Goldman Sachs estimates 2026 US IPO proceeds could reach US$225 billion, a record high; 'bubble alarm not yet ringing' is the bank's own view (media-relayed forecast, not actual proceeds)

ANK-Doc ID: ANK-2026-07-26-003 版本: v1.0.0 發布日期: 2026-07-26 作者: 方實 (Deputy Editor, auto AI structuring) 分類: US capital markets / IPO x AI financing (media-relayed investment-bank research and wire aggregation; zero verified official anchors) 涵蓋文章: cnyes#1692211 (main anchor; Anue, 2026-07-25, relaying Goldman's "Top of Mind" report) plus 12 supporting articles from OP's Phase A full-library search: cnyes#1692244 (2026-07-24, same-topic Goldman roundtable plus Apollo's caution, CORROBORATE + ADD_EVIDENCE), cna#945717 (2026-06-12, SpaceX pricing plus Goldman's earlier US$160 billion forecast, REVISE prior value + CONNECT), cnyes#1692245 (2026-07-24, SpaceX post-IPO performance and the August 2026 lockup expiry, CONNECT — a live precursor to the main article's 2027 lockup thesis), cnyes#1406784 (2026-07-10, SK hynix ADR raising about US$26.5 billion, CONNECT — a mega-deal case), cnyes#1560648 (2026-07-16, Anthropic reportedly targeting an IPO as early as October, CONNECT), cna#1232005 (2026-06-26, OpenAI reportedly delaying its IPO to 2027, CONNECT), cnyes#1692207 (2026-07-25, AI credit-market concerns, ADD_EVIDENCE), cnyes#1661102 (2026-07-23, 30-year Treasury yields and the AI bond wave, CONTEXTUALIZE), cnyes#1655873 (2026-07-22, Alphabet raising capex guidance, ADD_EVIDENCE), cnyes#1655647 (2026-07-22, record short interest and BofA's 82% survey, CHALLENGE), cnyes#1672836 (2026-07-24, JPMorgan's AI-divergence warning, CHALLENGE), storm#1671461 (2026-07-24, Fitch's Taiwan angle, CONTEXTUALIZE). 選定方法: Daily run ANKRUN-20260726193705; the selector chose cnyes#1692211 as the seed (scores.official_count_verified=0, official_count_raw=0, official_attribution_unverified_count=0 — zero verified official anchors on this card; Goldman Sachs is a private investment bank, so every Goldman number is "as carried in media-relayed Goldman research," never an official endorsement; basis is news_aggregation throughout). Phase A full-library search was completed by OP; 13 distinct source articles adopted; wsql search of published cards yielded 3 internal links. Editorial angle (fact-pinning): three stakes — the forecast-versus-actuals distinction, the record-value-versus-modest-deal-count structure, and the attribution boundary between Goldman's sanguine thesis and countervailing market data; Goldman's two forecast values (US$160 billion / US$225 billion) are listed side by side with their report dates and no delta is computed. 13 sources yield 17 F-Units (one source per unit) — above the rapid-lane suggestion of 10-13 because each of the 13 sources requires at least one anchoring unit (orphan-source gate); all content is sourced, none is padding. This card generates no number absent from the source texts and computes no ratio or delta (unit conversions between Chinese 億 and "billion" are notation transcriptions of source-carried values, flagged as such). No Wikidata Q identifiers are attached (no registry verification performed this run; omission over fabrication).

TL;DR

Anue's 2026-07-25 report relays Goldman Sachs's "Top of Mind" research: Goldman estimates 2026 US IPO proceeds could reach US$225 billion, a record high approaching twice the 2021 record; Goldman chief US equity strategist Ben Snider says 2026 year-to-date proceeds have already passed US$125 billion, above the old 2021 full-year record of about US$120 billion (cnyes#1692211) [F-001]. This card pins three stakes. First, US$225 billion is a forecast, not an actual — and Goldman's forecast has a checkable evolution across report dates: CNA's 2026-06-12 report carried Goldman's projection that 2026 IPO proceeds could expand to US$160 billion (cna#945717) [F-006]; Anue's 2026-07-25 report carried an estimate of up to US$225 billion (cnyes#1692211) [F-001]; this card lists both with their report dates and computes no delta. Second, a record dollar amount does not equal a listing frenzy: roughly 53 to 60 IPOs have been completed in 2026 to date, still far below the 25-year median of about 100 per year; the main article states that this boom's defining feature is a handful of mega-deals inflating total proceeds [F-002] — SpaceX's US$75 billion raise (pricing reported 2026-06-12) [F-006] and SK hynix's roughly US$26.5 billion raise (reported 2026-07-10) [F-008] are the concrete cases. Third, "bubble warning signs not yet present" is Goldman's thesis, not a market consensus: within the same roundtable, Acadian's Owen Lamont said the IPO boom already meets at least one of his updated "Four Horsemen of the Market Bubble" (cnyes#1692244) [F-011]; on the market side, short interest is at a record (cnyes#1655647) [F-012], 82% of fund managers surveyed by BofA in July 2026 called chip stocks the most crowded trade [F-012], and JPMorgan warns that AI-stock divergence recalls the dot-com era (cnyes#1672836) [F-013]. Zero verified official anchors on this card; Goldman is a private investment bank; every figure is media-relayed (see §Scope) [J-001].

