AI Power & Cyber Stocks Beyond Semis, per Monetary Matters & Pernas

Host Jack Farley of Monetary Matters interviews Dea and Dean Pernas of Pernas Research about where they see the most compelling AI-related opportunities as of August 2026. The brothers argue that investors are overly focused on semiconductors and hyperscalers, while underestimating second‑order beneficiaries in power, bandwidth, cyber security, and payments.

They frame the current pullback in the “memory and AI complex” (they mention SanDisk and SK Hynix) and the weakness in hyperscalers as a backdrop, but do not view it as the end of the AI story. Instead, Dean and Dea contend that open‑weight AI models, massive data‑center capex, and grid constraints are reshaping where the economics accrue in the stack.

Across the conversation, they highlight smaller or less mainstream names tied to “bring your own power” data centers, inter‑data‑center bandwidth, AI‑driven cyber security, and niche payments. Farley repeatedly stresses that their audited track record since 2017 shows over 1,300% gross returns, and positions the discussion as an insight into how those investors think, rather than as general macro commentary.

Dean outlines a four‑layer AI stack: chips, infrastructure/cloud, frontier models (like ChatGPT and Anthropic), and applications. He argues open‑weight models shift value away from frontier labs and toward cloud providers and GPU owners because cheaper weights spur more AI usage, invoking the “commoditize your complement” idea.

Dea adds that hyperscalers such as Oracle, Microsoft, Google, Amazon, and Meta face a step‑function rise in capex with uncertain ROI, while trading around what he calls roughly 10x EV/sales on average. In his view, those top‑heavy index leaders are being forced to subsidize AI for the rest of the economy, creating fertile ground for stock pickers in adjacent areas.

Instead of owning semiconductors, their portfolio themes center on:

• Energy / bring‑your‑own power for data centers.
• Bandwidth between geographically distributed data centers.
• Cyber security moving from identity to exposure management.
• Payments / fintech, especially cross‑border and niche verticals.

They argue these areas can capture hyperscaler and AI spend without bearing the same valuation risk as the mega‑caps.

On energy, Dean says U.S. electricity demand was roughly flat for 20 years around 4,000 terawatt hours, but he believes AI could reach about 20% of U.S. grid demand by 2030, or roughly 100 gigawatts. He argues the grid’s “two‑lane highway”‑like transmission and power‑fluctuation limits, plus 18–24 month interconnection queues, push hyperscalers toward behind‑the‑meter natural gas power.

To play this, he contrasts Bloom Energy’s high revenue multiple (he cites ~15–20x) with a much smaller turbine maker, Capstone, which he says traded around 3x revenue. Capstone’s micro‑turbines are smaller and cheaper, with lead times of 1–3 months versus 18–24 months for very large turbines. He notes Capstone emerged from bankruptcy, reached its first year of profitability recently, runs at roughly 10% capacity, and could serve data‑hall scale loads (tens of megawatts), edge data centers, and industrial sites strained by grid constraints.

Dean also mentions Stabilus Solutions, a small‑cap “virtual pipeline” provider that liquefies and trucks natural gas to data centers lacking pipeline access and to customers such as SpaceX. He references a contract worth a couple hundred million dollars over two years with a data‑center client, and anticipates rising launch‑driven LNG demand from SpaceX.

For bandwidth, he points to Smart Optics in Scandinavia, which provides hardware and software to convert data to light and modulate it across fiber between data centers. As mega‑sites meet community pushback and energy limits, he expects many 1‑gigawatt campuses instead of single, larger complexes, requiring up to 15x more inter‑data‑center bandwidth. He says Smart Optics is well‑known in Silicon Valley and interoperates with players like Broadcom and Cisco.

On robotics, Dean cites Vishay Precision Group (VPG) as a play on humanoid robots, stressing it is not to be confused with a passive component maker of a similar name. He says Vishay Precision makes strain gauges and sensors for robotic hands and is already working with at least three humanoid robotics customers, including Tesla and Figure. If AI enables robust robotic “brains,” he suggests humanoid robotics could become one of the largest industries in a decade.

In cyber security, Dea argues that AI models capable of scanning thousands of lines of code cheaply are reviving traditional hacking risk. Identity‑based attacks (like phishing) may have driven perhaps around 80% of recent incidents, but he says CISOs and CTOs now fear code‑level exploits as the cost of vulnerability discovery falls. Here, they favor Tenable, which they describe as leading in “exposure management,” using frontier models (via a partnership with Anthropic) to inventory assets, surface critical vulnerabilities, and help fix them. Dea notes Tenable trades around 4x EV/sales while CrowdStrike and Palo Alto trade above 20x, in his telling.

Despite their enthusiasm for certain themes, the Pernas brothers repeatedly frame ideas through valuation and risk. On hyperscalers, Dea points to what he sees as roughly 10x EV/sales valuations combined with massive, essentially mandatory AI capex, and calls the index “vulnerable” if ROI disappoints.

Within energy, they acknowledge Capstone is a roughly $300–350 million fully diluted company that historically made no money and only recently achieved profitability after emerging from pre‑packaged bankruptcy. Dean notes the stock rose about 1,131% from their August 5, 2025 initiation to May 29, 2026, then roughly halved, yet remains up about 545%, which makes position sizing and expectations important.

For Stabilus, Dea labels it a “speculative” holding, sized at 1–3% in their framework, partly because it has an approximately $75 million market cap and operates in a still‑small, consolidating niche. In payments, he sees opportunity in highly leveraged names such as Paysafe, which he says carries around $2.5 billion of debt versus roughly $600 million of book value, and whose headline revenue growth is modest and complicated by asset sales.

Dea also concedes that many SaaS and cyber names, including some they own, screen poorly on GAAP metrics due to high stock‑based compensation and decelerating growth. He implies that this makes them unattractive to traditional “deep value” investors despite large share‑price drawdowns.

Looking ahead, Dean argues that open‑weight models’ impact on AI unit economics is crucial. If enterprises can avoid paying what he characterizes as around $10 per million input tokens to frontier labs, he expects Jevons‑type effects: cheaper inference leading to more demand, higher GPU rental rates, and stronger economics for whoever controls compute capacity, particularly cloud providers.

On infrastructure, they are watching how quickly AI workloads push data‑center power toward the cited 20% share of the U.S. grid and whether grid bottlenecks and local opposition continue to force “bring your own power” builds and multi‑site architectures. Dean believes moratoriums and community pushback on 1–3‑gigawatt campuses are likely to persist, reinforcing demand for both behind‑the‑meter natural gas solutions and inter‑data‑center connectivity.

In cyber, Dea is focused on evidence that attackers are systematically using AI to discover and weaponize code vulnerabilities, which would validate the pivot from identity‑centric defense toward exposure management platforms like the one Tenable is building. He also tracks whether large incumbents successfully adapt or cede ground to more specialized platforms.

In payments and fintech, Dea downplays stablecoins as a near‑term threat to consumer and small‑business cross‑border providers, arguing most stablecoin volume is still crypto‑trading related. He is more interested in the ongoing shift of processing economics toward vertical SaaS platforms (such as restaurant‑oriented systems like Toast) and in niche processors positioned in gaming and prediction markets. Across all these areas, both brothers emphasize that their views reflect where they expect AI‑driven spending and regulation to move, not a static snapshot of 2026 conditions.