The 13 AI Giants, from Alphabet to Oracle, Sit on $1.13 Trillion in Cash. They Can Keep Spending on AI even if Markets Get Ugly

They’ve invested huge amounts in AI infrastructure, but they’ve raised cash even faster.

By Wolf Richter for WOLF STREET.

For the economy, the question is this: When will the tsunami of AI spending fizzle? Follow the spending. So if markets get rough, and yields spike, and stocks tank, and raising massive amounts of cash by issuing bonds, loans, or shares becomes difficult or impossible, then how long can the spending continue? That depends on how much cash the companies will still have on hand. And it’s a lot, and it has been growing as they’ve been raising funds faster than they’ve been investing them.

The 13 largest AI-related giants are sitting on $1.13 trillion in cash, cash equivalent, and marketable securities, as per their latest quarterly filings with the SEC. This is an astounding amount of cash. The top three – Alphabet, Anthropic, and Amazon – hold over half of that ($511 billion).

This table shows their prodigious cash, cash equivalent, and marketable securities on their balance sheets as per their latest quarterly filings (10-Qs except S-1 for Anthropic, 8-K for Micron), and the ending date of their most recent quarter that they reported.

Balance sheet: cash, cash equivalent, marketable securities

Billion $

As of:

Alphabet

265

Sep 30

Anthropic

132

Jun 30

Amazon

114

Sep 30

Nvidia

99

Jul 27

SpaceX

90

Sep 30

Meta

82

Sep 30

Microsoft

65

Sep 30

Apple

62

Jun 27

Intel

54

Sep 26

Broadcom

38

Aug 2

Tesla

45

Sep 30

Micron

43

Sep 3

Oracle

37

Aug 31

Total

1,127

If and when markets get messy and ugly enough that raising prodigious amounts of cash through debt or equity sales is no longer easy, these companies will still sit on so much cash that they can keep spending for a while longer, until perhaps markets calm down.

These companies have been raising, and continue to raise, gigantic amounts of cash via debt and equity issuance, and/or they’ve been generating massive cash from their operations, such as Nvidia, and they’ve been doing so even faster than they’ve been investing their cash in the AI infrastructure buildout – circular financing and all.

Several of the giants are now booking negative free cash flows, including Alphabet with a negative free cash flow of $9.1 billion in Q3, meaning that its capital expenditures (mainly the investments in AI infrastructure) exceeded its operating income by $9.1 billion. Anthropic is burning cash hand-over-fist.

But they’ve been burning cash less fast than they’ve been raising cash through debt issuance and equity issuance, and their cash balances have ballooned despite the massive spending.

The storm surge of corporate and investor money into AI infrastructure creates revenues and jobs at companies that are equipping and building the AI infrastructure, and at companies that catch the secondary and tertiary effects. And it ripples out from there.

Pushing this corporate and investor money into the economy has been a potent stimulus that then circulates throughout the economy.

We’ve seen the numbers on Thursday: a GDP measure for the private US economy that excludes government spending, imports and exports – “Real Final Sales to Private Domestic Purchasers” – soared by 4.9% in Q3 annualized, adjusted for inflation, the fastest growth rate since the reopening of the economy in 2021, and the second consecutive quarter, powered by a spiking investments in equipment, intellectual property, and nonresidential structures, and also by robust consumers.

Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the mug to find out how:




To subscribe to WOLF STREET...

Enter your email address to receive notifications of new articles by email. It's free.

Join 13.8K other subscribers

  87 comments for “The 13 AI Giants, from Alphabet to Oracle, Sit on $1.13 Trillion in Cash. They Can Keep Spending on AI even if Markets Get Ugly”

  1. Danlo says:

    Hahahahahhaha. That’s the BEST one I’ve heard in a long time. You think they’re going to burn through their own cash when things get ugly?
    The whole AI edifice is built on other people’s money!

    • Wolf Richter says:

      Most of IT IS other people’s money: they borrowed it, and the got via share sales, from other people. That’s what the entire article is about. Go ahead and read it.