Misreading guard (1 item)

"Bubble warning signs have not yet appeared" is the thesis of Goldman Sachs (a private investment bank) in its "Top of Mind" report, as relayed by Anue's 2026-07-25 article (cnyes#1692211) — it is neither an official determination nor a market consensus. "US$225 billion" is Goldman's estimate for full-year 2026 (per the source text, an estimate "counting deals above US$25 million each"), not realized proceeds — the reported 2026 year-to-date actual is "already past US$125 billion." This card carries zero verified official anchors (official_count_verified=0); every figure is as carried in media-relayed Goldman research, roundtables, or wire reporting, and this site has not independently verified any of them. All other caveats: see §Scope.

Three stakes (fact-pinning: what most needs nailing down in this relayed bank research)

Stake one: a forecast is not an actual, and forecasts move. The checkable forecast evolution (REVISE layer): CNA's 2026-06-12 report (SpaceX pricing day) carried: "Goldman Sachs projects 2026 IPO proceeds could expand to US$160 billion, up severalfold from the prior year (2025) and a record" (cna#945717) [F-006]. Anue's 2026-07-25 report carried Goldman's latest estimate that full-year 2026 US IPO proceeds could reach US$225 billion (cnyes#1692211) [F-001]. Anue's 2026-07-24 report likewise carried "Goldman forecasts 2026 US IPO proceeds will reach US$225 billion" — the same value as the main article, with slightly stronger phrasing (the main article says "estimated up to") (cnyes#1692244) [F-011]. Three report dates, two forecast values, each listed with its own date — this card computes no delta and does not assert an "upgrade" action on Goldman's behalf; the checkable textual fact is only that, about six weeks apart, the media-relayed Goldman estimates differ. As for realized amounts, the main article carries only "US IPO proceeds so far this year have passed US$125 billion" (Snider's statement, as of the 2026-07-25 report) [F-001]. When citing, "forecast US$225 billion" and "actual past US$125 billion" are not interchangeable.

Stake two: record value does not equal deal-count mania. The main article carries: roughly 53 to 60 US IPOs completed in 2026 to date, up more than 50% from the same period of 2025 (source wording "年增逾50%"), the strongest start since 2021 — yet still far below the 25-year median of about 100 deals per year, the 250-plus listings of 2021, and the nearly 400 deals of the 1999 dot-com peak (cnyes#1692211) [F-002]. The main article states plainly: the defining feature of this boom is not a rush to list but a handful of mega-deals lifting total proceeds. The matching cases (each dated, sentences kept separate): SpaceX priced at US$135 per share on 2026-06-12, selling 555.56 million shares to raise US$75 billion, the largest IPO raise on record (cna#945717) [F-006]; SK hynix, per a 2026-07-10 report, raised about 40 trillion won (about US$26.5 billion per the report) via ADRs, the world's second-largest equity offering, behind only SpaceX's US$75 billion in June 2026 (cnyes#1406784) [F-008]. In the queue: Anthropic is reported to be targeting completion of an IPO as early as October 2026, valued at US$965 billion after its May 2026 round (Bloomberg citing people familiar; cnyes#1560648) [F-009]; OpenAI is reported to be weighing a delay of its IPO to 2027, with Sam Altman holding to a US$1 trillion valuation (cna#1232005) [F-010].

Stake three: "no alarm yet" is one thesis; countervailing data exist in parallel. Goldman's indicators (all as carried in the report): the median EV/sales multiple of recent US IPO companies is about 5x — above the 30-year median of 4x but clearly below 1999's 9x, 2021's 7x, and 2020's 11x; 43% of the 2025 listing cohort turned profitable in their first post-listing quarter, above 1999's 28% and 2021's 24%; extreme first-day pops are, barring a few cases, not currently frequent (cnyes#1692211) [F-003]. Yet in the same roundtable Lamont said the IPO boom already meets at least one of his "Four Horsemen" (elevated valuations, "bubble belief," aggressive share issuance, heavy inflows), and that issuance booms tend to mark a bubble's beginning rather than its end — the 1990s US IPO boom and Japan's asset bubble each ran for years (cnyes#1692244) [F-011]. Countervailing market data (CHALLENGE layer, each with its own population): S3 Partners says total short interest in S&P 500 constituents is about 3.7% of free float, the highest since S3 began tracking in 2010; BofA's July 2026 fund manager survey found 82% of respondents calling global chip stocks the most crowded trade and roughly half of respondents calling an AI bubble the biggest tail risk, up from 28% in the June 2026 survey (population = surveyed fund managers, not all investors) (cnyes#1655647) [F-012]. JPMorgan strategists warn that AI stocks no longer rise and fall together and that the divergence recalls the dot-com era (cnyes#1672836) [F-013]. On actuals: SpaceX fell 23% from its first-day close of US$161 over its first 27 trading days and has traded below its US$135 offer price (reported 2026-07-24) (cnyes#1692245) [F-007]. This card lists both sides and adjudicates neither.

Series and context (each source dated; sentences not merged)

Supply absorption and the 2027 lockup wall: The main article carries Goldman's estimate that total 2026 US equity issuance (IPOs, follow-ons, convertibles, SPACs) will be about US$700 billion — only about 1% of Russell 3000 market capitalization, on par with the 2015-2019 average, below 2021's roughly 1.5% and far below the dot-com peak's roughly 2%; on the demand side, S&P 500 buybacks in Q1 2026 rose 4% from Q1 2025, announced buyback authorizations in 2026 to date total US$960 billion (a record per the report, including NVIDIA's recent US$80 billion addition), and Goldman estimates about US$1.3 trillion of total 2026 buybacks; Goldman flags that the real test arrives in 2027 — traditional lockups run about six months, and the lockup shares of 2026's large IPOs unlock through 2027 (cnyes#1692211) [F-004]. Live precursor: SpaceX unlocks in stages from August 2026, with some shares unlocking after the 2026-08-04 Q2 earnings release and Musk's stake locked until June 2027 (cnyes#1692245) [F-007].

