Agentic AI Boom Mints New Winners: Valuations Reach New Highs Across the Stack

The early Artificial Intelligence (AI) boom was underpinned by the demand for ever larger training runs mostly on densely interconnected NVIDIA Graphics Processing Units (GPUs) leveraging the CUDA stack to build frontier models expanding at an exponential scale. Much of the profits accrued to NVIDIA with silicon and systems still unmatched in that domain, paid for largely by hyperscalers’ largesse, the scale of which (continues to) dwarf prior tech cycles: The five largest U.S. hyperscalers are on track to spend roughly US$725 billion on AI infrastructure in 2026 alone, up from about US$450 billion in 2025.

The main driver of the current cycle is Agentic AI, with enterprises realizing value, in particular, for coding and software tasks. These emerging workloads may run dozens or hundreds of inference calls per task, orchestrate tool use, retrieve context, and execute multi-step workflows. This shift from training-heavy to inference-heavy—from one-shot to persistent—fundamentally changes what data center operators and enterprises need as they transition to more general-purpose Central Processing Unit (CPU) compute to orchestrate workloads. Second, the mushrooming power needs of today’s compute (and the transition to 800 Volt (V) Direct Current (DC) AI servers) is another spending driver, and power-conversion silicon leaders are also anticipating significant windfalls from their solutions that support racks that surpass the 100 Kilowatt (kW) mark. Third, the integration, deployment, and servicing capabilities for complex, fine-tuned clusters (with more memory-rich servers) optimized for inference economics is firmly in demand as token use is expanding exponentially.

Three companies with valuations that have surged in 2026—Intel, Infineon, and Penguin Solutions—sit squarely on each of those vectors.

Intel

Demand for Intel’s broad CPU portfolio has experienced unprecedented demand as its primary chipsets are placed at the heart of the orchestration layer and critical control plane for the entire AI stack—including Agentic AI. This has led to a significant increase in the CPU-to-GPU ratio in AI deployments with the expectation of a trajectory toward parity. Intel’s management flagged that demand continues to run ahead of supply, particularly for Xeon server CPUs, and that its advanced packaging has shifted from "hundreds of millions" to "billions of dollars per year" in expected revenue. Chief Executive Officer (CEO) Lip-Bu Tan remarked: "The next wave of AI will bring intelligence closer to the end user, moving from foundational models to inference to agentic. This shift is significantly increasing the need for Intel's CPUs.” The AI-driven business now represents 60% of Intel’s revenue and grew 40% Year-over-Year (YoY).

Intel's 1Q 2026 earnings release from April 2026 is an explicit articulation of the Agentic AI CPU thesis, especially when compared with other major chipmakers. Intel's stock has risen roughly 200% year-to-date in 2026, hitting an all-time high near US$130 in May before settling. The most recent catalyst was 1Q 2026, when the Data Center & AI segment revenue grew 22% YoY to US$5.1 billion, and the company has told customers that its server CPU capacity is essentially sold out for 2026 with price increases of 10–15% anticipated. Revenue again beats guidance, which validates the claim that demand outstrips supply.

Infineon

The expansion of power infrastructure is gaining momentum and becoming an increasingly important growth driver for Infineon’s industrial business. This includes a recently unveiled Trans-Inductor Voltage Regulator (TLVR) quad-phase power module, aimed specifically at next-generation AI server boards to service the increasing power of the next generation of high-performance NVIDIA and AMD servers. Demand is very strong for Infineon’s power solutions, which led to price increases on power switches and power Integrated Circuits (ICs) effective April 1, 2026, citing AI data center-driven shortages.

Infineon's shares have moved decisively in 2026, with a gain of around 120% year-to-date, and a particularly sharp gain in the days following its 2Q 2026 results on May 7. Infineon's 2Q earnings release explicitly calls out the growth in revenue from the AI infrastructure buildout: “power supply solutions for AI data centers are in very high demand.” In fact, management has raised FY 2026 Capital Expenditure (CAPEX) to €2.7 billion to expand the manufacturing capacity for power supplies for AI data centers. AI data center revenue is guided to €1.5 billion in FY 2026, increasing to €2.5 billion in FY 2027, and is characterized again as supply-limited.

Penguin Solutions

Penguin Solutions’ expertise in integrating highly optimized clusters, including from a memory standpoint, is in strong demand. This dovetails with enterprise adoption of inference and Agentic AI workloads. This is acknowledged by their management, and the inference-and-memory narrative is the underlying technical thesis: training was compute-bound, inference is memory-bound, and Penguin's 40-plus years of memory-subsystem expertise are positioned to capture that shift. Central to this growth is the need for memory as a critical scaling factor for AI inference, one of Penguin Solutions’ strengths and a significant part of its server portfolio. Relevant product announcements include the industry's first production-ready Compute Express Link (CXL)-based Key-Value (KV) cache server and the OriginAI Factory Platform optimized for AI inference.

Penguin Solutions’ valuation has also moved decisively year-to-date, marking an increase of around 260% from US$20 to over US$70. This growth is based on a string of catalysts. In 2Q 2026, integrated memory revenue jumped 63% YoY, the company increased its FY 2026 revenue-growth guidance to 12%, and AI-driven business now represents 60% of 1H revenue. April’s earnings release noted "enterprises, governments, and neocloud providers are racing to build AI factories, as platforms scale to power the next generation of inference workloads.”

This AI infrastructure cycle is moving on from single compute chip vendors and homogenous buyers such as hyperscalers. We can now observe a value chain story touching on all aspects of the AI hardware stack—from power delivery to compute. Advice for players across the stack follows:

• From Training to Agents and Inference: Reposition the messaging around inference and agents, not training. Training is concentrated among a handful of hyperscalers and largely NVIDIA-allocated. Inference is broader and far more sensitive to total cost per token. Salient is the quantification of customer outcomes in tokens-per-dollar and tokens-per-watt, not throughput alone.
• Capacity Allocation as a Commercial Lever: Explicit mentions of supply constraints signal to customers that allocation itself is a strategic asset. Multi-year agreements, capacity reservations, and priority-access tiers can now be used as legitimate parts of the GTM motion and a credible way to lock in revenue visibility ahead of competitors.
• The Transition to Systems-Level Offerings Has Fully Crystallized: For end-to-end solutions vendors like Penguin Solutions that package memory, software, and services into a "full-stack platform" for AI factories, this has been a boon. Pure-component vendors will face commoditization pressure and the erosion of value-add margins gained by those that bundle reference designs, validation, orchestration software and managed services.
• Preparation for Enterprise AI Lift-off: Moving beyond hyperscalers, the new growth pockets are neoclouds, sovereign AI initiatives, large enterprises deploying on-premises inference, and Tier One regulated buyers in sectors such as finance, telecommunications, and healthcare. Each vertical requires a different sales motion, channel, and dedicated reference architectures.