between current performance and long-term targets. While the platform has already demonstrated more than 20% spectral-efficiency gains, the advanced AI models aim for more than 2x (or 100%+) efficiency over time. This gap suggests the initial commercial value proposition relies on immediate capacity gains, while the broader strategic shift toward making the network an AI computing platform depends on achieving these higher efficiency thresholds in future deployments.

Nvidia’s AI Footprint Expands

Nvidia has built its AI dominance around data-center computing. AI-RAN opens another potential layer of demand within the telecommunications network. Nokia provides radio expertise and operator relationships, while Nvidia supplies accelerated computing and AI software.

The commercial opportunity remains developing. Operator trials must demonstrate that additional computing produces enough network capacity and economic value to justify large-scale deployment. As AI workloads spread from centralized data centers to the network edge, Nokia positions its radio business within Nvidia’s expanding AI infrastructure ecosystem.