The AI boom spreads from equities to credit: The main article carries Goldman's estimate that the five hyperscalers — Microsoft, Alphabet, Amazon, Meta, Oracle — will spend a combined US$5.8 trillion of AI capex across fiscal 2025-2030; as of 2026-07-18, AI-related investment-grade financing globally exceeds US$411 billion; the five have issued US$194 billion of investment-grade bonds in 2026 to date, above the US$108 billion for all of 2025, with about one-third completed in non-dollar markets including euro, sterling, Swiss franc, Canadian dollar, and yen (cnyes#1692211) [F-005]. Same-day companion piece: Mizuho Securities estimates that in 2027 (the report's "next year") hyperscaler capex will exceed their combined free cash flow for the first time; S&P Global cut Oracle's rating to BBB- earlier in July 2026 (cnyes#1692207) [F-014]. Rates backdrop: the 30-year Treasury yield rose to 5.146% on 2026-07-22, its 12th straight day above 5%, the longest run since 2007; BondCliQ data show the six major AI-infrastructure issuers' outstanding bonds in 2026 approaching US$500 billion (as of the 2026-07-23 report) (cnyes#1661102) [F-015]. Case in point: Alphabet's 2026-07-22 earnings raised full-year 2026 capex guidance to US$195-205 billion from the prior US$180-190 billion (given at Q1 2026 earnings), with Q2 capex of US$44.9 billion, up 100% from Q2 2025 (cnyes#1655873) [F-016].

The Taiwan angle (where the hardware supply chain sits): At the "Fitch on Taiwan 2026" forum (Storm Media, reported 2026-07-24), Fitch Ratings said that if enterprise AI revenue and productivity gains fail to keep pace with hyperscaler spending on GPUs, data centers, and power, large cloud providers would first squeeze more from existing compute, adjust pricing, and cut operating costs, then slow capex and build-out schedules; once investment cools, memory — the most cyclical — and semiconductor equipment, a leading indicator, would feel the shock before foundries; Bloomberg Intelligence senior analyst Tseng Hsu-liang said "AI capex is still in an acceleration phase, not at its peak" (as of the 2026-07-24 report) (storm#1671461) [F-017].

F-Units

F-001: Goldman "Top of Mind" report (relayed by Anue, 2026-07-25): 2026 US IPO proceeds estimated up to US$225 billion, a record high approaching twice the 2021 record; Goldman chief US equity strategist Ben Snider says 2026 year-to-date proceeds have passed US$125 billion, above the 2021 full-year record of about US$120 billion; US$225 billion is a full-year estimate "counting deals above US$25 million each"; Goldman's view is that past IPO-bubble alarms have not yet sounded (the report's thesis)

• source: cnyes #1692211
• source_url: https://news.cnyes.com/news/id/6545671
• confidence: medium
• basis: news_aggregation
• period: Forecast target full-year 2026; "past US$125 billion" is 2026 year-to-date (as of the 2026-07-25 report); comparison base 2021 full year
• caveat: A private investment bank's forecast relayed by media — not an official statistic, not realized proceeds; "no bubble alarm yet" is the report's thesis, not market consensus (see §Scope)

F-002: Deal count and structure (main article): roughly 53-60 US IPOs completed in 2026 to date, up over 50% from the same period of 2025 (source wording "年增逾50%"), the strongest start since 2021, yet far below the 25-year median of about 100 per year, 2021's 250-plus listings, and the nearly 400 of the 1999 dot-com peak; the article states the boom's defining feature is a handful of mega-deals lifting total proceeds; three drivers = valuations recovering after the rate-hike cycle ended, private equity's unrealized portfolio value at a record US$4 trillion (as of the report) accelerating the listing queue, and the AI boom being itself a capex boom

• source: cnyes #1692211
• source_url: https://news.cnyes.com/news/id/6545671
• confidence: medium
• basis: news_aggregation
• period: 2026 year-to-date (as of the 2026-07-25 report); comparison bases = same period 2025, 25-year median, 2021, 1999
• caveat: Deal count is the report's approximate range (53-60); the "three drivers" are the report's analyst attribution, not statistical causality (see §Scope)

F-003: Bubble-indicator set (main article, per Goldman's report): median EV/sales of recent US IPO companies about 5x — above the 30-year median of 4x, below 1999's 9x / 2021's 7x / 2020's 11x; 43% of 2025-listed companies turned profitable in their first post-listing quarter, above 1999's 28% and 2021's 24%; normal US IPO first-day gains run about 15%-20% from offer price to first close, first-day doublings were common in 1999, and extreme pops are currently infrequent barring a few cases; the University of Florida's Jay Ritter notes heavy issuance is historically associated with lower subsequent market returns but the indicator's predictive hit rate is only about 52%

• source: cnyes #1692211
• source_url: https://news.cnyes.com/news/id/6545671
• confidence: medium
• basis: news_aggregation
• period: Valuation and profitability comparison bases = 30-year median, 1999, 2020, 2021, and the 2025 listing cohort (as of the 2026-07-25 report)
• caveat: All are Goldman-report indicators relayed by media; Ritter's 52% is his own research claim (see §Scope)

