The importance of sector allocation

Sector allocation is one of the most consistent drivers of tracking error and alpha in active high yield bond portfolios. In today’s U.S. high yield market, sector tilts have become even more crucial as bond-level dispersion narrows. We expect sector dispersion to remain elevated, shaped by policy uncertainty and uneven technology risk – including those associated with AI.

Deep fundamental research is essential to uncover themes that will influence markets, and early identification and conviction around these themes are key to successful sector positioning. One such theme is the rapid buildout of AI infrastructure and the significant capital investment it attracts.

IPPs emerge as beneficiaries of AI theme

When ChatGPT reignited interest in AI, fixed income opportunities were initially concentrated in investment grade and private credit markets, leaving few options in high yield. By 2024, the U.S. high yield electric utilities sector emerged as an early beneficiary of the AI theme. Independent Power Producers (IPPs), which own and operate generation assets, stood out as direct beneficiaries of rising power prices. AI data centers consume enormous amounts of energy, driving both power consumption and prices higher.

Figure 1: Demand for power for data centers is expected to rise significantly in the US
US data center energy consumption (TWh) and share of US power demand (%)
Source: McKinsey & Company, Global Energy Perspective 2023, October 2023. Share of US data center energy consumption (TWh = Terawatt hour) and data center share of US power demand (%). Figures beyond 2024 are estimates. There is no guarantee that past trends will continue, or forecasts will be realized.

Hyperscalers have publicly acknowledged that power availability is one of their largest constraints in building new AI data centers. Earlier this year, Amazon and Talen Energy announced the largest nuclear power purchase agreement (PPA) to date—a 17-year, $18 billion agreement for up to 1.92 GW of capacity.¹ Power constraints have become so acute that some hyperscalers are now contracting with companies deploying modular microgrids using small gas turbines to quickly energize new facilities.

The AI datacenter ecosystem: New opportunities emerge

Throughout 2025, demand for AI data centers continued to surge, exposing severe power constraints and highlighting the challenges of building new generation capacity. Companies that had already secured grid access found themselves in a uniquely advantageous position.

Attention soon shifted to Bitcoin miners – firms that had locked in substantial power for energy-intensive mining, but historically struggled with rising network difficulty, high energy costs, and crypto price volatility. Among them, TeraWulf emerged as the first to pivot toward High-Performance Computing (HPC) and tap the high yield market. This strategic move re-rated TeraWulf from a cyclical, low-multiple business to one with higher margins and long-term contracted earnings. Its high yield bond deal, aimed at funding an AI data center buildout, carried construction risk and involved little-known counterparties, requiring an innovative structure backed by Google to secure financing. As shown in Figure 2, Google provides a lease backstop and lease cash flows from the tenant are deposited directly into an agent-controlled account (the lockbox). These funds are automatically applied first to mandatory debt amortization and interest payments for lenders, ensuring that debt service is prioritized and fully covered before any remaining cash can be used for operating expenses or other purposes. Initially, the rationale behind Google’s involvement was unclear, but we believe an opportunity to deploy Google’s Tensor Processing Unit (TPU) chips at the facility was a driver.

Also emerging onto the scene have been specialist data center operators known as “neo-clouds”: a “neo-cloud” is a new generation of cloud provider built specifically for AI and HPC. Unlike traditional hyperscalers, neo-clouds focus exclusively on leading-edge infrastructure, ultra-fast networking, and environments optimized for training and running large AI models. Fluidstack and CoreWeave are two leading providers. For example, Fluidstack leases a powered shell – a building with electrical infrastructure – from TeraWulf, installs Google-designed TPU clusters and supporting equipment, and sells compute capacity to downstream customers such as hyperscalers or AI labs that do not wish to undertake the data center operations themselves.

Figure 2: Simplified relationship structure
Cash flows protected by lockbox structure to help ensure prioritization of mandatory amortization

While Fluidstack has no public debt outstanding, CoreWeave has already issued two high yield bond deals this year and a convertible bond.² Hyperscalers typically issue in the investment grade market, and most AI labs remain privately funded, but notably, xAI (developer of the Grok LLM) entered the high yield market with a deal earlier this summer.

