US Software Sector Quarterly: AI Displacement Fears Drive Indiscriminate Selloff as Refinancing Risk Moves Into Focus

What followed was a relentless sequence of product announcements from the major foundation model developers, each interpreted by the market as incrementally disruptive to a different corner of the software universe. Anthropic set the pace, releasing new features and capabilities at a pace that left little time for the market to digest any single announcement before the next arrived. Volatility became the defining characteristic of the space, with entire subsectors swinging materially on individual model releases, leaked benchmarks and conference commentary.

In late March, a configuration error in Anthropic’s content management system inadvertently exposed thousands of internal documents, revealing the existence of a forthcoming model called Claude Mythos, which the company confirmed and described as a step change in capability and the most capable model it has built to date, with meaningful advances in reasoning, coding and cybersecurity. Anthropic’s own draft materials warned that Mythos was far ahead of any other model in cybersecurity capabilities and described it as presaging an upcoming wave of models capable of exploiting software vulnerabilities in ways that would far outpace the efforts of defenders, causing cybersecurity companies to sell off sharply, only to snap back about two weeks later as Anthropic announced Project Glasswing, a coordinated defensive security initiative granting Mythos preview access to roughly 50 organizations including Amazon, Apple, Microsoft, Google, CrowdStrike and Palo Alto Networks, reframing the model not as an existential threat to cybersecurity incumbents but as a capability multiplier that would flow through them.

Overall, software has sold off significantly, as shown below by the performance of IGV, a broad software-focused exchange-traded fund versus the performance of the Nasdaq 100 and the S&P 500 since the start of the year. As can be seen below by the performance of CIBR, a cybersecurity-focused ETF, cybersecurity has outperformed broader software but is still lagging overall tech performance. WCLD, a cloud computing-based ETF, also shown below, illustrates the intense pressure that cloud computing has faced.
 

Software credits have not been spared from the blanketed selloff. The chart below charts the weighted average performance, as measured by yield to worst, for a batch of the 50-largest high-yield software bonds (excluding bonds issued after the beginning date) versus the performance of the ICE BofA high yield index.
 

The chart below shows weighted-average price performance across the majority of high-yield software bonds over $100 million in principal over specified time periods versus the performance of the ICE BofA high-yield index and rating level indexes during the same time periods. As shown through the chart’s data, software spreads widened significantly more than the broader high-yield indexes over the trailing three months, though they have performed closer to in line with the broader indexes performance over the past month as software credits have partially recovered from a late-March trough.
 

Further, the chart below shows instrument-level pricing information across the same universe of software bonds described above. Specifically, the three-month change in spread shows the broad spread widening that applied indiscriminately to the majority of the software. More recently, we have seen credits that were wrongfully sold off meaningfully recover from the prior selloff.
 

A review of software and data company earnings calls from the most recently completed reporting cycle reveals a meaningful shift in the tenor of AI-related management commentary relative to the prior two quarters. Through the second half of 2025, AI risk was treated almost universally as theoretical. Seat compression concerns were actively rebutted by management teams at companies including ServiceNow and HubSpot, and the dominant framing across the space was AI as an upsell vector or platform differentiation tool rather than a demand headwind.

The most recent period tells a materially different story. Several management teams are now doing something they largely avoided in prior calls, embedding AI-attributable pressure directly into forward guidance. PagerDuty became one of the more direct examples, explicitly quantifying seat compression as a live headwind in prepared remarks and guiding to essentially flat annual recurring revenue growth as a result, a disclosure that would have been unusual six months earlier.

Beyond seat compression, Asana and HubSpot both named AI-driven search disruption as a structural top-of-funnel headwind affecting small- and medium-size business demand generation, while VTEX coined the phrase “AI wait-and-see effect” to describe enterprise customers deliberately elongating platform decision timelines amid uncertainty about AI’s implications for software categories, characterizing it as a marketwide phenomenon rather than a company-specific issue.

Critically, this pressure is not uniform. Infrastructure, cybersecurity and data platform companies continue to frame AI predominantly as a net positive, while the deterioration in forward-looking commentary is concentrated precisely in the seat-based software-as-a-service and search-dependent businesses most exposed to the displacement mechanisms described in this piece. That bifurcation, which has sharpened materially over the past two quarters, is the central analytical lens through which the relative value opportunities discussed below should be evaluated.

Octus has covered the space extensively in the wake of the extreme volatility over the past few months. In February, Octus published an analysis addressing the initial selloff and later followed with an analysis that dove into how specific issuers may or may not be affected by AI.