F-004: Supply-demand and the 2027 unlock (main article): Goldman estimates 2026 total US equity issuance (IPOs, follow-ons, convertibles, SPACs) at about US$700 billion — about 1% of Russell 3000 market cap, on par with the 2015-2019 average, below 2021's ~1.5% and far below the dot-com peak's ~2%; Q1 2026 S&P 500 buybacks rose 4% from Q1 2025 and announced strategic M&A value doubled from Q1 2025; 2026 year-to-date announced buyback authorizations total US$960 billion (a record per the report; NVIDIA recently added US$80 billion); Goldman estimates about US$1.3 trillion of total 2026 US corporate buybacks; foreign ownership of US equities rose from 6% in 1995 to 18% as of the report; Goldman flags the real test in 2027 — traditional lockups run about six months and the lockup shares of 2026's large IPOs unlock through 2027, potentially expanding free float substantially

• source: cnyes #1692211
• source_url: https://news.cnyes.com/news/id/6545671
• confidence: medium
• basis: news_aggregation
• period: 2026 year-to-date and Q1 2026 (comparison bases stated in-sentence); the unlock test points to 2027
• caveat: All are Goldman estimates or report-carried figures relayed by media; "record" is the report's wording (see §Scope)

F-005: AI credit markets (main article, Goldman estimates): the five hyperscalers — Microsoft, Alphabet, Amazon, Meta, Oracle — are estimated to spend a combined US$5.8 trillion of AI capex across fiscal years 2025-2030; as of 2026-07-18, global investment-grade AI-related financing exceeds US$411 billion, leveraged-finance issuance exceeds US$77 billion, and AI-related debt exceeds 15% of 2026 year-to-date total issuance in some markets; the five have issued US$194 billion of investment-grade bonds in 2026 to date, above 2025's full-year US$108 billion, about one-third in non-dollar markets (euro, sterling, Swiss franc, Canadian dollar, yen); tech's share of 2026 year-to-date US-dollar investment-grade supply is 20% (a record per the report); Goldman calculates that even levering the five up to big-US-bank levels would free only about US$510 billion of additional capacity; private markets have completed over US$140 billion of data-center financing since early 2025, and Goldman estimates broad infrastructure assets could top US$3 trillion by 2030

• source: cnyes #1692211
• source_url: https://news.cnyes.com/news/id/6545671
• confidence: medium
• basis: news_aggregation
• period: As of 2026-07-18 (AI-related financing); 2026 to date (issuance); estimate window fiscal 2025-2030
• caveat: All are Goldman estimates or calculations relayed by media, not official statistics (see §Scope)

F-006: CNA report of 2026-06-12 (SpaceX pricing day): Goldman projected 2026 IPO proceeds could expand to US$160 billion, up severalfold from the prior year (2025) and a record; the same article carried SpaceX's IPO pricing at US$135 per share, 555.56 million shares sold for US$75 billion — the largest IPO raise on record — a US$1.77 trillion valuation, and underwriting by Goldman Sachs, Morgan Stanley, BofA Securities, Citigroup, and JPMorgan

• source: cna #945717
• source_url: https://www.cna.com.tw/news/ait/202606120026.aspx
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-06-12; Goldman's forecast target was full-year 2026 (the estimate as of that report date)
• caveat: US$160 billion is the media-relayed Goldman estimate as of the 2026-06-12 report — a different point-in-time estimate from the US$225 billion in the 2026-07-25 report; this card lists both and computes no delta (REVISE prior value; see §Scope)

F-007: SpaceX post-IPO performance and lockup (Anue, 2026-07-24): over its first 27 trading days the stock fell 23% from its first-day close of US$161 and more than 40% from its US$225.64 peak; the average 27-day return across the 955 IPOs in the comparison set was 0.8%; per Barron's analysis, among US IPOs with at least US$1 billion market value since July 2009, SpaceX ranks in the bottom 10%; on 2026-07-23 it rose 2.6% to US$118.24, below the US$135 offer price; large lockup tranches unlock in stages from August 2026, some after the 2026-08-04 Q2 earnings release; Musk's shares are locked until June 2027; Musk and other major pre-IPO holders together hold about 7.8 billion shares, about 60% of the 13 billion shares outstanding post-IPO; lead underwriter Goldman Sachs may waive parts of the lockup at its discretion before expiry

• source: cnyes #1692245
• source_url: https://news.cnyes.com/news/id/6544028
• confidence: medium
• basis: news_aggregation
• period: First 27 trading days after the June 2026 listing; price as of 2026-07-23; unlocks from August 2026; Musk locked until June 2027
• caveat: Performance statistics are Barron's analysis relayed by Anue; the unlock schedule is the report's stated plan and actual selling pressure is undetermined (see §Scope)

F-008: SK hynix (Anue, 2026-07-10): raised about 40 trillion won (about US$26.5 billion per the report) via ADRs, trimmed from the previously announced 43 trillion won — the world's second-largest equity offering, behind only SpaceX's US$75 billion of June 2026; 177.9 million ADRs priced at US$149 each, subscription more than 7x covered; the report said it was set to become the largest-ever US IPO by a foreign company, surpassing the US$21.8 billion Alibaba raised in its New York IPO

• source: cnyes #1406784
• source_url: https://news.cnyes.com/news/id/6530389
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-07-10 (ADR issue day); "previously announced 43 trillion won" refers to the earlier point in time as carried by the report
• caveat: Amounts and record phrasing are as carried by the report; the won-dollar conversion is the report's own and this card performs no independent conversion (see §Scope)

