INVESTMENT NOTE NO.152026.10.22Credit · AI Infrastructure

AI Credit Concerns:
Where Are They Showing Up in Korea?

Tracing AI Credit Stress from CoreWeave to Korea Across Five Market Indicators

Three credit research notes published on October 20–21 reached different conclusions about how far the problems surrounding CoreWeave might spread. We examine those assessments alongside SK Securities' capital-expenditure and cash-flow forecasts, moving from CoreWeave through lower-rated borrowers, Big Tech and the wider US credit market to South Korea. Credit risk has been repriced sharply for CoreWeave and some lower-rated borrowers, and high-yield spreads have widened. But we cannot establish how much AI-related issuers contributed to the broader index move. In Korea, domestic factors complicate the credit-market picture. Shares of Korean data-center suppliers have fallen sharply, but an AI-driven spillover into Korean corporate credit has yet to be identified.

CoreWeave 5Y CDS: 1,100bp+Mid-October · Hanwha Investment & Securities¹
US High-Yield Spread: 333bpOct 20 · 263bp at end-August²
AAA Public-Sector Bonds, 3Y: +12.7bpYTD · BBB0 +14.4bp (our calculation)³
13 Suppliers: Median: −21.1%Oct 2–19 · All listed common shares −3.9%⁴

Figure 1 · Five Observation Points

Five Observation Points, from CoreWeave to Korea

▲ Schematic only. Circle sizes and spacing are not to scale and indicate neither the magnitude of risk nor the sequence of contagion. Figures at right are observations at each point (dates and sources in the report). Values read from brokerage charts are visual estimates.^{1, 2, 3, 4, 5}

AT A GLANCE · What the Data Show

One epicenter, three interpretations of the credit risk. Hanwha's Park Jun-young sees a CoreWeave-specific problem; Park Se-yeon at the same firm focuses on lower-rated borrowers and private credit; and Hana Securities' CrediVille team sees potential repercussions along the AI value chain. A separate SK Securities report does not assess credit risk but provides useful cash-flow context: its chart suggests negative consensus free cash flow (FCF) for six companies combined in 2026 and 2027 (visually estimated).
Credit protection became more expensive first for debt-funded infrastructure providers. Oracle's CDS reached a record high, as did those of several Big Tech firms. Yet Big Tech premiums, at an estimated 50–105bp, remained less than a tenth of CoreWeave's (based on chart readings). The US high-yield spread widened from 263bp on August 31 to 333bp on October 20, but neither CoreWeave's nor other AI issuers' contribution to that change has been established. Redemption pressure in private credit predates the CoreWeave episode.
Consensus forecasts put the six companies' combined CapEx above operating cash generation. Meta recorded negative FCF of $2.1bn in the third quarter, though its January–September total remained positive at $11.0bn. Long-term debt increased by $22.2bn over nine months. If the 2026–27 projections materialize, the companies must fund any shortfall through existing cash or outside financing, including bonds, joint ventures and private credit (our interpretation). The more they rely on outside funding, the more credit-market pricing can affect the pace of investment.
Korean three-year credit spreads have widened by roughly 13–19bp across ratings. In percentage terms, the increase exceeds 50% for some top-rated bonds but is only about 3% for BBB0, largely because the starting spreads differed (our calculation). Rate hikes and bank funding dynamics overlap with the spread moves, making an AI-specific contribution difficult to isolate. At its October 22 meeting, the Bank of Korea cited concerns about AI-related corporate bond supply as one factor behind rising long-term yields in international markets.
The clearest Korean price move has been in data-center suppliers' shares. The median return for 13 Korean data-center supply-chain stocks was −21.1% from October 2 to 19, more than five times the −3.9% median for listed common stocks. But this is a thematically selected group: the semiconductor cycle and valuation repricing may also explain the decline, so the effect of credit fears cannot be isolated. When the KOSPI fell another 3.52% on October 22, doubts about AI investment returns, Middle East tensions and semiconductor trading flows overlapped. Over a longer horizon, the credit quality of end customers placing on-site power-generation orders may matter most (deliveries in 2027–28).

After CoreWeave's five-year credit default swap (CDS) premium exceeded 1,100 basis points (bp) in mid-October, investors began asking how far the problem had spread.¹ We follow five sets of market data, working outwards from the epicenter and ending in South Korea. CoreWeave's contracted backlog and debt structure were discussed in NO.14.⁶

Three Credit Assessments, One Cash-Flow Perspective

Three credit-focused reports dated October 20–21 offer differing views of how far CoreWeave's difficulties could reach. A fourth examines the underlying cash-flow picture. The assessments below are attributed to their original authors.

Hanwha Investment & Securities · Park Jun-young · Semiconductors · Oct 21

A Problem Specific to CoreWeave

Single issuer

In this view, CoreWeave's model—borrowing heavily to acquire GPUs and data-center capacity, then recovering the investment through rentals—makes it unusually sensitive to interest rates and financing conditions. The argument rests on divergent CDS pricing and differences in investment scale. The chart puts combined 2026 CapEx for neoclouds CoreWeave and Nebius at roughly $60bn, versus about $680bn for four hyperscalers (visually estimated).¹

Condition: provided credit-market turmoil does not lead hyperscalers to cut investment.

Hanwha Investment & Securities · Park Se-yeon · ESG · Oct 20

A Problem for Debt-Funded Customers and Their Lenders

Part of the industry

Her assessment is that demand has not changed; the cost of financing has risen for businesses borrowing to build capacity. Spreads have widened most for lower-rated capacity lessees and their capital providers, rather than AA-rated Big Tech firms.⁴

Watch point: demand is concentrated among a handful of AI laboratories. If they lose financing, cancelled orders could begin to signal a genuine demand slowdown.

Hana Securities · Kim Sang-man · CrediVille · Oct 21

Potential Spillovers Across the AI Value Chain

Possible industry-wide impact

Hana suggests that BB−-rated CoreWeave is likely a major driver of the widening US high-yield spread and that the consequences may not remain confined to speculative-grade issuers. Its report also includes a chart showing a sharp second-half rise in spreads on 30-year US AI corporate bonds, while spreads excluding AI issuers were broadly flat (visually assessed).³

Limit: no stand-alone CoreWeave credit figures were supplied; the claim relies in part on its rating.

SK Securities · Han Dong-hee · Semiconductors · Oct 21

The Cash-Flow Backdrop

Cash-flow evidence

SK argues that high penalties for breaching long-term memory supply agreements (LTAs), together with rising GPU rental prices, make this an attractive entry point for memory stocks. Its Bloomberg consensus charts imply combined FCF of roughly −$30bn in 2026 and −$120bn in 2027 for six companies (visual estimates).⁷

Caution: the chart aggregates six companies and does not identify which account for the deficit.