Most recently, Octus published a U.S. Software Analysis on April 16 examining AI displacement risk across six high-yield software and data issuers, including ZoomInfo, Clarivate, RingCentral, Twilio, Fair Isaac and Consensus Cloud, arguing that the broad software credit selloff has been applied with insufficient precision across names with meaningfully different competitive profiles. The analysis is organized around three displacement mechanisms: seat compression exposure, proprietary versus publicly available data, and regulatory and institutional entrenchment.

On seat compression, the piece argues that more consequential near-term risk for seat-based vendors is not direct product obsolescence but AI-driven enterprise head-count reductions that compress the unit of monetization at contract renewal without requiring any active platform replacement decision, a dynamic most acute for ZoomInfo given its concentration in sales head count, more gradual for RingCentral given broader distribution across knowledge-worker populations and structurally inverted for Twilio, whose consumption-based model means AI agent proliferation drives volume rather than compressing it.

On the proprietary-versus-publicly-available data question, ZoomInfo is identified as more exposed given that a portion of its core contact data can already be loosely approximated using foundation models, while Clarivate is assessed as more defensible at the data layer given the nonreconstructible expert curation underlying its Derwent patent records and Web of Science citation index, though the analytics and workflow products built above that data face growing replication risk within the bond maturity horizon.

At the lower end of the displacement spectrum, Fair Isaac and Consensus Cloud are characterized as occupying a distinct risk category where AI can replicate the technical function but cannot replicate the regulatory approval, institutional adoption and legal authority that make each an entrenched industry standard, with the more relevant monitoring risks for both credits being competitive and secular rather than technological.
 

The widespread and relatively indiscriminate nature of the software selloff has created what appear on the surface to be constructive entry points across a number of credits. Spreads have widened materially across names with meaningfully different competitive profiles, and in several cases current fundamental metrics do not obviously justify the move. Evaluating relative value on near-term fundamentals alone, however, understates the analytical challenge, as the displacement risks driving the selloff are in many cases years away from full materialization. The more relevant question for credit investors is not whether a business can service its debt today but whether capital markets will remain open to it on reasonable terms when the maturity wall arrives.

That question is complicated further by the limited options available to management teams trying to address refinancing risk proactively. For software companies whose equity is under pressure from the same displacement narrative driving spread widening, redirecting free cash flow toward early debt retirement signals precisely the concern they are trying to hedge against. The result is that many management teams remain structurally biased toward buybacks and other methods of shareholder compensation despite approaching maturity walls, leaving refinancing contingent on capital markets access at a point when the displacement narrative may be more developed. With this in mind, we believe investors may gravitate toward credits with business models that are perceived to be relatively protected from AI displacement and balance sheet flexibility sufficient to invest in their own AI capabilities.

We continue to view Fair Isaac as relatively well placed given the regulatory moat that makes it one of the more insulated credits to AI-native entrants. Its unsecured notes due 2033 and 2034 indicate spreads moderately wide of the indexes after the 2033 notes’ spread widened approximately 80 bps to 220 bps since late January, and fundamentals remain strong with adjusted EBITDA margins in the mid-to-high 50% range and net leverage of approximately 2.6x.

Its most significant risk remains the emergence of a formidable competitor in VantageScore following the Federal Housing Finance Agency’s 2025 decision permitting lenders to choose between VantageScore and Classic FICO when selling loans to Fannie Mae and Freddie Mac, ending decades of mandatory FICO use in the conforming mortgage market. VantageScore’s structural distribution advantage through its joint ownership by the three major credit bureaus adds a further competitive dimension, though the operational burden of transitioning underwriting models and revalidating loan pricing against a new scoring standard limits the pace at which lender adoption can realistically accelerate, preserving meaningful defensibility within the bond maturity horizon.

Consensus Cloud presents a similar displacement profile, with its 6.5% unsecured notes due 2028 at 288 bps widening less than 5 bps over the past 90 days, reflecting the market’s assessment that a business whose core product is mandated by healthcare compliance requirements faces a categorically different displacement risk than the seat-based and data-dependent names elsewhere in this table. The more relevant monitoring risk is secular fax volume decline rather than any AI-driven displacement scenario, though with the notes already indicating modestly inside the single-B index and net leverage of approximately 2.6x, the spread reflects the low displacement thesis fairly fully at current levels.

CrowdStrike’s 3% unsecured notes due 2029 at 90 bps tell a similar story from a different direction. The notes have been essentially unmoved over the past 90 days, widening less than 1 bp against a backdrop of significant broader software spread widening, reflecting the market’s read that the AI-driven threat landscape is a net demand tailwind for endpoint security rather than a displacement risk, a dynamic consistent with the Project Glasswing framing discussed in the introduction.

The 670-bps spread on ZoomInfo’s 3.875% unsecured notes due 2029 reflects considerable AI displacement risk, and we remain cautious despite manageable near-term metrics.