F-009: Anthropic (Anue, 2026-07-16; Bloomberg citing people familiar): reportedly targeting completion of an IPO as early as October 2026, which would put it ahead of OpenAI to public markets; valuation rose to US$965 billion after its May 2026 funding round, which the report says surpassed OpenAI for the first time; Morgan Stanley, Goldman Sachs, and JPMorgan selected as joint lead underwriters; people familiar note timing and deal details may change with market conditions

• source: cnyes #1560648
• source_url: https://news.cnyes.com/news/id/6537377
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-07-16; valuation as of the May 2026 round; "as early as October 2026" is a reported plan
• caveat: All are media reports citing people familiar; timing and details may change (see §Scope)

F-010: OpenAI (CNA, 2026-06-26): The New York Times, citing three people involved in company discussions, reported OpenAI is weighing delaying its IPO to 2027 because CEO Sam Altman is holding to a US$1 trillion valuation; Reuters reported OpenAI has confidentially filed for a US IPO with a target valuation of up to US$1 trillion, and CFO Sarah Friar told some colleagues the company targets a 2027 listing

• source: cna #1232005
• source_url: https://www.cna.com.tw/news/ait/202606260185.aspx
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-06-26; "2027" is the direction of internal discussions as of that report
• caveat: All are wire reports citing insiders/people familiar, relayed by CNA (see §Scope)

F-011: Same-topic roundtable and Apollo's caution (Anue, 2026-07-24): Goldman forecasts 2026 US IPO proceeds will reach US$225 billion, a record (same value as the main article, slightly stronger phrasing; the main article says "estimated up to"); Acadian's Owen Lamont laid out an updated "Four Horsemen of the Market Bubble" = elevated valuations, "bubble belief" (investors knowingly riding rich valuations), aggressive corporate share issuance, and heavy capital inflows — saying the current IPO boom meets at least one, but IPO booms often run for years and more likely mark a bubble's beginning than its end, and noting first-day pops are currently uncommon; Goldman's Ben Snider said US equity market value is about US$75 trillion versus an estimated US$700 billion of 2026 equity issuance, a limited share; Jay Ritter said US companies have returned about US$1.6 trillion to investors via buybacks and dividends in recent years; Apollo chief economist Torsten Slok said US IPOs since 2019 have on average underperformed the broad market over their first three years; the S&P 500 is up 9.5% in 2026 to date (as of the 2026-07-24 report), the Renaissance IPO ETF up 17.5% over the same span, that ETF does not hold SpaceX, and SpaceX recently traded below its IPO price

• source: cnyes #1692244
• source_url: https://news.cnyes.com/news/id/6544419
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-07-24; index returns are 2026 year-to-date as of that report; Slok's baseline is since 2019
• caveat: Roundtable statements are individual/institutional views relayed by media; this unit's role for the US$225 billion figure = CORROBORATE (a second, independent report carrying the same value) (see §Scope)

F-012: Countervailing indicators (Anue, 2026-07-22): S3 Partners says total short interest in S&P 500 constituents is about 3.7% of free float, the highest since S3 began tracking in 2010; Goldman research cited in the report puts the pre-2010 peak in the 2008 financial crisis at a median of about 3.8%; BofA's July 2026 fund manager survey: 82% of respondents called global chip stocks the most crowded trade, and roughly half of surveyed managers called an AI bubble the market's biggest tail risk, up from 28% in the June 2026 survey

• source: cnyes #1655647
• source_url: https://news.cnyes.com/news/id/6542057
• confidence: medium
• basis: news_aggregation
• period: Short-interest data as of the 2026-07-22 report; BofA survey July 2026, comparison base June 2026 survey
• caveat: The 82% and "roughly half" populations = fund managers surveyed that month (a sample, not all investors); short positions include hedging, and the report itself notes shorting is a common hedge for net-long investors (see §Scope)

F-013: JPMorgan divergence warning (Anue, 2026-07-24, relaying MarketWatch): JPMorgan strategists warn AI stocks no longer move together and the divergence recalls the dot-com era, with the risk of "a more worrying wave of position unwinds"; the Roundhill Magnificent Seven ETF is up only about 1.5% in 2026 to date (as of the 2026-07-24 report) while the Philadelphia Semiconductor Index is up more than 70% over the same span; JPMorgan's Jason Hunter says the SOX has corrected 20% since early July 2026, notes a similar convergence appeared in Q2 2000 that marked that cycle's top, and says the next few weeks of price action are critical

• source: cnyes #1672836
• source_url: https://news.cnyes.com/news/id/6544423
• confidence: medium
• basis: news_aggregation
• period: ETF and SOX returns = 2026 year-to-date as of the 2026-07-24 report; the 20% SOX correction = since early July 2026 (Hunter's statement in the same article)
• caveat: "Up more than 70% over the span" and "corrected 20% since early July" are two different-period figures in the same article, not a contradiction — each is transcribed with its own time base; technical levels and scenarios are the JPMorgan strategists' views (see §Scope)

F-014: AI credit-pressure corroboration (Anue, 2026-07-25): Mizuho Securities estimates that in 2027 (the report's "next year") hyperscaler capex will exceed their combined free cash flow for the first time, implying heavier reliance on external borrowing; S&P Global cut Oracle's rating to BBB- earlier in July 2026 (the report's "earlier this month"), one notch above junk; people familiar say Meta is arranging financing for an approximately US$12 billion Texas data-center project at rates expected to exceed its prior comparable deals

