Executive Summary
Artificial intelligence is creating one of the most significant shifts in corporate credit risk in recent years. While public discussion has focused largely on equity valuations, productivity gains and technology adoption, the implications for creditworthiness are only beginning to emerge.
This report examines how banks’ collective credit assessments have evolved across the AI value chain, identifying the sectors benefiting from AI investment alongside those facing growing structural pressure from automation. Drawing on Credit Benchmark’s consensus credit data, it provides a forward-looking view of where default risk is improving, where it is deteriorating, and how these trends have changed over the past three years.
The findings reveal a clear divergence.
Cloud & AI Platform Operators, semiconductor companies, networking infrastructure providers and power utilities have generally experienced improving credit profiles as investment in AI infrastructure accelerates. By contrast, sectors whose business models depend heavily on repetitive knowledge work or labor-intensive service delivery—including Customer Experience & Business Process Outsourcing (BPO), IT services, staffing and education—have seen materially weaker credit trends as automation begins to reshape demand.
The most striking contrast is between Cloud & AI Platform Operators, whose consensus probability of default has fallen 15% over the past three years, and Customer Experience & Business Process Outsourcing providers, where default risk has increased 92%. These changes suggest AI’s impact is already extending well beyond technology companies into industries that support enterprise operations across the global economy.
Importantly, the report also demonstrates that AI’s effects are not uniform. Several sectors widely expected to be disrupted – including advertising, media and publishing – have shown improving credit fundamentals, reflecting the resilience of established market leaders. Conversely, some apparent beneficiaries are beginning to show signs of increasing leverage as the cost of building AI infrastructure grows.
Rather than presenting AI as a simple story of winners and losers, the analysis highlights a more nuanced transition: one in which credit quality is diverging within and across industries as business models adapt to a rapidly changing competitive landscape.
For lenders, investors, insurers and risk managers, these trends provide an independent perspective on how AI is reshaping credit risk – and where structural change may create opportunities, vulnerabilities and emerging concentrations of risk before they become visible through traditional measures.
Camp: winners = AI beneficiaries; losers = AI-exposed; “mixed” = improved in credit despite equity-market disruption fears. PD in basis points; ΔPD is the three-year change in default risk (negative = improving).
Scope & Method – This review uses Credit Benchmark’s consensus credit data — the probability of default (PD) that contributing banks assign to each borrower, aggregated and mapped to the CB21 rating scale (aaa strongest to c weakest; the investment-grade / high-yield boundary is ~55 bps, between bbb- and bb+). We screened 337 companies across the AI value chain into ten custom segments; 206 carry a live bank-sourced consensus. Aggregate trends run May-2023 to May-2026; rating changes compare to twelve months earlier. Because the data is bank-sourced, coverage is deep on large borrowers and thin on small, asset-light equities.
On Sample Size – The service segments (outsourcing, staffing, education, content, advertising and IT services) were deliberately broadened with global and privately-held names so the aggregates do not rest on a handful of companies. This changed one headline materially: education’s three-year deterioration falls from +131% on the original eight names to +58% on twelve, because the extreme figure was partly small-sample noise. Staffing’s decline also proved broad-based rather than driven by a single collapse. Where a segment still has few consensus names (outsourcing and education, twelve each), the figures should be read directionally.
Segment Analysis
The remainder of this report examines each of the ten AI-related sectors individually, including:
- Sector definition and rationale
- Aggregate consensus credit trends
- Three-year probability of default evolution
- Rating distribution
- Key observations and emerging risks
1. Semiconductors & Chip-Making Equipment
WINNER · Consensus rating bbb+ · PD 16 bps · 3-yr ΔPD -3% · 32 of 48 screened names carry a consensus
What is in this segment, and why – Companies that design and fabricate the chips used to train and run AI — GPUs, AI accelerators, memory and logic — with the equipment and materials makers that build the fabrication plants. They are the physical supply of AI compute (the “picks and shovels”), so demand reaches them first as model training scales.
How it is performing (aggregate) – Solidly investment grade and essentially flat over three years — a marginal 3% improvement in default risk. A stable winner rather than an improving one.