What Would Count as a Spillover?Evidence at each observation point and this note's assessment; classifications are our own.
Station	Measure	Evidence	Assessment
ST.01 CoreWeave	Issuer-specific credit stress	CoreWeave CDS and bond yields	Observed
ST.02 Borrowers / lenders	Repricing of similar borrowers	Record Oracle CDS · private-credit redemption pressure	Observed · redemption pressure began earlier
ST.03 Big Tech	Transmission to stronger borrowers	Record CDS · bond spreads · FCF	Same direction · much lower levels
ST.04 US credit	Market-wide transmission	High-yield index spread	Widening confirmed · AI contribution unknown
ST.05 Korean bonds	Transmission to domestic credit	Credit spreads by rating	Cannot isolate · domestic factors overlap
ST.05 Korean equities	Repricing of suppliers	Returns of 13 supply-chain stocks	Decline confirmed · causes not isolated

ST.01Epicenter · CoreWeaveClear signal

Why Did CoreWeave Come Under Pressure First?

Credit markets first repriced the risk of a heavily indebted company building data-center capacity.

1,100bp+Five-year CDS · mid-October (Hanwha)

16%+Bond yield · briefly on Oct 15 (Hanwha / Reuters)

BB−Credit rating (as cited by Hana; rating agency unspecified)

Approx. 25%Q2 net interest expense / revenue (our calculation)

CoreWeave uses substantial borrowing to secure GPUs and data-center infrastructure upfront, rents that capacity to customers, and aims to recoup its investment over time. Hanwha analyst Park Jun-young argues that the structure leaves the company particularly exposed to higher rates and tighter financing conditions.¹ Its five-year CDS premium exceeded 1,100bp in mid-October, while the yield on one of its bonds briefly topped 16% on October 15. Hanwha's Park Se-yeon refers to a bond maturing in October 2031; Reuters cites a 2032 maturity.^{1, 4, 5} The 16% figure is a secondary-market yield on an already issued bond—not its coupon rate or the average interest rate on all of CoreWeave's debt. But refinancing or raising comparable new debt at that yield would be costly. Reuters said a CDS premium of 1,000bp implied roughly a 63% cumulative probability of default over five years, compared with approximately 30% a year earlier. This is a market-implied, model-dependent estimate that assumes a recovery rate, not a forecast of the company's actual default probability.⁵ CoreWeave's remaining performance obligations (RPO)—contracted revenue not yet recognized—totaled $103.7bn at the end of June. Only 41% was expected to be recognized as revenue within the following 24 months. Second-quarter net interest expense equaled approximately 25% of revenue (our calculation).⁸

Figure 2

CoreWeave Above 1,000bp; Big Tech at Roughly 50–105bp Even at Record Highs

Five-year CDS premiums · observation dates and sources differ by issuer

CoreWeave: Hanwha's statement that five-year CDS exceeded 1,100bp in mid-October. SoftBank (up 24bp in one day) and Oracle: Reuters, October 15; Oracle's record 288bp is from S&P Global Market Intelligence. Big Tech: approximate 50–105bp endpoints read from Hanwha's chart, alongside Reuters' report of record-high levels. Exact company-by-company readings were not published.^{1, 4, 5}

ST.02Lower-Rated Borrowers · Capital ProvidersClear signal

The Cost of Credit for Borrowers—and Redemption Pressure on Lenders

Comparable borrowers and their funders face pressure, but redemptions in private credit were already elevated before the CoreWeave episode.

288bpOracle five-year CDS · record Oct 15 · rated BBB−

39%Blue Owl OTIC Q3 redemption requests · Q1 40.4%, Q2 38.1% · tender offer capped at 5%

$27.3bnBonds issued by Meta's Hyperion joint venture

Approx. $800bnEstimated private-credit contribution through 2028 (Morgan Stanley)

Hanwha analyst Park Se-yeon argues that what has changed is not underlying demand, but the cost of financing for companies borrowing to build infrastructure. Credit spreads have widened most for borrowers and lessees with weaker ratings than AA-rated Big Tech, and for the firms supplying their capital.⁴ Oracle was downgraded in July to BBB−, the lowest investment-grade rating on S&P's scale. Its five-year CDS premium reached a record 288bp on October 15.^{4, 5}

The funding side has also evolved. Meta's Hyperion project in Louisiana is being built by a joint venture in which a Blue Owl affiliate holds an 80% interest; the venture issued $27.3bn of bonds. In July 2025, Morgan Stanley estimated that private credit would provide around $800bn of the $1.5tn in external funding required for global data-center investment by 2028.⁴ Private-credit funds were already experiencing redemption pressure. At Blue Owl Technology Income Corp. (OTIC), redemption requests represented 39% of outstanding shares in the third quarter, broadly unchanged from 40.4% in Q1 and 38.1% in Q2. The company said most requests were resubmissions of previously unfilled tenders. Its quarterly repurchase offer remained capped at 5% of outstanding shares for a third consecutive quarter. Blue Owl had offered quarterly tenders of that size in 2025 as well; this was not a newly imposed emergency restriction.^{4, 9, 10} The announcement came on October 2, before CoreWeave's CDS surge. Rather than a fresh shock caused by CoreWeave, OTIC illustrates a pre-existing liquidity vulnerability worth monitoring as AI credit concerns grow.

On October 13, the IMF identified 14 circular-financing arrangements in the AI industry since 2023, linking approximately $300bn in equity investments and $650bn in debt with $1.4tn in future revenue commitments. It warned that stress at key recipients of investment could lower expected revenues and investment valuations, in turn weakening the value of debt accumulated in those deals.¹¹

ST.03Big TechPartial signal

Is Big Tech Credit Still Sound? What Six Companies' Cash Flow Reveals

Heavy investment is tightening cash flow, but Big Tech balance sheets differ substantially from CoreWeave's.

Record highsAmazon, Alphabet, Meta, Nvidia CDS · Oct 15 (Reuters)

+30bpBig Tech bond spreads, YTD · broader investment grade +2bp (August)

Approx. −$120bnConsensus FCF, six firms combined · 2027 (chart estimate)

The cost of credit protection for Big Tech has risen too, but the absolute premiums remain far below CoreWeave's. Hanwha's Park Jun-young argues that even after the increase, Big Tech CDS prices remain distinct from those of neocloud providers such as CoreWeave. Ample liquidity and strong credit ratings give hyperscalers greater financial flexibility.¹ The direction of travel is nevertheless similar. On October 15, CDS premiums for Amazon, Alphabet, Meta and Nvidia were trading at record levels, although they had risen much more slowly than those for Oracle or CoreWeave.⁵ By August, spreads on Big Tech corporate bonds had already widened by roughly 30bp from the start of the year, against just 2bp for the broader investment-grade market (J.P. Morgan Asset Management, cited by Hanwha).⁴ The September Federal Reserve minutes recorded a report from the open-market desk that spreads on hyperscaler bonds used to finance AI infrastructure remained wide, reflecting heavy issuance volumes and relatively long maturities. Fed staff noted that, after widening substantially from low levels earlier in the year, those spreads had narrowed slightly between the July and September meetings.¹²

When Capital Spending Exceeds Operating Cash Flow

A Bloomberg consensus chart reproduced by SK Securities suggests that combined capital expenditure for Google, Amazon, Microsoft, Meta, Oracle and CoreWeave could rise from roughly $380bn in 2025 to $1.12tn in 2027. Aggregate FCF is estimated at approximately −$30bn in 2026 and −$120bn in 2027. These are approximate readings from a published chart, not raw consensus data.⁷ If those forecasts prove accurate, the companies would need to bridge the aggregate funding gap over those two years by drawing on cash reserves or raising funds externally through bonds, joint ventures, private credit and similar channels (our interpretation). The Hyperion joint-venture bond in ST.02 is one example of such external financing.