      • Anton says:

        They borrow money from each other and buy each others shares and debt. It’s one giant circle jerk so when one of them gets in big trouble they all do.

    • dang says:

      When will the AI spending fizzle

      I think it happened last week when the AI industrial complex was the pablum subject by the networks news programs

      Thank goodness for the substack journalism and the vibrant podcast community.

      Freedom’s just another word for nothing left to lose

      Janice Joplin

  2. Michael says:

    Do you know how much of that $1.13 trillion is already committed to projects?
    Also, I’m guessing that the cash number will last longer for Alphabet and Amazon than it will for someone like Anthropic since the first two have other sizable businesses with positive margins.
    Thanks

    • Wolf Richter says:

      Commitments? 🤣

      Those with 30-day and 30-day cancellation periods?

      Those that are spread out over the next 10 years and are going to be paid for out of operating cash flow?

      Those that are reciprocal, such as you take my chips and I take your compute and we’re even?

      The relationships these companies have with each other is like a pile of spaghetti ai frutti di mare

      • foo says:

        I don’t see these investments negatively, as long as the companies have cash to lend.

        And these companies *do* have a lot of cash.

        • CasOneTwoSeven says:

          Or…they could return some fraction of that cash to the putative owners of these companies…I know that share buybacks etc are viewed as the devil’s spawn in certain quarters but…

          Substitute the “Metaverse” (rapidly memory-holed despite being the hype d’jour *immediately* before AI) and the “evils” of returning more profits to the shareholders start to look not quite so…evil.

          It is the,

          1) apparent vast scale,

          2) insane speed, and

          3) inability to provide crystal clear use cases with highly likely ROI for ultimate end users (not just circular financing hype-meisters)

          that rightfully fuel the pessimism/cynicism.

        • Wolf Richter says:

          Please revert to numbers (“127”) in your screen name.

        • cas127 says:

          Testing.

          I switched to number-wording because the commenting system was completely refusing to accept the traditional CAS127.

      • Tyler says:

        “such as you take my chips and I take your compute”
        Just curious, does anyone know what this means. That along with tokenization etc. I am familiar with gross sales, profit margin etc. Now we hear a lot about compute and tokenization, yet I have no idea what this means in terms of $$ money. If I take your compute or I tokenize it, does that mean I am making money?

    • foo says:

      Profits for semiconductors are up nearly 200% while the stocks are up a little more than 40% on the year. And they are currently in a drawdown of more than 20% from the highs.

      Differently from the typical bubble where valuations go up, valuations are going DOWN.

      • MM says:

        Early late 2007/early 2008 was like this. Wild swings. A lot of banks having big drops despite not looking bad on paper. I kept buying the dip, thought the volatility was great…..then at one the bottom fell out they just kept going down.

        Not saying it will happen here. It’s an eery market that’s all.

        They have a lot of cash, but execs care almost more than anything about their stock price. So hard to know what will happen. Likely layoffs at least….

      • RH says:

        Future income realizable versus commitments/losses is the issue. While AI advancements are real, the issue is if their subscribers will be willing to pay enough for the kind of profits that their valuations imply, before their equipment depreciation or leasing commitments drive each of those companies under. Doubt it. Some companies will survive but they may transfer ownership via bankruptcy.

        More amusingly, before they go down, they will drive down crypto garbage: the drug trafficking kingpins (like El Chapo), human traffickers, zrls, Meyer Lansky’s Murder, Inc. (whatever they call it now), assassins, etc., that love crypto for its ability to convey wealth deceptively will be very annoyed as advanced AI enables the hacking of more of their wallets. Sure, wallets in external, unplugged, hard disks, etc., will be fine, but if the hard disk crashes, lots of crying ensues.

  3. andy says:

    That seems like a lot, but it’s only 3% of their combined market cap. Having 3% in cash is not bad, but it’s nothing to write home about. And if you think about it, after all these years of mega-hyper-scaling, that 3% in cash is money they had to borrow.

