CITIC SEC: What signals do CDS spreads reveal amid the AI financing boom?

CITIC SEC released a research report stating that the scale of AI capital expenditure continues to increase, financing costs for related entities are increasingly high, and combined with changes in the macro environment in Q3, this has driven investors' attention to credit risk of AI technology companies and hedging demand to rise simultaneously, driving the CDS market for AI technology companies to show a trend of both trading volume and spreads rising. Although changes in market sentiment have pushed CDS spreads of related companies to widen broadly, this has not yet evolved into a comprehensive rise in industry-wide credit risk, and CDS spreads of major AI-related companies show tiered differentiation. Against the backdrop of AI infrastructure construction and bond financing still being in an expansion phase, the cross-sectional differentiation of CDS spreads and their dynamic evolution may become important indicators for observing the accumulation and transmission of credit risk in this round of the AI capital expenditure cycle. CITIC SEC's main views are as follows: AI capital expenditure expansion combined with rising financing costs continues to heat up demand for credit risk hedging among technology entities. Computing infrastructure investment pushes up financing demand, and the front-loading of capital expenditure alongside delayed cash flow realization exacerbates maturity mismatch, strengthening market attention to debt-servicing capacity and refinancing pressure. Since Q3, changes in the macro environment have increased the market's sensitivity to credit risk, and CDS trading activity and spreads have risen simultaneously. According to Bloomberg data, in Q3 2026 the average monthly trading volume of single-name CDS for AI-related technology entities increased by 156.6% QoQ, and CDS spreads of major technology entities widened broadly, continuing to rise since October, reflecting deepening market concerns about the realization of investment returns and financing sustainability. Bond financing in AI-related industries shows a misaligned distribution between the number of entities and issuance scale: the software services industry has more companies, while telecom operators have a larger financing volume. According to Bloomberg's scope, since 2023 there have been a total of 1,490 effective bond issuers in AI-related industries. Among them, the software and services industry had 659 issuers, corresponding to a bond issuance scale of USD 494.23 billion since 2023. In contrast, the wired and wireless telecommunications services industry had a total of 243 issuers, with cumulative bond issuance reaching USD 891.55 billion. In terms of rating distribution, there were 225 entities with effective ratings from the three major international rating agencies, corresponding to 66.1% of the bond issuance scale in related industries since 2023. Among them, there were 128 investment-grade entities, corresponding to 50.5% of the total sample's bond issuance scale, with the BBB rating range accounting for the highest share of issuance scale at 22.7%. CDS spreads of AI technology entities show tiered differentiation, with credit repricing of highly leveraged companies particularly prominent over the past three months. According to Bloomberg's scope, as of October 20, 2026, CoreWeave's 5-year CDS spread reached 1,120.7bps, widening by 270.9bps over the past three months, driven by its high-leverage model of expanding computing power and persistent pressure on its own free cash flow. Oracle, whose credit rating had previously been lowered, also saw its 5-year CDS spread rise to 329.1bps, widening by 115.3bps over the past three months. In contrast, Microsoft and Apple, relying on ample cash flow and stable leverage, had average CDS spreads of 51.5bps and 42.9bps respectively over the past three months, with an overall differentiation pattern that is evident. Compared with historical cycles, bond financing in industries related to this round of the capital expenditure cycle shows the characteristics of large scale and long tenor. According to Bloomberg data, annual bond issuance in AI-related industries rose from USD 305.91 billion in 2023 to USD 779.89 billion in 2025, and further reached USD 964.67 billion in 2026 as of October 20, significantly higher than the annual bond financing scale of the internet telecom and real estate cycles. In terms of maturity structure, the trend toward longer-tenor bond issuance is evident, with the share of bonds issued with maturities above 10.33 years rising from 14.4% in 2024 to 25.0% in 2026, among which the share of ultra-long-term bonds above 20.33 years rose from 8.4% to 13.8%, reflecting that long-term capital investment is further extending corporate financing duration. The direction of credit evolution in the subsequent AI capital expenditure cycle depends on whether the gap between capital investment and cash flow realization can converge. If related entities subsequently recover cash flow through order fulfillment to cover debt principal and interest and fixed expenditures, leverage pressure is expected to gradually ease. Looking ahead, compared with single-bond spreads, which are susceptible to differences in tenor, liquidity, and covenants, single-name CDS uses a specific issuer as the reference entity and is more suitable for reflecting the overall credit risk pricing of the entity, though it is still subject to hedging demand and liquidity disturbances. Going forward, it is recommended to combine order fulfillment and cash flow realization, continuously track CDS spreads and differentiation among entities, and identify the accumulation and transmission of credit risk. Risk factors: AI demand and commercialization progress falling short of expectations; overseas monetary policy exceeding expectations; impact of individual credit risk events.