Figure 3

Consensus FCF for Six Firms Turns Negative in 2026–27

Google, Amazon, Microsoft, Meta, Oracle and CoreWeave combined · approximate endpoints in Bloomberg consensus charts (Sep–Oct 2026), visually read from SK Securities

Values are visual estimates (≈) read from the endpoints of annual consensus series in SK Securities' chart (page 5). The six-company aggregate does not reveal individual contributions. Figures for 2026–28 are forecasts, not reported results.⁷

Oracle and CoreWeave may account for a substantial share of the negative FCF (our hypothesis). Hanwha's Park Jun-young also presents a chart in which projected 2027 CapEx for four hyperscalers exceeds 2026 levels (visual estimate).¹

Meta Q3 results · released after the US close on Oct 21 (early Oct 22 KST) · includes follow-up analyst call

Meta's Q3 FCF Was −$2.1bn; Nine-Month FCF Remained +$11.0bn

−$2.1bnQ3 FCF · +$10.6bn a year earlier · +$11.0bn Jan–Sep

$34.9bnQ3 CapEx (including finance lease principal) · operating cash flow $32.8bn

$80.9bnLong-term debt at end-Sep · $58.7bn at end-2025

$135–145bn2026 CapEx guidance · floor raised from $130bn

Meta spent $34.9bn on CapEx in Q3, including finance lease principal repayments, exceeding $32.8bn in cash generated from operations. FCF was −$2.1bn, versus +$10.6bn a year earlier. Importantly, FCF remained positive at approximately $11.0bn for the first nine months: an annual funding deficit is not established. Long-term debt rose from $58.7bn at end-2025 to $80.9bn at end-September, an increase of $22.2bn (our calculation). The $24.9bn of long-term debt issued during 2026 was all raised in the first half; there was no issuance in Q3 and no share repurchases in January–September. At end-September, cash and marketable securities totaled $82.45bn against $83.7bn of debt, according to the CFO's prepared remarks.^{13, 14}

Meta lifted the lower end of its 2026 CapEx guidance to $135bn–$145bn from $130bn–$145bn. It gave no numeric figure for 2027, saying only that it expects to keep investing aggressively in its core business and AI initiatives. The board also approved a new non-voting Class C share category, scheduled to take effect on January 19, 2027. The first stated rationale was to provide flexibility for raising capital to support substantial investment during the AI transition. Existing holders are to receive one Class C share for each share held as a stock dividend; that distribution does not, by itself, bring cash into the company.¹³ In a follow-up call, CFO Susan Li stressed that the filing's reference to 'hundreds of billions of dollars' was a theoretical maximum calculated to identify the amount of equity financing that would threaten founder control—not a statement of Meta's current financing needs or plans. She described compute capacity as still constrained, with demand for capacity likely to rise into next year, and said near-term needs would be supplemented through third-party cloud contracts.¹⁴ Revenue of $64.6bn and EPS of $7.40 excluding litigation expenses exceeded market expectations of $63.37bn and $5.74, respectively. The shares rose approximately 3% in after-hours trading immediately after the release.¹⁵

The six-company annual consensus chart from SK Securities and one quarter of Meta's reported results measure different things. Still, Meta's actual Q3 numbers show a period in which investment outlays exceeded operating cash generation. Debt has also increased over the year, and the new share structure, although not currently intended for fundraising, could provide future equity-financing flexibility. In our assessment, this points to more aggressive investment alongside a broader set of funding options; one negative FCF quarter is not, on its own, evidence of financial distress. Of Park Se-yeon's three questions, continued investment plans were confirmed. Neither the earnings release nor the follow-up call separately specified whether higher spending plans were driven by memory prices or volumes, or detailed the timing of that spending.⁴

ST.04US Credit MarketPartial signal

How Far Has the High-Yield Market Moved?

US high-yield spreads have widened, but we cannot establish what portion is attributable to AI companies.

333bpHY spread · Oct 20 · highest since July

+70bpVersus 263bp on Aug 31 (our calculation)

+56bpSep 22 → Oct 1 · concentrated widening (our calculation)

The ICE BofA US High Yield Index option-adjusted spread (OAS) rose by 70bp, from 263bp on August 31 to 333bp on October 20—the highest level since July (our calculation). Much of that increase occurred in late September: the spread widened 56bp from 268bp on September 22 to 324bp on October 1, before fluctuating in the low-300s. The sharpest widening therefore preceded the October 15 report that CoreWeave's CDS had exceeded 1,000bp.^{2, 5} Hana Securities' CrediVille team considers BB−-rated CoreWeave a likely major driver of the widening. Its charts also show spreads on 30-year US AI corporate debt climbing sharply in the second half, while a comparable series excluding AI companies remained broadly flat (visually assessed).³ However, the ICE BofA index measures a broad, weighted universe of US speculative-grade bonds. Without knowing whether and at what weight CoreWeave's bonds are included—and how much their individual spreads contributed—its share of the overall move cannot be established. Higher rates may have increased borrowers' funding pressure at the same time that investors were reassessing AI credit risk, but the data available here cannot disentangle the two. iM Securities noted that a high-yield spread remained below its post-2019 median of 356bp; because its index definition was not specified, we do not compare that benchmark directly with the ICE BofA series in this note.¹⁶

Figure 4

Which Came First? A Timeline of the Evidence

Sep 1–Oct 22 · key events across five observation points · bars show periods, dots show dates

US high-yield OAS: ICE BofA US High Yield Index (FRED; selected observations only). Blue Owl: AltsWire on Oct 2. CDS: Reuters, Oct 15. Meta: Oct 21 earnings. Korean events: Hanwha for suppliers (Oct 2–19), Snowshagal Daily and News1 for the KOSPI, Bank of Korea (BoK) for Oct 22. Chronological order does not establish causation.^{2, 4, 5, 9, 13, 17, 18}

ST.05South KoreaCredit: not isolated · equities: clear move

Where Has the Risk Shown Up in Korea?

Shares of Korean data-center suppliers have fallen sharply, but an AI-specific spillover into domestic corporate credit has not been identified.

+12.7bpAAA public-sector 3Y spread, YTD · BBB0 +14.4bp (our calculation)

−30.9bpA1 91-day CP spread, YTD (our calculation)

−21.1%13 data-center suppliers, median · Oct 2–19

₩1,328KRW per US dollar · Oct 20 · won strengthened

Korean Credit: Spreads Up Roughly 13–19bp Across Ratings

Hana Securities' CrediVille team argues that the early warning signs in Korean credit lie not in lower-rated, high-yield bonds (BBB and below), but in short-dated and top-rated paper. Its reasoning is that Korea's 'canaries in the coal mine' had already reacted to two credit events in the first half, and the domestic credit cycle may have peaked ahead of the global one. With issuance and trading concentrated at shorter maturities this year, new stresses may also show up there first. Its table reports year-to-date percentage spread changes of more than 50% for three-year AAA public-sector and bank bonds, about 3% for BBB0, and around −40% for commercial paper (CP).³

Converting the same table into basis-point moves tells a different story. AAA public-sector bonds widened from 22.9bp to 35.6bp, a gain of 12.7bp. The increases for AAA bank bonds, AA− corporates, AA− specialized credit finance companies and BBB0 corporates were 13.2bp, 16.9bp, 19.4bp and 14.4bp respectively (our calculations). Across those ratings, the absolute widening was broadly similar—around 13–19bp. The large variation in percentages reflects the starting point: the top-rated spread began in the low 20s, compared with roughly 500bp for BBB0. A 14bp move is substantial relative to 23bp but small relative to 500bp. Hana's use of proportional changes therefore still offers a valid perspective.