  4. sufferinsucatash says:

    Don’t fear the Bear
    He’s here to stay
    He’ll take down Tech
    Making them his Bear Wreck
    Nothing sweeter than honey
    Oh wait maybe taking Tech’s Money!

    🐻

    Haha

  5. James says:

    From the above, the revenue sources of AMZN, GOOG. MSFT, AAPL seem solid, (could AAPL come in and clean up after the fall) . Anthropic not very much. Ellison reportedly leveraged to the hilt (upcoming fire sale on a Hawaiian island? Haha). NVDA part of the big circular finance game the boyz are playin’?

  6. Robert says:

    It would be interesting to see the total carried debt of these companies as well. I could do a cash advance on my cc to show a healthier checking account balance but net/net I am still only even or even slightly negative.

    • Wolf Richter says:

      You completely missed the point of the article. Did you even read it?

      What it says it that their spending might not slow when they have trouble raising new funds. They sit on this huge cash cushion that they can spend even if markets turn against them.

      The whole point of the article is to lean against the theory that if they slow their fund-raising, it will immediately slow their spending and tank the economy. But that’s not the sequence because they will keep spending their cash cushion, and their cash cushion is still growing at this point.

      • MM says:

        But will they??? Every exec I’ve ever met cares a lot about their share price….

        • yippee says:

          exactly correct. anyone who actually deceives themselves into knowing what these nihilist tech bro founders will do, makes me LOL. smells to me like a combo of roaring 20s, 1929 uh oh moment, smoot hawley tariff debacle, .COM bubble and bursting, and the circular financing of the r/e bubble and bursting of panic of 2007 and 2008…….btw i think 9.11.01 saved the .com bubble burst recession we were feeling in late 2000 and 2001. government started spending massively. an odd note, the best r/e investments of the 21st century were investing in kabul and baghdad. my russian investment banker pals told me their oligarch clients were piling in there knowing the trillions of USA dollars into those cities would make the r/e boom. in spite of the booms from us bombers.

      • Duke DeGuise says:

        A cash cushion, much of which is debt, at a time when a lot of debt is imperiled? That’s don’t sound so hot.

  7. Pain in your ass Steve says:

    These large numbers confuse my tiny brain.

  8. Mitchv says:

    Echos of the dotcom bust. When the stuff hit the fan, hundreds, thousands? of companies disappeared. A few companies like Oracle, Amazon, Apple, and Microsoft survived and eventually thrived. Lots of big name tech companies with pretty solid revenue streams like Intel, Motorola, Cisco, IBM, ATT survived and probably paid their executives really well but stockholders generally got crappy returns for decades. Dangerous times.

    • Glen says:

      Much of that is a given but the .dot bull market ran for 3+ years after most thought it was inevitable to collapse. The question really isn’t around that there will be winners and losers but which ones and more importantly the time horizon. This could easily keep accelerating through 2030 just as much as 2027 could be a reckoning.

  9. Alan says:

    Not surprising, just confirms the magnitude of how these companies all what a piece of A-I world dominance.

    What about Nvidia, the fuel to this fire and how many data centers built or are being built by these companies, including the leased ones?

  10. Gabriel says:

    I did not realize that xAI makes up for 40% of SpaceX.

  11. JFMcNamara says:

    This is still heavily dependent on revenue growth. If revenues aren’t growing, they have to service the debt with that cash. They will cancel the commitments so that they don’t both overbuild and to need to save it for debt service if revenues don’t grow.

    At its core, this whole house of cards is built on existing companies giving AI companies a significant part of their revenue and the belief that we need to build endless compute for AGI and new users. At my to 20 U.S. company, we have strict controls over token use, because it isn’t good enough to completely replace people and its expensive. On the other topic, the models from a year ago are more than enough if we don’t have AGI. Real users aren’t compute constrained and the larger models aren’t improving enough to justify building a massive amount more. The lack of moat means people will just leave if you spend the compute, have a high price, and don’t get enough benefits to justify price.