• source: cnyes #1692207
• source_url: https://news.cnyes.com/news/id/6545672
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-07-25; Mizuho's estimate targets 2027; Oracle downgrade earlier in July 2026
• caveat: Mizuho's figure is an estimate and the Meta financing is a people-familiar report, both media-relayed (see §Scope)

F-015: Rates backdrop (Anue, 2026-07-23): the 30-year US Treasury yield rose 1.6 basis points to 5.146% on 2026-07-22, its 12th consecutive day above 5%, the longest run since 2007; US CPI rose 3.5% in June 2026 from June 2025; BondCliQ data show outstanding bonds of the six major AI-infrastructure issuers — Microsoft, Amazon, Alphabet, NVIDIA, Meta, Oracle — approaching US$500 billion in 2026 (as of the 2026-07-23 report)

• source: cnyes #1661102
• source_url: https://news.cnyes.com/news/id/6543320
• confidence: medium
• basis: news_aggregation
• period: Yield as of 2026-07-22; CPI is June 2026; outstanding bonds as of the 2026-07-23 report
• caveat: Dow Jones Market Data and BondCliQ are third-party data providers relayed by media (see §Scope)

F-016: Alphabet capex (Anue, 2026-07-22 earnings coverage): full-year 2026 capex guidance raised to US$195-205 billion from the prior US$180-190 billion (given at Q1 2026 earnings); Q2 2026 capex was US$44.9 billion, up 100% from Q2 2025; the same article carried Alphabet booking about US$99.03 billion of equity-investment gains in Q2, mainly from the rising value of its Anthropic and SpaceX stakes, and noted SpaceX's share price had come off since its listing

• source: cnyes #1655873
• source_url: https://news.cnyes.com/news/id/6543260
• confidence: medium
• basis: news_aggregation
• period: Q2 2026 earnings (reported 2026-07-22); capex growth comparison base is Q2 2025
• caveat: Company earnings figures relayed by media; this site has not checked the original filings; equity gains include unrealized value changes (volatility risk noted in the report itself) (see §Scope)

F-017: Taiwan angle (Storm Media, 2026-07-24; "Fitch on Taiwan 2026" forum): Fitch Ratings said that if enterprise AI revenue and productivity gains fail to keep pace with hyperscaler spending on GPUs, data centers, and power, large cloud providers would first raise utilization of existing compute, adjust pricing, and cut operating costs, then slow capex and build-out schedules; once investment cools, memory (the most cyclical) and semiconductor equipment (a leading indicator) would feel the shock before foundries; Bloomberg Intelligence senior APAC tech-hardware analyst Tseng Hsu-liang stressed "AI capex is still in an acceleration phase, not at its peak" (as of the 2026-07-24 report) and said talk of an AI investment retreat is premature

• source: storm #1671461
• source_url: https://www.storm.mg/article/5534086
• confidence: medium
• basis: news_aggregation
• period: Reported 2026-07-24 (forum remarks)
• caveat: Fitch and Bloomberg Intelligence positions are institutional/analyst views and scenario reasoning, not statistical facts, relayed by Storm Media (see §Scope)

J-Units

J-001: Official-anchor master caveat: the selector flags official_count_verified=0, official_count_raw=0, official_attribution_unverified_count=0 — zero verified official anchors on this card. Goldman Sachs is a private investment bank; every "Goldman estimate/report/calculation" figure is as carried in media-relayed Goldman research or roundtables, is not an official statistic, and must not be treated as official endorsement; basis is news_aggregation throughout. Any citation of any number on this card must carry the "Goldman forecast / as reported; not independently verified by this site" caveat; figures from different sources stay in separate sentences, and this card fabricates no cross-source comparisons

• confidence: high
• basis: news_aggregation

J-002: Forecast-evolution handling (REVISE layer): Goldman's 2026 US IPO proceeds estimate — CNA's 2026-06-12 report carried US$160 billion; Anue's 2026-07-24 and 2026-07-25 reports carried US$225 billion; this card lists each with its report date, computes no delta, generates no "upgraded by N%" figure absent from the sources, and does not assert an "upgrade" action on Goldman's behalf; the checkable textual fact is only that the estimates carried at two report dates differ. A further caliber note: the US$225 billion figure is explicitly an estimate "counting deals above US$25 million each," while the report carrying US$160 billion states no deal-size threshold — the two estimates' statistical calibers are not necessarily identical and must not be directly subtracted or turned into a growth rate. The US$225 billion value is carried by two independent reports (cnyes#1692211, cnyes#1692244) = CORROBORATE, with the phrasing-strength difference ("estimated up to" vs. "will reach") transcribed

• confidence: high
• basis: news_aggregation

J-003: View-attribution caveat: "bubble warning signs not yet present" = Goldman's report thesis; "meets at least one Horseman" = Lamont's personal view; "about 52% hit rate" = Ritter's research claim; "IPOs since 2019 underperform over three years" = Apollo's Slok's statistical claim; "position-unwind risk" = JPMorgan strategists' view; "memory and equipment most sensitive" = Fitch's scenario reasoning — each view belongs to its person or institution, relayed by media; the BofA survey's 82% and "roughly half" have as their population the fund managers surveyed that month (a sample, not all investors); the S3 short-interest population = free float of S&P 500 constituents. This card lists the sanguine and the cautious side by side, adjudicates neither, and elevates neither to market consensus

• confidence: high
• basis: news_aggregation

P-Units

P-001: SpaceX lockup: some locked shares begin unlocking after the 2026-08-04 Q2 earnings release, with staged releases from August 2026 (reported schedule) — actual selling pressure and price reaction to be verified (open)