Figure 5

In Basis Points, Top-Rated and BBB Flat Bonds Widened by Similar Amounts

YTD 2026 credit-spread changes · 3Y bonds measured over Korean Treasuries; CP over 3-month Bank of Korea Monetary Stabilization Bonds · latest around Oct 19 (table date not specified)

Start-of-year and latest spreads, and percentage changes: Hana Securities CrediVille Chart 3 (Infomax). Changes in bp: our calculation. Percentage changes can appear large when the starting spread is low.³

That does not make Hana's conclusion irrelevant. The firm also notes that Korean lower-rated bond prices are relatively poor indicators of actual market conditions and that their correlation with overseas credit markets is limited.³ A fair reading is that top-rated and shorter-dated instruments, where price discovery is more reliable, may provide earlier signals (our interpretation).

Most of the Rise in Yields Came from Government Bonds

The yield on three-year AAA public-sector bonds rose 105bp from 3.18% at the start of the year to 4.23%. Of that rise, about 93bp came from Korean Treasury yields and 13bp from the credit spread. The 91-day CP yield rose 49bp: three-month Monetary Stabilization Bond yields rose 80bp, while the CP spread narrowed by 31bp (as of October 19).³ Hana attributes the recent rise in short-term rates to a funding premium driven by banks absorbing liquidity from the market. Net issuance of both bank bonds and securities backed by term deposits has increased since August.³ Two Bank of Korea rate hikes in July and August (from 2.50% to 3.00%) coincided with changing bank funding conditions. That overlap makes it difficult to isolate any AI-related contribution to wider Korean spreads.¹⁹

Figure 6

Most Yield Increases Came from Sovereign Rates; Short-Term Credit Spreads Actually Narrowed

Breaking down yield changes since the start of 2026 · as of Oct 19

Hana Securities CrediVille Charts 4 and 5 (Infomax). For public-sector bonds, 93 + 13 = 106bp versus 105bp for the total yield change; the 1bp discrepancy is due to rounding.³

Spreads widened in the most recent week as well. Between October 12 and 16, the 10-year Korean Treasury yield fell 24.7bp, but credit yields did not fall by as much; three-year spreads widened by 0.8–1.7bp across ratings. Hana expects overseas credit developments and other pressures to leave Korean credit spreads broadly weak or wider in the near term.³

On October 22, the Bank of Korea left its policy rate at 3.00%, with all seven Monetary Policy Board members voting in favor. It said it needed to assess the effects of its two earlier hikes before deciding the timing and pace of any further increases. The Bank expects growth this year and next to substantially exceed its August forecasts of 3.3% and 2.9%, with inflation staying above target for some time. The hold should therefore not be read as a dovish turn. In discussing the sharp rise in long-term government bond yields internationally, the Bank cited inflation, expectations of rate increases in major economies, fiscal sustainability concerns and 'concerns about AI-related corporate bond supply.' It did not attribute widening Korean corporate credit spreads to an AI-driven spillover.¹⁸ The decision matched market expectations. In a pre-meeting survey, all 10 securities-industry analysts expected a hold; most anticipated further hikes in November and the first quarter of 2027.²⁰ This note does not incorporate the bond-market reaction after the Monetary Policy Board decision, including changes in short-term yields or bank- and public-sector bond spreads.

Equities: The Biggest Moves Were in Supply-Chain Stocks

Among Korean assets, data-center supply-chain shares have moved most conspicuously in the same direction as AI credit concerns. According to Hanwha's Park Se-yeon, eight of 13 selected Korean data-center suppliers lost more than 20% between October 2 and 19. Their median return was −21.1%, over five times the −3.9% median for listed common shares. Memory (−25.8%) and electrical equipment (−22.3%) were hit hardest; only two backup-battery suppliers posted gains.⁴ However, the 13 stocks were selected by theme, spanning semiconductors, electrical equipment, cooling and on-site power generation. The decline may also reflect the semiconductor cycle and a reassessment of valuations. The Bank of Korea likewise cited concerns about weaker profitability in semiconductors as one factor in the Korean equity selloff.¹⁸ As with Korean credit spreads, it is not possible to separate the share-price impact of AI credit concerns from other forces.

Figure 7

Eleven of Thirteen Suppliers Fell; Only Two Backup-Battery Stocks Rose

Cumulative daily returns, Oct 2–19, for 13 Korean data-center suppliers · groups ordered approximately by how soon orders feed through into earnings

Hanwha Investment & Securities, Park Se-yeon, Figure 6 (Quantiwise). The broad listed-stock median covers 2,428 common shares traded on all nine trading days.⁴

The benchmark index was volatile too. The KOSPI fell 3.73% to 5,552.74 on October 20, then rose 3.29% to 5,735.66 the next day, recovering 85% of the previous session's point loss.²¹ AI cloud provider Elice, which listed on October 20, fell 32% below its IPO price on debut despite a retail subscription ratio of 408 to 1. Kiwoom Securities saw this as a possible sign that questions over returns on AI-infrastructure investment—arising in the wake of CoreWeave—were affecting newly listed AI shares, alongside more conventional factors such as an ambitious IPO valuation.²²

The KOSPI could not sustain the rebound on October 22. It lost 3.52% to close at 5,533.77, below its October 20 close. Samsung Electronics fell 4.03% and SK hynix 5.49%.¹⁷ Aju Business Daily cited several overlapping drivers: doubts about AI investment profitability, including Meta's lack of specific 2027 CapEx guidance and its negative Q3 FCF; geopolitical tensions in the Middle East; profit-taking in semiconductors following the prior day's rebound; and reduced buying support after corporate share-repurchase programs ended.²³ Given the mix of factors, the October 22 decline cannot be described as further evidence that CoreWeave's credit problems had spread to Korean equities. To trace a genuine credit channel, we would need to examine spreads on individual Korean corporate bonds and companies' actual financing terms. Figure 7 retains its original October 2–19 comparison because the same stock universe and benchmark group have not been recalculated through October 22.

The Korean won moved the other way. The USD/KRW exchange rate fell to ₩1,328.10 per dollar on October 20, meaning the won strengthened.²⁴ iM Securities economist Park Sang-hyun described this divergence as two markets reading the same economy differently: equities were pricing in doubts about the AI cycle, while foreign exchange was reflecting expectations that the semiconductor boom would continue to support Korea's trade and current-account surpluses.²⁵

The Longer-Term Exposure: Customers' Credit Quality

Beyond share prices, a lasting vulnerability may lie in the buyers placing orders. Park Se-yeon argues that on-site power-generation equipment, with delivery dates extending into 2027–28, faces financing risk for longer than other parts of the Korean supply chain. Even within one supply chain, exposure depends on the credit standing of the ultimate customer.⁴ On October 6, Doosan Enerbility shipped its first 380MW gas turbine to SpaceXAI, an AI company led by Elon Musk, according to the cited report. Under the five-turbine order from the same US Big Tech customer, three units added in December 2025 are scheduled for delivery—one in 2027 and two in 2028.^{26, 27} Publicly disclosed contract details, however, are insufficient to determine the credit quality of the entity legally responsible for payment or whether a parent-company guarantee exists.