Management repurchased $407 million of shares in 2025 against $390 million of free cash flow generated and approved a new $1 billion buyback authorization alongside fourth-quarter results, leaving debt repayment at the 2029 maturity fully contingent on capital markets access. For the notes to represent compelling value, investors need conviction both that the business will stabilize sufficiently to support refinancing on reasonable terms and that management will pivot its capital allocation posture before the maturity window narrows – two conditions that are difficult to achieve simultaneously in the current environment.

Clarivate’s capital structure presents a related but distinct set of concerns. The secured 2028s at 244-bps spread and unsecured 2029s at 560-bps spread represent a 316-bps step-up for one additional year of maturity and structural subordination, with net leverage of 4.1x through the unsecured. Although the core data moat remains defensible at the data layer, the analytics and workflow products built above that data face growing replication risk within the bond maturity horizon, and the company’s elevated leverage makes the unsecured notes difficult to consider relative to ZoomInfo at 150 bps of incremental spread with meaningfully lower leverage.

RingCentral’s 8.5% unsecured notes due 2030 at 330 bps, approximately 160-bps wide of the BB index, appear to offer reasonable compensation relative to the underlying credit profile and stand out within the table on a risk-adjusted basis. Net leverage of 1.7x and consistent free cash flow generation provide a substantial cushion, and although the seat-based model carries structural exposure to gradual enterprise head-count compression, retention above 99% suggests limited near-term materialization of that risk. Critically, RingCentral does not face the capital allocation tension that complicates ZoomInfo’s investment rationale.

The signaling trap is less substantial for a business whose equity narrative is less directly defined by the displacement debate, and the leverage profile provides a meaningful runway to absorb incremental pressure before refinancing risk becomes relevant. The roughly 200-bps spread differential to Twilio’s 2031 notes looks difficult to justify on fundamentals given that platform displacement risk is low, the AI add-on layer is early but growing, and the leverage profile is only modestly wider. For investors seeking software credit exposure with a more straightforward credit case than the higher-yielding names in this table, the 2030 notes present a profile that warrants consideration in the current environment.

Ziff Davis’ 4.625% unsecured notes due 2030 at 192 bps present a credit profile that is more nuanced than the spread level suggests. The company’s balance sheet is conservatively positioned at approximately 0.5x net leverage, and the March announcement of the $1.2 billion sale of its connectivity division to Accenture, which included Ookla, Speedtest and Downdetector, materially reinforces that position.

The more relevant credit concern is the underlying organic revenue trajectory, which has been declining in the low single digits for several years, driven in part by structural pressure on the tech and shopping segment as AI-driven search disruption erodes the high-intent affiliate commerce traffic that segment depends on, a dynamic management has explicitly guided to persist through 2026. Further asset sales could continue to surface embedded value and improve the balance sheet, but they also progressively concentrate the remaining business in segments facing more acute structural headwinds. At 192 bps, the notes offer limited cushion for a business with negative organic revenue growth and a revenue mix that is becoming less diversified with each divestiture.
 

ZoomInfo notes declined more than 11 points in the past quarter to 83.7, with the market viewing the company as one of the more fundamentally exposed businesses to potential AI-displacement risk. We noted in our recently published analysis of a group of high-yield software issuers that the company faces pressure on both the seat-compression displacement framework, given that it is primarily tied to seat-based pricing, and the replication framework, as a portion of its core data offering can already be loosely replicated with the use of large language models.

The chart below shows spread performance for ZoomInfo’s 2029 unsecured notes versus the ICE BofA single-B index spread performance over the last three months:
 

Virtusa’s 7.125% unsecured notes due 2028 declined more than 25 points over the past quarter to just over 72. The company, a private equity-backed IT services and outsourcing provider with delivery centers concentrated in India and Sri Lanka, likely faces direct structural headwinds from AI-assisted coding tools that are compressing enterprise demand for offshore engineer-hours. Its capital structure, which Moody’s Ratings recently cited as being more than 5x levered, leaves the company with limited margin for top-line or margin deterioration ahead of the notes maturity in 2028.
 

Both series of CDK’s (Central Parent LLC) secured notes due 2029 sold off significantly over the past few months, with its 8% notes declining about 25 points over the last three months to 64 and its 7.25% note declining about 22 points to 63. The selloff was magnified by broad sector-level AI displacement fears, though Octus has pointed out that its current trading levels are anchored in company-specific risks and weakening fundamentals. Its competitive position has deteriorated in the wake of its June 2024 cyberattack incident, which further calls into question whether it can sustain leverage over 9x, excluding aggressive pro forma addbacks.
 