Other suppliers to the AI ecosystem

As investors gained clarity on what goes into an AI data center, new opportunities emerged among suppliers to other parts of the ecosystem—particularly in memory and networking.

Memory remains highly cyclical, and we believe we are in the early stages of a supercycle, supported by rising price expectations and low supply. Also emerging is NAND memory: as AI evolves from text-based to multimodal (image and video generation), demand for flash memory should accelerate, as it’s the only NAND type fast enough to support these workloads. Kioxia, a Japanese NAND memory leader, tapped the U.S. high yield market with its inaugural bond deal earlier this year, and has emerged as an additional way to gain exposure to AI hardware within the high yield investment universe. Networking providers are also benefiting from the need to interconnect GPUs within racks, link racks across buildings, and connect data centers to each other and the broader internet. High yield names like Ciena and Coherent supply critical networking equipment, while Level 3 Communications’ fiber network enables AI data centers to communicate across regions and with end-users.

The competitive landscape: Forming alliances

Recently, Google’s Tensor Processing Unit (TPU) has drawn attention as a rival to Nvidia’s Graphics Processing Unit (GPU) for AI data centers. GPUs are versatile and support a wide range of workloads, while TPUs are designed for certain AI tasks—faster in those areas, though less flexible and largely specialized for Google’s ecosystem and corresponding workloads. Beyond chips, Google competes with Microsoft in enterprise software through Google Workspace (Docs, Sheets, Gmail, Meet) versus Microsoft 365 (Word, Excel, Outlook, Teams), and its Gemini large language model (LLM) challenges OpenAI’s ChatGPT. In short, Google is fighting on three AI fronts: hardware against Nvidia, AI-enabled enterprise software against Microsoft, and LLMs against OpenAI. Two major alliances have emerged: the ‘Google complex,’ including Broadcom (which co-design TPUs), Fluidstack, and TeraWulf/Cipher Mining; and the ‘OpenAI complex,’ featuring Nvidia, CoreWeave, Applied Digital, and non-Google hyperscalers such as Microsoft and Oracle. Price action is moving swiftly as the market constantly reevaluates which players are gaining a competitive edge in the AI race.

Figure 3: Price action in equity markets reflects shifting sentiment towards alliances
Price performance of Google complex and OpenAI complex
Source: Bloomberg, Janus Henderson calculations, Google complex is market cap weighted basket comprising Google, Broadcom, Lumentum, Cipher Mining, and TeraWulf; OpenAI complex comprises NVIDIA, Microsoft, Oracle, Softbank, Coreweave, Applied Digital. Price performance in US dollars from 31 December 2024 to 9 December 2025. Past performance does not predict future returns.

Looking ahead

Opportunities to invest around the AI theme in high yield have steadily expanded, and we expect this trend to continue. This broadens the ways portfolio managers can drive alpha through both sector allocation and security selection in AI-exposed industries. Over time, many of these companies may become less reliant on the high yield market as they gain access to other funding sources, including securitized credit and private debt. Some existing high-yield bonds could be called early and refinanced into ABS structures. Meanwhile, the massive capex programs of investment grade hyperscalers will benefit the entire ecosystem, as spending flows across the AI data center supply chain – where many participants remain high yield issuers.

¹Source: Talen Energy presentation, ‘Executing our strategy: Validating the thesis on power & data intersection’, 11 June 2025.
²Source: Bloomberg, correct at 11 December 2025.

Fixed income securities are subject to interest rate, inflation, credit and default risk. The bond market is volatile. As interest rates rise, bond prices usually fall, and vice versa. The return of principal is not guaranteed, and prices may decline if an issuer fails to make timely payments or its credit strength weakens.

High-yield or “junk” bonds involve a greater risk of default and price volatility and can experience sudden and sharp price swings.