COUNTERARGUMENTWhy the Spillover May Be Overstated

Demand and Relative Scale Tell a Different Story

Demand

Demand Indicators Have Not Broken Down

North American data-center vacancy was 1.4% in the first half of 2026, according to CBRE, and over 80% of the 7,481MW under construction had already been pre-leased. Five major tech companies raised CapEx guidance a total of 15 times between January 2025 and July 2026 (Hanwha's Park Se-yeon, citing CBRE).⁴

Scale

CoreWeave Is Small by Comparison

Projected combined CapEx for neoclouds in 2026 was around $60bn, less than one-tenth of the roughly $680bn projected for four hyperscalers (Hanwha's Park Jun-young; visual chart estimates).¹

Investment recovery

GPU Rental Prices Are Rising

SK Securities argues that rising GPU rental rates shorten payback periods, while long-term memory supply agreements carry sufficiently high breach costs to discourage easy cancellation.⁷

These counterarguments speak to demand and relative company size. They do not negate the fact that financing costs have already risen for debt-funded infrastructure providers. Park Se-yeon likewise sees resilient demand indicators, but locates the risk primarily in the creditworthiness of the parties expected to pay for that demand.⁴

ST.06What to Watch NextCalendar

Earnings, the FOMC and Private-Credit Redemptions: The Next Tests

10.28~30Korea time (KST)
SK hynix, Samsung, Alphabet, Microsoft and Amazon Earnings
SK hynix on Oct 28; Samsung Electronics, Alphabet and Microsoft on Oct 29 (the US releases arrive early morning KST); Amazon on Oct 30 (early morning KST). Watch CapEx budgets, volume commitments under long-term memory agreements, and whether other hyperscalers can illuminate the pace of 2027 CapEx growth and the route back to positive FCF that Meta did not quantify (a Kiwoom Securities monitoring point).^{4, 22}
10.2903:00
FOMC Decision
How the remaining scale and duration of Fed tightening feed into high-yield spreads and borrowing costs (NO.14).⁶
11.26Korea morning
Bank of Korea Monetary Policy Board
The timing and pace of any further hikes. Changes in short-term yields and AAA bank- and public-sector bond spreads following October's hold may help distinguish domestic drivers of Korean credit spreads from outside influences.¹⁸
ONGOINGDaily / weekly
Key Indicators at Each Observation Point
Whether CoreWeave's CDS exceeds its October high and Oracle's exceeds 288bp (S&P Global MI, Oct 15) · US HY spread versus 333bp (Oct 20) · bp changes in AAA three-year public-sector and bank spreads (35.6bp and 38.6bp) and BBB0 (511.1bp) · returns of the 13 suppliers versus all listed common shares · USD/KRW
2027.1Expected
Q4 Private-Credit Redemption Results
Whether Blue Owl and other funds continue to face elevated requests and maintain 5% tender limits (timing inferred from the quarterly redemption schedule cited by Hanwha's Park Se-yeon).⁴

KNOWN

CoreWeave five-year CDS above 1,100bp (mid-October, Hanwha)
Oracle five-year CDS at a record 288bp (Oct 15, S&P Global MI); rated BBB−
US HY spread: 263bp (Aug 31) → 331bp (Oct 16) → 333bp (Oct 20)
Korean three-year spreads widened 12.7–19.4bp YTD; 91-day CP spread narrowed 30.9bp (Hana table; our calculations)
Median return: 13 suppliers −21.1% vs listed common shares −3.9% (Oct 2–19)
KOSPI: Oct 20 −3.73%, Oct 21 +3.29%, Oct 22 −3.52%
Meta Q3 FCF −$2.1bn; long-term debt $58.7bn → $80.9bn (end-2025 → end-Sep); 2026 CapEx guidance $135–145bn
BoK held its policy rate at 3.00% on Oct 22 (unanimous); statement cited 'concerns about AI-related corporate bond supply'

ESTIMATE

Six-company consensus FCF: approximately −$30bn in 2026 and −$120bn in 2027 (visually estimated)
Big Tech CDS roughly 50–105bp (estimated from Hanwha chart)
Private credit's expected contribution: approximately $800bn (Morgan Stanley)
Korea's credit cycle may already have passed its peak (Hana Securities)

UNKNOWN

Meta's 2027 CapEx (not quantified in the release) and spending plans at Alphabet, Microsoft and Amazon
Which companies account for the negative aggregate six-company FCF
CoreWeave's and other AI companies' contribution to wider US HY spreads
The AI-specific portion, if any, of widening Korean credit spreads
Financing conditions for on-site power-generation customers in 2027–28
Whether Meta's new Class C shares are actually used to raise capital (CFO says no current plan)

CoreWeave's elevated credit risk is plain in its CDS and bond prices. Around the same time, the cost of credit protection rose for Oracle and Big Tech, while US high-yield spreads widened. Yet we lack the data to apportion the index-level move to AI issuers. Consensus forecasts suggest that six companies' combined CapEx will exceed cash from operations in 2026–27, and Meta reported negative FCF for Q3 alone. In Korea, credit spreads have widened since the start of the year, but rate hikes and bank funding pressures have also been at work. Data-center suppliers' shares fell much more sharply than the market, yet an AI-specific spillover into Korean corporate credit has not been identified. Over the longer term, the creditworthiness of customers placing on-site power-generation orders deserves scrutiny. The next immediate tests are the October 28–30 earnings reports, particularly Big Tech CapEx plans and funding choices.

This report assesses the reach of AI-infrastructure credit concerns using publicly available indicators and securities-firm research. It neither forecasts a credit event nor recommends a trade. Source dates and calculation methods appear below.

TERMS · GLOSSARY

CDS: Credit default swap. The CDS premium, quoted in basis points, is the cost of protection against a borrower's default. Higher premiums generally signal a higher market-implied level of credit risk.
High-yield spread: The yield premium on speculative-grade corporate bonds relative to a benchmark government yield. The ICE BofA series cited here is an option-adjusted spread (OAS).
Neocloud: A specialized cloud provider built around renting GPU capacity for AI workloads, such as CoreWeave or Nebius.
Hyperscaler: A large cloud platform operating data centers at enormous scale, such as Amazon, Microsoft or Google. The composition of the 'four Big Tech firms' differs between source reports.
Private credit: Direct lending to companies by investment funds and asset managers rather than banks. Fund redemption windows and tender limits can restrict when investors receive cash.
Free cash flow (FCF): Cash generated by operations after capital investment. If FCF is negative, the company must draw on cash balances or obtain additional funding.
RPO: Remaining performance obligations: contracted amounts not yet recognized as revenue; broadly comparable to a contracted revenue backlog.
Credit spread (Korea): The difference between yields on corporate, bank or similar bonds and Korean Treasury bonds of comparable maturity (or Monetary Stabilization Bonds for short-dated instruments).
Public-sector bonds / specialized finance bonds: The former are bonds issued by Korean public corporations and government-related entities; the latter are issued by specialized credit finance companies, including card issuers and consumer or commercial finance firms.
Commercial paper (CP) / term-deposit securitization: CP is short-term debt issued by companies. Term-deposit-backed securities are short-term instruments issued against bank term deposits.
Long-term agreement (LTA): A multi-year supply contract in which a memory manufacturer and its customer agree in advance on volumes and pricing terms.