Despite the majority of software credits selling off materially during the past few months, Sabre’s secured notes, with maturities ranging between 2029 and 2030, rallied on the back of aggressive balance sheet deleveraging, including the $1.1 billion sale of Hospitality Solutions to TPG, with net proceeds used almost entirely for debt repayment, a series of refinancings that eliminated near-term maturities. These capital structure actions were reinforced by improving operating fundamentals and increased liquidity, which dramatically reduced the probability of a distressed outcome that bond markets had previously been pricing in.

Octus recently published an analysis forecasting that Sabre Corp. will generate negative levered free cash flow through 2027 and face significant refinancing risks for its $4.1 billion debt maturity wall beginning in 2029 due to a demand-side shock from the Middle East conflict and structural pressures on its global distribution system.
 

Select sub-investment-grade software credits with upcoming maturities are shown in the chart below:
 

The maturity for Alludo’s $450 million first lien term loan is approaching in July 2026, though the loan has continued to sell off over the past few months, now indicating at a bid price of about 86, down from the mid-90s in late 2025.

Octus reported in November that the company was exploring refinancing options with its sponsor, KKR, and suggested that it could still opt for a refinancing in the broadly syndicated market on the back of improving performance. The company announced in February that it will separate its business into two independent companies, with Vector Capital acquiring the creativity and productivity software portfolio and KKR retaining ownership of Parallels.

The transaction is expected to close in May 2026, ahead of the July maturity. Although financial terms were not disclosed, we would expect the deal to necessitate a refinancing of the existing term loan, which would resolve the maturity ahead of its July deadline. The loan has since recovered from its lows, though it continues to trade at a modest discount to par as investors price residual close risk ahead of the anticipated May completion.
 

Sophos has nearly $2.6 billion U.S. dollar-equivalent of secured debt maturing in 2027, inclusive of more than $2 billion of term loans due March 2027, though the loans indicate near par at a bid of about 97 as the market prices in a likely refinancing. Octus reported that the company is expected to gauge interest from both the private credit and broadly syndicated markets. Moody’s downgraded the company’s corporate rating to B3 from B2 in March, citing weaker-than-expected operating performance over the past few quarters and heightened refinancing risk, though it forecasts that the company will be able to refinance its credit facilities with a moderate increase in its cost of debt.
 

Primary market activity remained relatively frozen for the majority of the quarter as the steep software selloff, paired with geopolitical tension due to the Iran war, left investors with a skittish risk appetite. Gaming developer Electronic Arts, however, did manage to push through its unparalleled $18 billion leveraged buyout financing package in late March.

The broader March primary market was dominated by its landmark $55 billion take-private, the largest LBO in history. To fund the transaction, a consortium led by PIF and Silver Lake raised $18 billion in debt across a multitranche structure that transformed EA from a near-unlevered investment-grade credit into a highly levered sub-investment-grade issuer (B1/BB-).

The financing, which included $11.375 in term loans split between U.S. dollar and euro tranches, $4.125 billion of secured notes split between U.S. dollar and euro tranches, and $2.5 billion of unsecured notes, was met with exceptional demand, closing more than 2x oversubscribed as investors sought the yield premium offered by EA’s “must-have” sports licensing moats.

The company’s pro forma capital structure is shown below:
 

The deal significantly reshapes EA’s credit profile, increasing annual cash interest to an estimated $1.3 billion from $56 million. Although the company marketed the deal at 5.3x leverage, this figure relies on $709 million of unrealized cost savings and R&D addbacks; stripping these out reveals a reported LTM gross leverage of 6.8x.

Critically, the LBO was sized at peak-cycle EBITDA following the landmark “Battlefield 6” launch, leaving through-cycle leverage materially higher during “gap years” between major non-annual releases. Despite these concerns and the broader selloff in software-adjacent credits, EA’s unsecured notes priced at a significant premium to the single-B index, providing what many investors view as appropriate compensation for the structural subordination of its international revenue and the inherent volatility of the AAA release calendar.

The secured and unsecured notes have performed strongly since pricing, as shown below:
 

Fair Isaac priced $1 billion of Ba1/BB+ rated 6.25% of unsecured notes due 2034 on March 11 in a standard refinancing, with proceeds used to pay down its revolver and redeem in full its 5.25% unsecured note due May 2026.

CACI International priced a $500 million add-on to 6.375% senior unsecured notes due 2033 at 102.875, indicating a yield to worst of about 5.6% on Feb. 26 alongside $800 million of incremental term loans at SOFR+1.75% to finance its $2.6 billion acquisition of ARKA Group.

Octus reported on April 15 that Yahoo! was premarketing a refinancing package, including a five-year $1.1 billion loan and a five-year $500 million high-yield bond, with initial price talk coming at SOFR+550 bps with 98.5 OID and the high-9% area, respectively. The launch of the deal is expected in the coming weeks. Octus’ private team published a primary analysis for the deal on April 13.
 

Notable software credit rating downgrades and upgrades from the past three months are shown below: