Two of the biggest banks on Wall Street just started letting investors trade the debt behind the AI boom as its own separate risk.
Goldman Sachs and JPMorganChase have each built tradable baskets of bonds from AI-infrastructure companies, the data-center and compute names funding the buildout, so investors can bet on or hedge against that debt directly. Goldman’s basket groups eighteen high-yield issuers at an average yield of 7.45 percent, against 7.3 percent for the broader high-yield market, in sizes from 50 to 250 million dollars.
When banks carve something into its own tradable category, they are telling you the market now sees it as a distinct risk rather than part of the broader credit pile.
For any finance leader raising capital for an AI project, that is a signal worth reading. The cost of that debt is starting to be priced on its own terms, and those terms will only get sharper as more of it comes to market.
Pole vaja. Sul on juba täna olemas alkolukud, AI kaamerad ja kõikvõimalikud sensorid, mis sinu käitumist rooli taga mõõdavad. Pole probleem need ka kontorisse üle kolida. Ja ka kontor peab olemas “sertifitseeritud” jne. Kui juhtub õnnetus, on “süüdlase leidmine” palju lihtsam ja kiirem kui ei kunagi varem.
Üks cräsh, automaatreageerimise näss… See parandatakse, aga kui vastu panna kui palju AI suudab seaduslikult reklaami toota ja seda ajudesse laiali laotada on väga väikse murega…
Kaugjuhitud auto sõitis küll foori keelava tulega Endla tänava ja Sõpruse puiestee ristmikule, kuid pidurdas kohe ning peatus ülekäigurajal ehk enne ristuvaid sõiduradasid.
Autot juhtinud inimene tegi aga otsuse jätkata keelava fooritulega ristmiku ületamist ja kiirendas. Ristuval teel õigetel sõiduradadel paiknenud juhid märkasid ohtu ja pidurdasid, kuid vales kohas ehk ühissõidukirajal paiknenud takso juhi vaatevälja varjasid õigetel sõiduradadel olnud autod ja seetõttu põrkasid autod kokku.
See on nii valus jutt, et ma praeguse palava ilmaga lihtsalt ei suuda… aga las ta olla pealegi, ma katsun ennast korral pidada, et mitte midagi obstsöönset või ebaviisakat sõnuda…
Proovisin siin ükspäev teisese arvamuse saamiseks Claude tasuta versiooni. Ilma liialdamata, ma olin šokeeritud, kui rumal see oli. Siiamaani veits olen.
Selle postituse iva on selles, et kui su arvamus AIst põhineb ainult tasuta saadaoleval, siis sul puudub tegelikult AI kasutamise kogemus. See on nii erinev. Minu ühekordse võrdluse põhjal.
Nii tore kui, mitte kusagil peale fantaasia kasjastuva fakti miljaridsest SKP kasvust, peale ehitatakse teooria, kuidas kohe kasvab veel terve miljardi…
Aga miks piiruda mingi näkase miljardiga? Äkki kohe 10 miljardit või 100 või lajataks kohe triljoni ära?
aastal tõid e-residendid Eesti riigile otsest tulu ligi rekordiliselt 125 miljonit.
Sama hakata tegema AI agentidega, anda neile “isikutunnistus”. Visionäärid missugused, vaataks kuidas see AI regulatsioon hakkab maailmas välja kujunema.
GS hoiatab, et raporteerimisreeglite omapära tõttu on hyperscaleritel triljon $ tulevikuvõlga off-balance sheet… kuidagi tuleks kõik see tagasiteenida (+kasum) kah. Meeletud summad.
5 of world’s biggest tech co’s have racked up $1 trillion in off-balance sheet spending, potentially masking the long-term debt liabilities underpinning the AI boom, Goldman Sachs has calculated. The indebtedness of the so-called hyperscalers could be “understated” as a result of the accounting practice.
Alphabet, Amazon, Meta, Microsoft and Oracle’s combined active and future spending on leasing space in data centres hit $1.5 trillion. About $1 trillion in spending is tied to leases that have not yet started, meaning they do not have to be included in financial statements. Under US accounting rules these are not recognised as an asset or liability from the agreement. Instead, the leases are listed as footnotes in financial reports.
Microsoft has pipeline of $329.1 billion, with leases set to commence 2027-2033. Oracle has $260 billion of commitments, Meta $347 billion. Alphabet reported $85.2 billion of uncommenced leases, while Amazon disclosed $137.2 billion, a figure that also includes some other spending.
Fitch and Moody’s only incorporate operating leases into the adjusted leverage metrics after the lease has commenced. S&P does incorporate the leases into its adjusted debt metrics. “Although these leases may not yet be recognised on the balance sheet, they reflect committed obligations that are unlikely to be avoided without significant cost or disruption,” the company explained.
Hyperscalers have vastly increased their capital spending. They will spend $745 billion this year on chips, power generation and constructing data centres, which is more than double last year and has put pressure on free cashflow, which fell to its lowest level in a decade in Q2/26, and has pushed the giants to tap debt markets. So far this year they have issued a combined $194 billion in debt, up from $108 billion last year.
Amanda Lynam, at Goldman, said: “What they’re doing is getting ahead of a multi-year investment cycle. We think of it as a waterfall of capital, almost exhausting the depth of these different capital markets — debt, equity, their own internal cashflows — to really position themselves to be able to invest in this business, going forward.”
Nvidia on Monday signed deals with 6 of the largest finance houses on Wall Street to mobilise a $500 billion package that will provide capital to help frontier AI labs, enterprises and neoclouds access Nvidia’s chips. -the Times
Kas sain asjast õigesti aru, et GPU’sid hakatakse laenutagatisena aktsepteerima? Et kui 2008. a. finantskriisi sildiks oli subprime mortgages, siis järgmise sildiks tuleb subprime datacenters?
Finantseerimise mudeleid on 1000 erinevat, aint hoone, +jahutus, +varugenekad, +solar, +muu elekter, jne ++++ kombod. vt. siit. Mingis versioonis on vb abs. kõik, sh. prosed, tagatiseks.
Selle originaalloo alt kommidest: "The interesting question is who is in the other side of these commitments - the hyper scalers are in effect lending out future compute. Who is borrowing it from them… 53% has been taken on by just two companies, Anthropic and OpenAI. The first has $46 billion in revs, growing rapidly and has Google’s support. The second has no clear path to breakeven this decade, and has revs which are thought to be less than half Anthropic’s. That means the only way it can pay for these commitments is to raise more and more capital. Every financial crisis begins with an over-extended sub-prime borrower. This time it might be OpenAI. Oracle is also looking very exposed.
Ultimately the agnostic data and AI platforms that run open source will clean up. Business are already catching on to cheaper models from China for many workloads and only a matter of time that models will reach a ceiling where most will catch up.
That will be way before Anthropic and OpenAI makes a penny.