SOURCES · DATES · METHODS

Published October 22, 2026 · Information cut-off After the Korean market close on October 22, 2026 (KST). Korean equity prices and Bank of Korea announcements are included through October 22. Korean credit spreads are from Hana Securities data around October 19; the 13-supplier returns cover October 2–19; the US high-yield series is observed through October 20; Meta's October 21 results and follow-up call are included. Visual estimates from brokerage charts are labeled. Bond-market responses to the BoK decision and Meta's regular-session share price after the release are not included.

Observation dates	CoreWeave CDS: mid-October (Hanwha Oct 21). Oracle and SoftBank CDS: Oct 15 (Reuters). Big Tech CDS: final visible points in Hanwha Oct 21 Figure 1 (date unspecified). US HY OAS: through Oct 20 (FRED). Korean spread table's 'current' column: date unspecified; Charts 4–5: Oct 19; weekly change: Oct 12–16 (Hana Oct 21). Supplier returns: Oct 2–19 (Hanwha Oct 20). KOSPI closes: Oct 20, 21 and 22. BoK announcement: Oct 22. USD/KRW: Oct 20. Six-company CapEx and FCF consensus: chart endpoints around Oct 2026 (SK Securities Oct 21).
Values estimated from charts	SK Securities, page 5 (Bloomberg consensus, six-company aggregate): CapEx about $380bn in 2025, $760bn in 2026C, $1.12tn in 2027C and $1.24tn in 2028C; FCF about +$180bn in 2025, −$30bn in 2026C, −$120bn in 2027C and +$15bn in 2028C. Hanwha (Park Jun-young), Figure 1 (CDS): Big Tech four-company endpoints around 50–105bp, CoreWeave October peak around 1,125bp, Oracle chart endpoint around 320bp. Figure 3 (2026F CapEx): neoclouds around $60bn and hyperscalers around $680bn.
Our calculations	Korean three-year credit spread change (bp) = latest less start-of-year: AAA public-sector 35.6 − 22.9 = +12.7; AAA banks 38.6 − 25.4 = +13.2; AA− corporates 67.7 − 50.8 = +16.9; AA− specialized finance 74.0 − 54.6 = +19.4; BBB0 corporates 511.1 − 496.7 = +14.4; A1 91-day CP 47.0 − 77.9 = −30.9. Reproduced percentage changes: +55.5%, +52.0%, +33.3%, +35.5%, +2.9%, −39.7% (Hana reports +55%, +52%, +33%, +36%, +3%, −40%). US HY: 333 − 263 = +70bp; Sep 22 → Oct 1: 324 − 268 = +56bp; 333bp on Oct 20 was the highest since Jul 1. Meta long-term debt: $80.922bn − $58.744bn = +$22.18bn. Meta FCF: cash from operations $32.786bn − PP&E purchases $33.942bn − finance lease principal $0.964bn = −$2.12bn (matches Meta's reported figure). CoreWeave Q2 net interest expense/revenue: $0.640bn / $2.575bn = 24.9%. The KOSPI regained 182.92 of the 215.27 points lost on Oct 20 (5,735.66 − 5,552.74), or 85%; Oct 20 decline: 215.27 / 5,768.01 = 3.73%.
Differences across sources	CoreWeave CDS: Park Jun-young at Hanwha says 'above 1,100bp'; Reuters and Park Se-yeon say 'above 1,000bp'. CoreWeave bond yield: Park Jun-young gives approximately 15–16% (citing news reports); Park Se-yeon reports a yield briefly above 16% on an Oct 2031 bond on Oct 15; Reuters cites a 2032 issue trading around 78 cents on the dollar with a yield just over 16% (Tradeweb). Oracle CDS: Reuters / S&P Global MI gives 288bp on Oct 15; Park Jun-young's text says 'approaching 300bp'; chart endpoint visually reads around 320bp (possibly reflecting a different date). AAA public-sector yield decomposition: 93 + 13 = 106bp versus 105bp for the reported total (rounding).
Source verification notes	Hana's CrediVille report gives CoreWeave a BB− rating without naming the rating agency. SK Securities charts aggregate six companies without reporting individual figures. Hanwha's Park Se-yeon names the gas-turbine buyer 'xAI'; this note uses 'SpaceXAI' following Doosan Enerbility shipment coverage dated Oct 6. Figures from Blue Owl, J.P. Morgan Asset Management and Morgan Stanley are cited within Park Se-yeon's report. The four Big Tech names differ by source: Reuters uses Amazon, Alphabet, Meta and Nvidia; Hanwha's chart uses Microsoft, Google, Amazon and Meta. iM Securities' post-2019 high-yield median of 356bp is omitted from Figure 4 because its underlying index is unspecified. ICE BofA index data carry redistribution restrictions on FRED, so only selected observations, not a daily time-series chart, are reproduced here.
Estimates and interpretations	Statements that the projected funding gap requires cash reserves or external financing, that top-rated and short-dated Korean bonds may offer earlier signals, and that Meta's results show both investment expansion and diversification of funding options are this note's interpretations. The possibility that Oracle and CoreWeave account for a large share of negative aggregate FCF is our hypothesis. Classifications at the five observation points and judgments in the spillover table are our assessments. Figure 1 is schematic.