    I really like AI. I use it every day. It has improved my life. I also don’t think the market is as big as they think because only three people in my family need it enough to pay for it, and I now use a Chinese model for 1/10 the cost and get similar results at home. Despite the cash hoards, there’s plenty that can go wrong.

  12. Andrew pepper says:

    What does a con artist do? He raises more money than he invests. AL looks potentially to me like a huge con. Keep an eye on the owners of AI companies and see how they behave. New huge yatch anyone?

    Perhaps those huge data centers will never all get built because your telephone will house the new AI quickly and all data centers will soon be obsolete.

    • Arizona Slim says:

      ISTR reading that today’s cell phones have more computing power than the NASA Mission Control room had during the 1960s.

  13. yappymutt says:

    When that stock market levy breaks, there won’t be enough fingers and thumbs to block that water break. Today’s Market caps of the biggest publicly held, like nvda has a market cap of Germany’s total annual gdp, apple of California’s annual gdp, and msft bigger than Texas gdp. Simply put in my mind those companies market capitalizations seem way too big and bloated with comparable countries/states gdp’s. Something is gonna give and it’s gonna be a doozy my man. Combined, the top five USA listed public companies are valued around 20 trillion dollars. No wonder the buffet valuation indicator is flashing bright red at 295% over valued historically rite now.

    • andy says:

      Right you are. When this bubble pops, it will make all previous bubbles look like day-trading in Berkshire Hathaway Class A shares.

  14. KIMS says:

    Anthropic is preparing for IPO. And it has not made profits yet.
    How come this private company can have such a huge cash unlike OpenAI?

  15. Delusional about inflation says:

    CDS are rising to insure that debt that provided the cash on their books. They may look to be flushed with cash but they have debt obligations to pay it back. Meta upcoming deal for cash is nothing more than vendor financing “ Meta is planning to raise $95bn in one of the largest-ever stock offerings to fuel its massive AI infrastructure spending, backed by chipmakers Nvidia, AMD and Broadcom, who will act as cornerstone investors” actually META and co remind me of a drug addicts fronting dope to avoid withdrawal. The chipmaker are drug sellers. Who will be the heavy to collect the debt?

    • Wolf Richter says:

      You completely missed the point of the article.

      What it says it that their spending might not slow when they have trouble raising new funds. They sit on this huge cash cushion that they can spend even if markets turn against them.

      The whole point of the article is to lean against the theory that if they slow their fund-raising, it will immediately slow their spending and tank the economy. But that’s not the sequence because they will keep spending their cash cushion, and their cash cushion is still growing at this point.

      • Delusional about inflation says:

        I understood your point, it’s a bullish argument with real facts to support the runaway has more time and distance to go, no offense it’s factual argument or theory. I am 100% committed to the bear case. I am always more than 60% committed to bear case but right now is the time to be peak bearish. Stock up on water and food bearish. Thanks for putting out so much material, you are committed to factual detailed information I am a consumer of the material you put out. I know you have time and energy involved. Thanks for the format to share our thoughts. Sincerely, thank You!
        The crash is coming. We have 28 years of thoughtful action to keep the wealth bubble growing at the expense of fiscal health of we the people, the seeds were planted long ago. It takes time for karma or equilibrium to return to basis, no free lunches of wealth given to the participants for the harm of we the peoples. So many retired Americans own annuities that are loaded with AI debt. Blackstone, Apollo etc stacked them with their dog Sh#% to hard working retired Americans. Life insurance did the same too, The banks are getting richer by taking haircuts fees from all this debt. The American public will be burden with being the heavy! We the people got bent over. Thanks for the I bond story I am getting them for my kids for the holidays, I am not too bearish that I don’t believe in the future, just bearish enough to know what’s coming.