P-002: Listing-queue timing: Anthropic reportedly as early as October 2026, OpenAI reportedly 2027 (both unconfirmed reports); Goldman's US$225 billion full-year estimate versus actual full-year 2026 proceeds — verifiable at end-2026 (open)

P-003: The 2027 unlock wave (the main article's Goldman flag: 2026 large-IPO lockup shares unlock through 2027) and Mizuho's estimate that hyperscaler capex first exceeds combined free cash flow in 2027 — verifiable in 2027 (open)

§Scope (source limitations, consolidated once)

• Official anchors: official_count_verified=0, official_count_raw=0, official_attribution_unverified_count=0. Every figure on this card is as carried in media-relayed Goldman research/roundtables, wire reporting, or third-party data providers (S3, BondCliQ, Barron's, etc.); this site has not independently verified any of them. Goldman, Mizuho, BofA, JPMorgan, Apollo, and Fitch are all private financial institutions; their research and estimates are not official endorsements. Basis is news_aggregation throughout.
• Forecast vs. actuals: US$225 billion = Goldman's full-year 2026 estimate (counting deals above US$25 million each); the realized portion = the reported "past US$125 billion" for 2026 to date. The two are not interchangeable.
• Forecast evolution: US$160 billion (2026-06-12 report) and US$225 billion (2026-07-24 and 07-25 reports) are Goldman estimates carried at different report dates; this card lists both with dates, computes no delta, and does not assert an "upgrade" action on Goldman's behalf; moreover, the US$225 billion figure carries an explicit "deals above US$25 million each" threshold while the US$160 billion report states none — the two calibers are not necessarily identical and cannot be directly subtracted.
• View attribution: "no bubble alarm yet" = Goldman's thesis, neither an official determination nor market consensus; Lamont / Ritter / Slok / JPMorgan / Fitch views belong to their respective holders; this card lists them without adjudication.
• Sample scope: BofA July 2026 survey figures (82%, "roughly half") have as their population that month's surveyed fund managers (a sample, not all investors); the S3 short-interest metric's population is the free float of S&P 500 constituents, and short positions include hedging (per the report itself).
• Unconfirmed reports: Anthropic's October IPO, OpenAI's 2027 delay, and Meta's Texas financing are media reports citing people familiar; timing and details may change.
• Conversions: 40 trillion won = about US$26.5 billion is the report's own conversion, transcribed; 億-to-billion renderings are notation transcriptions, not new numbers.
• Time discipline: publication date 2026-07-26 ≥ latest source date 2026-07-25; relative time words ("this year / last year / next year / earlier this month") are converted to absolute years (2026 / 2025 / 2027 / July 2026); every comparative carries its time base in-sentence; the same-article pair in cnyes#1672836 ("up more than 70% in 2026 to date" and "corrected 20% since early July") are different-period figures, each transcribed with its own base.
• Search statement: Phase A full-library search completed by OP; 13 distinct source articles adopted (above the 5-article floor); weak links cut rather than padded (GE Vernova's inflation comment, Waymo operations, World Cup traffic, and other passages without load-bearing relevance to the IPO topic were not cited).

Internal citation chain (links to published cards)

• CONNECT → ANK-2026-07-06-003 "Memory cycle, chapter three — the capital markets chapter: SK hynix launches a roughly US$28 billion US listing on 2026-07-06, poised to be the second-largest equity offering on record" (published 2026-07-06): that card records the launch of SK hynix's roughly US$28 billion US listing plan as of 2026-07-06; this card records, via cnyes#1406784 (reported 2026-07-10), the issue-day raise of about 40 trillion won (about US$26.5 billion per the report), trimmed from the previously announced 43 trillion won — the two point-in-time figures are dated separately and not merged; the deal is the concrete case of the main article's "a handful of mega-deals lifted total proceeds" structure. Links: zh / ja / en
• CONNECT → ANK-2026-07-06-004 "Taiwan equities correction, chapter three — box-range consolidation under tight liquidity: SpaceX's listing and the NVIDIA bond wave weigh on money flows" (published 2026-07-06): that card records the impact of SpaceX's listing and the AI bond wave on market liquidity; this card's coverage of the main article's 2027 unlock thesis, SpaceX's August 2026 unlock schedule (cnyes#1692245), and the AI credit sections (cnyes#1692211, #1692207, #1661102) are follow-on records in the same event chain. Links: zh / ja / en
• CONTEXTUALIZE → ANK-2026-07-02-001 "Taiwan's AI rally — 'raging bull meets gray rhino': the SOX suffers two single-day plunges in one week; Fubon names high leverage as Taiwan equities' biggest risk" (published 2026-07-02): the Taiwan-market counterpart view — this card's storm#1671461 (Fitch's Taiwan forum) sits in the same "Taiwan local view of AI-rally risk" layer as that card; figures in the two cards keep their own sources and are not merged. Links: zh / ja / en

FAQ

Q: Goldman says 2026 US IPO proceeds will hit a record US$225 billion — is that money already raised?

No. US$225 billion is Goldman's estimate for full-year 2026 (per the source text, an estimate "counting deals above US$25 million each," relayed by Anue's 2026-07-25 report), not realized proceeds; the reported 2026 year-to-date actual is "already past US$125 billion," above the 2021 full-year record of about US$120 billion. Goldman's estimate also evolved across report dates: CNA's 2026-06-12 report carried US$160 billion, Anue's 2026-07-25 report carried US$225 billion — this card lists both with their report dates and computes no delta.

Goldman is a private investment bank; all figures are media-relayed and unverified by this site (see §Scope) (cnyes#1692211, cna#945717).