Hanwha Investment & Securities, Park Jun-young (research associate Kim Na-woo), 'The CoreWeave Episode Does Not Mean the End of AI Investment,' Semiconductor Industry Analysis (Positive), Oct 21, 2026. Page 1: 'CoreWeave's five-year CDS premium exceeded 1,100bp in mid-October'; some bond yields at approximately 15–16% (citing media reports); Oracle CDS approaching 300bp. CoreWeave borrows heavily to secure GPUs and data centers in advance, then seeks to recoup the cost by renting capacity to customers, making it sensitive to rates and financing conditions. Hyperscalers have greater financial flexibility thanks to ample liquidity and high credit ratings. Unless credit-market stress actually leads hyperscalers to cut investment, growth drivers for HBM4 and server DRAM remain intact in 2027. Figure 1: CDS trends for neoclouds and hyperscalers. Figure 2: hyperscaler CapEx, 2026 vs 2027. Figure 3: 2026 CapEx comparison; neoclouds = CoreWeave + Nebius; hyperscalers = Microsoft + Google + Amazon + Meta.
ICE BofA US High Yield Index Option-Adjusted Spread, FRED series BAMLH0A0HYM2 (retrieved Oct 22, 2026). Observations: Aug 31, 2.63%; Sep 22, 2.68%; Oct 1, 3.24%; Oct 16, 3.31%; Oct 19, 3.29%; Oct 20, 3.33% (highest since Jul 1). Data copyright ICE Data Indices, LLC; FRED notes redistribution restrictions. This report quotes selected observations only. fred.stlouisfed.org
Hana Securities, Kim Sang-man, 'Not High Yield (HY) but High Sensitivity: Korea's Barometers Are Short-Dated and Top-Rated Bonds,' CrediVille, Oct 21, 2026. Page 1: with CoreWeave rated BB−, it is 'likely a major driver of recent high-yield spread widening'; the effects may not remain limited to speculative-grade issuers. Korean high-yield (BBB or lower) spreads remained firm. The report suggests Korean lower-rated bond pricing indicators are less reflective of actual conditions, while correlations between domestic and foreign credit markets are limited. Korea may already have passed its credit-cycle peak. Short-term rate increases reflect an 'additional premium from supply and demand pressures.' Chart 1: US AI vs non-AI 30-year corporate spreads and HY spreads. Chart 3: year-to-date spread percentage changes (Infomax). Chart 4: AAA public-sector 3Y yields 3.18% → 4.23%, with +93bp from KTB yields and +13bp from the spread. Chart 5: CP 91D yield +49bp = +80bp in 3M MSBs and −31bp in the spread. Chart 7: net bank-bond and term-deposit-backed issuance rose from August. Weekly credit conditions: Oct 12–16, 10Y KTB −24.7bp and 3Y spreads +0.8 to +1.7bp (Chart 2); foreign credit events and other pressures are expected to keep spreads steady to wider. The report saw a low probability of a further BoK hike on Oct 22.
Hanwha Investment & Securities, Park Se-yeon, 'Permits Stalled Data Centers; Investment Kept Growing,' ESG, Oct 20, 2026. Page 1: eight of 13 Korean data-center suppliers fell more than 20% between Oct 2 and 19; five Big Tech companies raised CapEx guidance 15 times; what changed was financing costs for companies borrowing to build, not demand (CoreWeave bond yield briefly >16% on Oct 15). Page 2: 2,428 listed common stocks traded in all nine sessions had a median return of −3.9%, versus −21.1% for the 13 suppliers; memory −25.8%, electrical equipment −22.3%, cooling −21.1%, on-site generation −18.9%; only the two backup-battery names rose (Samsung SDI +11.9%, LG Energy Solution +16.5%). Figure 6: Quantiwise. North American data-center vacancy: 1.4% in H1 2026, with more than 80% of 7,481MW under construction pre-leased (CBRE). Alphabet, Meta, Amazon, Microsoft and Oracle raised guidance 15 times between Jan 2025 and Jul 2026. Section 4: demand concentrated in a handful of AI labs; Meta Hyperion joint-venture bonds of $27.3bn (Blue Owl affiliate: 80% ownership; Meta 10-Q / IFR); Morgan Stanley's Jul 2025 estimate that private credit would supply about $800bn of the $1.5tn in outside capital through 2028; two Blue Owl private-credit funds offering quarterly tenders capped at 5% for a third consecutive quarter; OTIC requests around 39% of shares. Big Tech bond spreads +30bp YTD vs broader investment grade +2bp (J.P. Morgan Asset Management, August). Oracle downgraded to BBB− on Jul 9 (S&P), with record 5Y CDS 288bp on Oct 15 (S&P Global MI); CoreWeave's Oct 2031 bond yield briefly >16% on Oct 15, CDS >1,000bp, a record. Spreads widened most for lower-rated capacity lessees and their lenders rather than AA-rated Big Tech. On-site generation deliveries in 2027–28; gas turbine buyer identified as xAI. Three issues for Meta earnings: maintenance of the CapEx plan, price versus volume behind any increase (particularly memory), and spending timing. 'The risk lies less in demand than in who is expected to pay.' Figure 9 calendar: Meta Oct 22 KST; SK hynix Oct 28; Samsung, Alphabet, Microsoft Oct 29 (early morning for US releases); Amazon Oct 30 (early morning); Q4 private-credit redemption results expected in early Jan 2027.
Reuters, Tom Westbrook and Amanda Cooper, 'AI Credit Fears Push Tech Default Insurance to Record Highs,' Oct 15, 2026 (republished by BusinessDay). Amazon, Alphabet, Meta and Nvidia CDS traded at record levels; Oracle five-year CDS at 288bp, up from 260bp one week earlier; SoftBank five-year CDS at 424bp, up 24bp on the day; CoreWeave CDS >1,000bp (S&P and LSEG), debt approximately $50bn (LSEG). A 2032 bond traded near 78 cents on the dollar with a yield just over 16% (Tradeweb). The report estimated that 1,000bp implied a five-year cumulative default probability of about 63%, versus about 30% a year earlier. Big Tech CDS rose much less rapidly than Oracle's or CoreWeave's. businessday.co.za
Snowshagal Investment Note No. 14, 'The US 10-Year Treasury Yield Tends to Peak Around the Final Rate Hike,' Oct 21, 2026. snowshagal.com
SK Securities, Han Dong-hee, 'This Time Is Different,' Semiconductors, Oct 21, 2026. Long-term memory supply agreements carry high breach penalties; rising GPU rental prices shorten payback periods; argues that memory stocks offer a buying opportunity. Page 5 charts: 'Aggregate CapEx for Six North American Companies: Trends and Outlook' and 'Aggregate FCF for Six North American Companies: Trends and Outlook' (Bloomberg; the six are Google, Amazon, Microsoft, Meta, Oracle and CoreWeave; evolution of annual consensus from Jul 2025 to around Sep–Oct 2026).
CoreWeave, Form 10-Q for quarter ended June 30, 2026. RPO of $103.7bn, with 41% expected to be recognized over the next 24 months; Q2 revenue of $2.575bn and net interest expense of $0.640bn. sec.gov
AltsWire, 'Blue Owl's OCIC Redemption Requests Ease to 16.8%; OTIC Holds Near 39% in Q3,' Oct 2, 2026. OTIC redemption requests: Q1 2026 40.4% (about $1.2bn), Q2 38.1% (about $1.1bn), Q3 approximately 39% (about $1.1bn). Q3 repurchases: $135m, or approximately 13% of amounts requested. Both funds again limited tenders to 5% of outstanding shares. A shareholder letter said that 'the vast majority of requests were resubmissions of previously unfulfilled tenders rather than new demand.' altswire.com
US SEC, Blue Owl Technology Income Corp., Schedule TO tender-offer document, Exhibit 99(a)(1)(B), 2025. Stated repurchase size: '5.00% of the aggregate number of the Company's Shares outstanding'; stated intention to conduct quarterly tender offers, subject to board discretion. sec.gov