        • Delusional about inflation says:

          Wolf, a while back when you shared; when you compete with China profits tend to go away . That was Gold! AI is a doom game for ROI. One of my kids went to CMU for computer science. Chinese students are capable they work relentlessly. Our best universities in stem, MIT, Berkeley, cal tech ,Carnegie Mellon (not in order) provide them with a great education. We need to make friends with China, cooperate as humans for humanity, it’s going to be impossible to out work them in competition. It’s numbers game. Our American exceptionalism was based on easy credit from the reserve currency. That’s not the case going forward
          Cheers

        • No Rant says:

          Woah! What a rant. Yes, the debt needs be paid back, but it can be paid back from future cash flows and/or future growth…in any case, Wolf’s point here was that the cash on hand is already here to be spent, come hell or high water in the near future. Debt repayment is a question for later, not now.

        • Wolf Richter says:

          OK, but I wasn’t talking about stocks, and my comment wasn’t bullish. I’m bearish out the wazoo on stocks. I’m buying long-term Treasuries to weather what’s coming. In my mind, the only question is when, not if.

          I was talking about the economy being boosted by AI spending, and that this spending and boosting of the economy would continue for a while even if markets lock up on the AI companies, such as by stocks tanking and yields spiking (corporate bonds tanking, spreads widening) so that fund raising would dry up.

        • Gattopardo says:

          “I am always more than 60% committed to bear case”

          Damn, bro, you’re either super young or have had one horrific run. God speed!

        • Delusional about inflation says:

          Thanks for the God Speed, I appreciate that and accept the kind words :) I’ve had great shame in the past with margin calls to cover short positions, true but better than great shame from other activities. it’s been over well over a decade since i made a short position. However, limit buy SPXS at $25.00 looks like it would get filled Monday morning and limit sell SPXS @ $28.75 would be quick 15 % gain in less than week without much risk. stop loss at 23.16 to protect the downside but it won’t be needed. i Remember how good the intoxication felt trading, i am smart enough to know, BRK/A outperformed the spy last month by 10.18% that’s where smart money is trying to hide out, also in BIIB, it outperformed spy last month by 13.31%. BRK/A will see a reversal by Wednesday, and money will flow out of it. not sure about BIIB that may have legs for a while longer. Honestly, I am happy with preservation not growth. will see in the end who is up and who is down, it only a game!!!, The reality is knowing it’s about being present in this earth school adventure program we call life. Tell the truth, love everyone and feed people are words to live by.

  16. Cem says:

    The value add is already apparent. Forget the old reports on value not being seen. The latest models are providing value and leadership is pushing for higher budgets in 2027.

    We’re at the bottom of the first inning with capabilities of the models right now. Anything you read regarding a research paper take note of the models being used. A year ago the models code some, could draft crappy emails, could make a horrific photo/video. Those days are done.

    The last 6months of capabilities are astounding, and these companies are all in. The first one to hit AGI has an asymmetric advantage over all competition. They all know that, they can’t stop spending now even if it’s zero sum and they may not win.

    Coming from a finance bro inside a tech company. Remember the models are general use right now, like the internet was in ‘95 so new use cases are being discovered every week.

    The safe harbor is infra, physical goods (not the $55k base model F-150 with the 5.0 coyote, cloth seats…really?), manufacturing, military contractors. Even luxury is being hit by dropping equities, only ultra luxury is resilient right now.

    Not saying to go out and buy 6 months of MREs but the music is slowing, some (companies) will come out the other side stronger than ever and others will be wiped out.

    Good luck, and remember to have fun.

    • JFMcNamara says:

      This is an interesting take. As a heavy user, I have had a slightly different experience. I haven’t seen many new use cases that couldn’t be handled by Opus 4.6 launched in February. The models are better, but the gains are incremental.

      I don’t think AGI is close. I don’t think simply adding compute will help. Every model really, really struggles with tasks that aren’t derivative. If I feed it years of stock market data, it won’t come up with a unique idea to extract profit. It will apply what it learned in training and spit out (bad) results. The only insights it gives are led by answers to questions you ask it. I felt like I was training it instead of it helping me. It was just a good calculator.

      They are probably going to take everything I asked the model, train on it, then either share it with others or have a “breakthrough” despite saying they won’t. Be careful what you feed the AI. Your proprietary info might be there for everyone in a year, because that non public data is what they need to continue to improve.