Concentrated in single-A and triple-B, with a high-yield (bb) tail of smaller equipment and fabless names. No distressed names.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
2. Cloud & AI Platform Operators (Hyperscalers)
WINNER · Consensus rating bbb+ · PD 20 bps · 3-yr ΔPD -15% · 21 of 21 screened names carry a consensus
What is in this segment, and why – The large cloud providers and platform-software firms that monetize AI at scale and are the biggest buyers of AI hardware. They capture AI demand directly through cloud consumption and AI features — but they are also funding the build-out with capital, so their balance sheets carry the capex.
How it is performing (aggregate) – The strongest three-year improvement of any group (-15%), but it peaked in mid-2025 and has been giving ground since, as debt-funded capital spending mounts. The recent direction is the warning sign.
Bar-belled: a cluster at aa/a (the cash-rich mega-caps) against a large bb block (debt-funded GPU-clouds and unprofitable software). The widest quality spread of any group.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
3. Networking & Data-Center Hardware
WINNER · Consensus rating bbb · PD 22 bps · 3-yr ΔPD -4% · 31 of 34 screened names carry a consensus
What is in this segment, and why – The equipment that connects and physically houses AI compute — network switches and optics, servers, and the electrical and cooling systems inside data centers. AI clusters need dense networking and far more power and cooling than traditional IT, pulling spend toward these suppliers.
How it is performing (aggregate) – Roughly flat over three years, with a 2024 wobble now recovered. A middle-of-the-pack investment-grade group where the AI benefit is real but spread across many diversified industrials.
Centered on single-A and triple-B, with a bb/b tail of smaller hardware and distressed cabling names. Broadly healthy.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
4. Power Generation & Utilities
WINNER · Consensus rating bbb · PD 25 bps · 3-yr ΔPD -9% · 18 of 22 screened names carry a consensus
What is in this segment, and why – The companies that supply the electricity AI data centers consume — power generators, regulated utilities, nuclear and uranium, and grid and turbine equipment. Power availability has become the binding constraint on AI expansion, so load growth and long-term supply contracts flow to this group.
How it is performing (aggregate) – Improving, with most of the gain in the last three quarters as the AI-electricity thesis converts into firmer credit — and, notably, no downgrades at all in the past year.
Evenly split between single-A and triple-B, with a small bb/b tail of merchant-power and fuel-cell names.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
5. IT Services & Consulting
LOSER · Consensus rating bbb- · PD 32 bps · 3-yr ΔPD +43% · 29 of 35 screened names carry a consensus
What is in this segment, and why – Firms that sell technology implementation, systems integration and managed services, billed largely by people and time. At risk because AI code-generation and automation compress the billable hours their revenue depends on.
How it is performing (aggregate) – A steady three-year deterioration (+43%), and the aggregate has just slipped from bbb to bbb-. It remains tail-driven: a healthy single-A/triple-B core of majors is offset by a weak tail, including one distressed name and several further names in high yield.
A solid single-A/triple-B core (21 of 29 names) with a bb/b/c tail of eight, including one distressed name.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
6. Customer-Experience & Business-Process Outsourcing (call centers & back-office)
LOSER · Consensus rating bb- · PD 138 bps · 3-yr ΔPD +92% · 12 of 33 screened names carry a consensus
What is in this segment, and why – Firms that run outsourced customer support — call and chat centers — and back-office processing for other companies. The single most AI-exposed segment: chatbots and AI agents can deflect the very call and transaction volumes these firms are paid to handle. (“CX” is customer experience; “BPO” is business-process outsourcing.)
How it is performing (aggregate) – Still the worst performer: consensus default risk rose ~92% over three years and the group now sits at bb-, deep in high yield. The picture holds after the sample was broadened — and the split is starker: commoditized voice/back-office operators are collapsing while analytics-led processors are improving.