IMF Blog, Athanasios Vamvakidis and Jason Wu, 'Global Financial Markets Face Rising Risks From AI, Bond Yields, and Cyberattacks,' Oct 13, 2026, accompanying the October Global Financial Stability Report. Fourteen circular-financing arrangements since 2023 linked about $300bn of equity investment and $650bn of debt with $1.4tn in future revenue commitments. As the authors warn, 'stress at key companies receiving investment could lower revenue forecasts and investment valuations, which in turn would weaken the value of the accumulated debt in those deals.' imf.org
Federal Reserve, 'Minutes of the Federal Open Market Committee, September 15–16, 2026,' published Oct 7, 2026. Open-market desk: 'yield spreads on hyperscaler debt used to finance AI infrastructure remained wide, given the large volume of issuance and the relatively long duration of the securities being issued.' Staff financial-conditions assessment: 'Spreads for corporate bonds issued by AI hyperscalers decreased slightly over the period after widening notably from low levels at the start of the year.' federalreserve.gov
Meta, 'Meta Reports Third Quarter 2026 Results,' Oct 21, 2026. Revenue $64.601bn (+26%); operating income $12.988bn (including $9.89bn of litigation expense; $22.88bn excluding it); diluted EPS $3.83 ($7.40 excluding litigation expense). Operating cash flow $32.786bn; purchases of property and equipment $33.942bn; finance lease principal $0.964bn; FCF −$2.120bn (versus +$10.625bn in Q3 2025); Jan–Sep FCF +$11.049bn (year-earlier +$29.508bn). CapEx including finance lease principal $34.91bn. Cash and marketable securities $82.45bn. Long-term debt $80.922bn (end-2025 $58.744bn). Jan–Sep long-term debt issuance $24.910bn, with no share repurchases (vs $26.248bn a year earlier). Distribution received from data-center joint venture $1.037bn. 2026 CapEx guidance $135bn–$145bn (previously $130bn–$145bn). No numerical 2027 guidance: 'we expect to continue investing aggressively to support both our core business and emerging AI initiatives.' New non-voting Class C shares effective Jan 19, 2027, with one Class C share distributed per existing share. First stated rationale: 'Creating a New Class of Shares for Capital Raising, Employee Compensation, Potential Acquisitions.' Earnings call at 4:30 p.m. US Eastern, Oct 21. investor.atmeta.com
Meta, 'Q3 2026 Prepared Remarks' and 'Q3 2026 Follow Up Call Transcript,' Oct 21, 2026. Prepared remarks: CapEx including finance lease principal of $34.9bn on servers, data centers and networking; cash and marketable securities approximately $82.4bn; debt approximately $83.7bn; 2027 budgets and capacity planning still in progress, with aggressive investment expected to continue. Follow-up call, CFO Susan Li: the filing's 'hundreds of billions of dollars' was 'a theoretical maximum that was basically was computed to determine what level of equity financing would trigger a loss of founder control. It is not actually a description of our present equity financing needs or plans.' 'The landscape for compute remains pretty constrained'; capacity demand continues into next year; near-term capacity supplemented by third-party cloud contracts. Prepared remarks · Follow-up call
Benzinga, 'Meta Posts Q3 Double Beat: We're Seeing Velocity Across the Board,' Oct 21, 2026, 4:21 p.m. US Eastern. Revenue $64.6bn vs consensus $63.37bn; adjusted EPS $7.40 vs $5.74 (Benzinga Pro). Shares +3.12% at $719.30 in after-hours trading at publication time. benzinga.com
iM Securities Credit, 'Bull Flattening Alone Is Not Enough to Sustain Credit Strength,' Oct 19, 2026. US high-yield spreads stood below the post-2019 median of 356bp.
News1, 'KOSPI Closes Down 3.52% at 5,533.77; KOSDAQ Drops 3.62%' (second update), and Seoul Shinmun market-close recap, Oct 22, 2026. KOSPI −201.89 points (−3.52%) to 5,533.77, after touching 5,485.98 intraday. Samsung Electronics −4.03%, SK hynix −5.49%. news1.kr · seoul.co.kr
Bank of Korea, 'Monetary Policy Decision,' press release 2026-10-15, Oct 22, 2026. Policy rate kept at 3.00%; need to 'assess the transmission effects of the two rate increases.' On international markets: 'long-term government bond yields increased sharply due to higher inflation, stronger expectations of rate increases in major economies, and concerns over fiscal sustainability and AI-related corporate bond supply.' Korean stocks fell sharply 'amid concerns about a slowdown in semiconductor profitability.' The Bank would 'decide on the timing and pace of any further rate increases.' All seven Monetary Policy Board members voted in favor. Next meeting: Nov 26. Korean-language press-release PDF (Oct 2026) · Monetary Policy Board materials index
Snowshagal Investment Note No. 13, 'Will the Bank of Korea Hike Even as Stocks Tumble?' Oct 20, 2026. BoK raised the base rate July 16 from 2.50% to 2.75% (first increase since Jan 2023), and again on Aug 27 from 2.75% to 3.00%. snowshagal.com
Financial News, reporting a News1 survey, 'All 10 Expect a Hold This Month; Two Further Hikes Could Take Rates to 3.5%,' Oct 18, 2026. All 10 securities analysts expected the BoK to hold in October; most anticipated one hike in November and another in Q1 2027, for an eventual 3.5%. fnnews.com
Snowshagal Daily, 'AI Earnings Spark a 3% Rebound, Yet Half the KOSPI Falls,' Oct 21, 2026 close. KOSPI +3.29% to 5,735.66 versus prior close 5,552.74, recovering 85% of the previous day's 215.27-point loss.
Kiwoom Securities, Han Ji-young, 'Untangling Positioning and Weak Sentiment: Equity Risks to Work Through,' Investment Strategy / Market Commentary, Oct 21, 2026. Monitoring points for Meta earnings: 'the pace of 2027 CapEx growth and the route to FCF recovery.' KOSPI fell 3.7% on Oct 20. Elice IPO: retail subscription ratio of 408:1, debut down 32% versus offering price; conventional concerns about valuation alongside questions over AI-infrastructure returns following the CoreWeave episode.
Aju Business Daily, 'AI Profitability and Trading-Flow Worries Return as KOSPI Slumps 3.52%,' Oct 22, 2026. Meta's lack of specific 2027 CapEx guidance and Q3 FCF deficit of $2.12bn weighed on sentiment; Middle East tensions; semiconductor profit-taking; diminished buying support as Samsung Electronics' (Oct 6) and SK hynix's (Oct 21) share-repurchase programs ended. ajunews.com
Eugene Investment & Securities, Kim Ji-na and Cha Young-hu, 'Eugene's FICC Update,' Oct 21, 2026. USD/KRW closed at 1,328.10 on Oct 20 (−0.2% day on day, −7.7% YTD), down ₩2.20.
iM Securities, Park Sang-hyun, 'USD/KRW and the KOSPI: Two Different Readings,' Economy Brief, Oct 21, 2026. Equities priced in concern that an AI-cycle slowdown could weaken Korea's semiconductor export boom; the FX market priced in continued strength in semiconductors, trade and the current-account surplus.
Money Today, 'Doosan Enerbility Ships 380MW Gas Turbine to Elon Musk's SpaceXAI,' Oct 6, 2026. Shipment ceremony at Doosan's Changwon headquarters for its S2 380MW turbine (DGT6-300H). Describes the buyer as 'SpaceXAI, a US artificial-intelligence company led by Elon Musk.' Doosan had contracts for five turbines with this US Big Tech customer. mt.co.kr
iNews24, 'Doosan Enerbility to Supply Three Gas Turbines to US Big Tech Firm,' Dec 17, 2025. Three 380MW units, 'one to be delivered in 2027 and two in 2028'; in addition to two units contracted in October, total orders from the same buyer stood at five. inews24.com

AI Credit Concerns:Where Are They Showing Up in Korea?