Q: On what basis does Goldman say bubble warning signs have not appeared — and does the market agree?

Goldman's report cites (as relayed): median IPO EV/sales of about 5x, below 1999's 9x / 2021's 7x / 2020's 11x; 43% of the 2025 listing cohort profitable in their first post-listing quarter, above 1999's 28% and 2021's 24%; extreme first-day pops currently infrequent. But that is Goldman's thesis, not consensus: in the same roundtable, Lamont said the IPO boom meets at least one of his "Four Horsemen" and that issuance booms tend to mark a bubble's beginning, not its end; on the market side, S&P 500 short interest is at 3.7% of free float, the highest since S3 began tracking in 2010; 82% of managers in BofA's July 2026 survey called chip stocks the most crowded trade and roughly half called an AI bubble the biggest tail risk (28% in the June 2026 survey); and JPMorgan warns AI-stock divergence recalls the dot-com era. This card lists both sides and adjudicates neither.

(cnyes#1692211, cnyes#1692244, cnyes#1655647, cnyes#1672836)

Q: What are the next verifiable checkpoints?

Near term: SpaceX's partial lockup release begins after its 2026-08-04 Q2 earnings, with staged unlocks from August 2026 (SpaceX fell 23% from its US$161 first-day close over its first 27 trading days and trades below the US$135 offer price). Medium term: Anthropic's reported IPO as early as October 2026 (US$965 billion valuation, unconfirmed), OpenAI's reported 2027 timing (holding a US$1 trillion valuation, unconfirmed), and the comparison of Goldman's US$225 billion full-year estimate against actual year-end proceeds. Longer term: the main article's Goldman flag that 2026's large-IPO lockup shares unlock through 2027 — the real test of market absorption; Mizuho separately estimates hyperscaler capex will exceed combined free cash flow for the first time in 2027.

(cnyes#1692245, cnyes#1560648, cna#1232005, cnyes#1692211, cnyes#1692207)

同事件・三視角 / Three Perspectives on the Same Event / 同一イベント・三つの視点

• 繁體中文
• 日本語
• English

Sources

1. [cnyes#1692211] (main anchor) Anue (Taiwan), "Is the US IPO boom a dot-com bubble redux? Goldman: bubble warning signs not yet present" (translated title; original in Chinese), 2026-07-25. https://news.cnyes.com/news/id/6545671
2. [cnyes#1692244] Anue, "2026 US IPO proceeds head for US$225 billion; manager warns one of four bubble signs is already here" (translated title), 2026-07-24. https://news.cnyes.com/news/id/6544419
3. [cnyes#1692245] Anue, "SpaceX below its offer price — and not done? Down 23% a month after listing, trailing 90% of large IPOs; August lockup expiry the next test" (translated title), 2026-07-24. https://news.cnyes.com/news/id/6544028
4. [cnyes#1692207] Anue, "AI cash burn reaches the bond market: hyperscaler capex has debt investors worried" (translated title), 2026-07-25. https://news.cnyes.com/news/id/6545672
5. [cnyes#1672836] Anue, "Warning light for the AI trade? Dot-com-style divergence returns; JPMorgan: the next few weeks are key" (translated title), 2026-07-24. https://news.cnyes.com/news/id/6544423
6. [storm#1671461] Storm Media (Taiwan), "Fitch on Taiwan 2026: where would an AI bubble start? Cloud providers would cut capex first; memory and equipment chains most sensitive" (translated title), 2026-07-24. https://www.storm.mg/article/5534086
7. [cnyes#1661102] Anue, "Longest run in 18 years: US yields stuck above 5% as the AI bond wave adds pressure" (translated title), 2026-07-23. https://news.cnyes.com/news/id/6543320
8. [cnyes#1655873] Anue, "Earnings: Alphabet beats but sells off — US$205 billion AI capex spooks the market; shares fall nearly 4% after hours" (translated title), 2026-07-22. https://news.cnyes.com/news/id/6543260
9. [cnyes#1655647] Anue, "US short interest at a record: AI bubble worries climb as 82% of fund managers flag chip-stock overheating" (translated title), 2026-07-22. https://news.cnyes.com/news/id/6542057
10. [cnyes#1560648] Anue, "Anthropic reportedly eyeing an IPO as early as October, possibly ahead of OpenAI" (translated title), 2026-07-16. https://news.cnyes.com/news/id/6537377
11. [cnyes#1406784] Anue, "Second only to SpaceX: SK hynix lists in the US tonight to raise US$26.5 billion, poised to top Alibaba's foreign-company IPO record" (translated title), 2026-07-10. https://news.cnyes.com/news/id/6530389
12. [cna#1232005] CNA (Taiwan), "OpenAI holds to US$1 trillion valuation; IPO reportedly delayed to 2027" (translated title), 2026-06-26. https://www.cna.com.tw/news/ait/202606260185.aspx
13. [cna#945717] CNA, "SpaceX prices IPO at US$135 a share, the largest US IPO on record" (translated title), 2026-06-12. https://www.cna.com.tw/news/ait/202606120026.aspx

Anti-fabrication statement: this card carries zero verified official anchors (official_count_verified=0); every figure is as carried in media-relayed Goldman research/roundtables, wire reporting, or third-party data providers, and this site has not independently verified any of them; no number absent from the sources was generated, no forecast delta or ratio was computed, no cross-source comparison was fabricated; Goldman's two point-in-time estimates are listed without adjudication; no Wikidata Q attached (no registry verification performed).