No single-A names at all; a heavy bb block and a quarter of the group (three names) already in the c near-default bucket.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
7. Education & Edtech
LOSER · Consensus rating bb- · PD 136 bps · 3-yr ΔPD +58% · 12 of 39 screened names carry a consensus
What is in this segment, and why – Companies selling courses, learning platforms, tutoring and educational content. At risk because free, capable AI tutors substitute directly for paid study help, homework tools and course material.
How it is performing (aggregate) – Deteriorated ~58% and sits at bb-. Broadening the sample matters here: on the original eight names the figure was +131%, so much of that extreme was small-sample noise. The larger set adds several leveraged education publishers, which sit in single-B and confirm a genuinely weak, debt-heavy segment — just not a collapsing one.
Concentrated in bb and single-B (the leveraged publishers), with one distressed name in the c bucket and no single-A support.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
8. Content, Media & Publishing
MIXED · Consensus rating bbb · PD 29 bps · 3-yr ΔPD -12% · 17 of 32 screened names carry a consensus
What is in this segment, and why – Producers and licensors of written, visual and reference content — news, stock imagery, professional information and creative tools. Exposed where generative AI substitutes for stock media, freelance content and search-driven traffic.
How it is performing (aggregate) – Still improving (-12%) after the sample was broadened — confirming that the consensus-covered content universe is the diversified information incumbents, which are healthy. The pure disruption targets remain largely uncovered by banks. AI content disruption is, so far, an equity story rather than a credit one.
Anchored in single-A/triple-B (the information incumbents), with a bb tail of smaller media names and one clearly AI-hit name in single-B.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
9. Advertising & Marketing
MIXED · Consensus rating bbb- · PD 43 bps · 3-yr ΔPD -30% · 17 of 31 screened names carry a consensus
What is in this segment, and why – Advertising agency holding companies and ad-technology firms that plan, create, buy and measure advertising. At risk where AI automates creative work and media buying; helped where ad-tech names benefit from AI-driven targeting.
How it is performing (aggregate) – Improved even more clearly after adding ad-tech and agency names (-30%), and was upgraded from bb+ to bbb-. Consolidation among agency holding companies and fast-growing ad-tech names outweigh the weaker legacy agencies.
Centered on triple-B, with a bb tail of mid-tier agencies and ad-tech and two single-B names; no distressed names.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
10. Staffing & Recruitment
LOSER · Consensus rating bbb- · PD 39 bps · 3-yr ΔPD +56% · 16 of 38 screened names carry a consensus
What is in this segment, and why – Firms that place temporary and permanent workers and provide related HR and payroll services. Exposed where AI reduces demand for white-collar hiring — and for recruiters themselves — though payroll processors are more insulated.
How it is performing (aggregate) – Deteriorated ~56% and lost a notch to bbb-. The larger sample answers the single-name worry: the decline is broad-based, with several further names joining the downgrades, not just one collapse.
Split across single-A, triple-B and bb, with one distressed name in the c bucket; payroll processors anchor the top.
Consensus probability of default for the segment, monthly, May-2023 to May-2026.
Notes & Limitations
- Bank-sourced coverage. Many pure-play AI-loser equities, most white-collar recruiters, and large private call-center operators carry no consensus; their stress shows in equity, not here. 131 of the 337 screened names have no live consensus.
- Sample expansion. Outsourcing, staffing, education, content, advertising and IT services were broadened with global and private names. This lifted consensus coverage (e.g. outsourcing 8→12, staffing 11→16, IT services 19→29) and revised education’s three-year move from +131% to +58%.
- Small samples remain. Outsourcing and education still rest on twelve consensus names each; treat their aggregates as directional.
- Tail effects. The deterioration in IT services is still amplified by one distressed name; staffing’s, by contrast, is now broad-based.
- Entity choices. Where a listed parent had no consensus, the debt-bearing operating company was used instead. Each such instance is noted above.
- Source. Credit Benchmark Consensus Credit Ratings (CCR), CB21 / CB7 scales, May-2026 publication. Analytical summary for internal use, not investment advice or a rating-agency opinion.
Credit Benchmark provides ongoing consensus credit data, analytics, and research. For more
detailed analysis and custom data extracts, please contact us at [email protected] or visit creditbenchmark.com to learn more.