      • andy says:

        There will be a steep learning curve on what not to do with AI. People ‘black out’ SSNs or bank account numbers on a PDF file and dump it into an AI without realizing all that sensitive info is still there. Later, these same people will be surprised when the AI transfers their money to some Caribbean island.

        • sufferinsucatash says:

          I won’t even use Monarch money.

          I do everything in spreadsheets and without any reference.

      • BS ini says:

        A trillion dollar available cash with only a few billion a quarter negative cash flow can last a long long time and keep the economic activity high . As you pointed out GDP very positive . At least on my street with upper middle class boomers I see no evidence of slowdowns on spending.
        My neighbor across the street 75 years old wanting to move from East Texas to Denver .
        His financial advisor told him no problem buying his Denver home before he moves and then sell his home here .
        The tax roles value of his home is 1 million 4000 sq ft and he says his newer Dell Webb neighborhood home with a view is 1.7 million but the size is 3000 or smaller but with a basement available for additional space

    • George says:

      I’ve heard “it only got good in the last 6 months” over and over for the last 2 years. That line doesn’t work on me any more.

      As a programmer, it’s impressive what they can do, but they can’t be trusted to work independently. The worst and most common outcome is people using them to write code and english that is incomprehensible.

      • Cem says:

        Agreed, they shouldn’t be let loose and hopefully the final product is good.

        From the SWE side what I’m hearing (from the heads of product) a marked increase in productivity. Of course these men and women don’t push commits without checking, but when you have a 50% increase in productivity. What is your strategy to then help margin?

    • Paul S says:

      Good comment. The MRE line made me wonder about the effects of extreme hacking and more recent digital attacks on our modern integrated and very vulnerable supply lines. What could go wrong? Apparently freezer units were also hit on US military bases as was done in Paris.

      Easy to notice many countries suggesting citizens now bulk up on a few days of emergency supplies. Obviously smart, but what’s with making it official?

      Lots out of our control….but I must say I am sick to death of hearing about AI and watching people mindlessly scrolling on their phones. It’s starting to remind me of the Donald Sutherland remake of the Invasion of the Body Snatchers. Do people want to live in this world? Obviously good medical advances will result but mostly all we’ll hear about is money money money.

      • CasOneTwoSeven says:

        “Obviously good medical advances will result but mostly all we’ll hear about is money money money.”

        Indirectly, this is a very good point (regarding medicine…lamenting profiteering is to lament the motivation for 70%-80% of all human activity…not saying it is good or bad…it just…is).

        Back to medicine.

        In Pubmed, we have a closed universe database, fairly well indexed, heavily abstracted, with over *25 million* medical research papers.

        That is an extensive resource very, very much in our current sorta-AI-but-really-basically-Regression-Correlation-of-the-Gods wheelhouse.

        But where are the AI generated cures? Or even advances?

        In three years of hype-meistery and shrimp Jesus/underpants Gnomes oohs-and-aahs?

        Nothing would win more validation/public support than AI derived medical advances…the fact that AI Hollywood use cases get hyped more than already structured medical database use cases…says something about AI sellers’ faith in/honest understanding of their own (heavily, heavily) pitched product.

    • Jeziel says:

      Torsten Slok’s analysis differs here. The AI “payoff” hasn’t really been reflected out of the magnificent 7 tech sector. And most articles I have seen have point out how most big corps are cutting back on token spend, not expanding it.

      https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector

    • Gattopardo says:

      “The first one to hit AGI has an asymmetric advantage over all competition.”

      Color me extremely skeptical of that. Why? The second or Nth to hit AGI will immediately catch up, and maybe have a better mousetrap to sell.

      • CasOneTwoSeven says:

        Because lord knows only the first company to use a telephone or a train captured 100% of the value add and those companies now rule the Earth…

        In a vague, indirect way this “Absolutist Capture” argument reminds me of the similarly titanically wrong-minded “deflation-is-*always*-evil” absolutism.

  17. John Wilkin says:

    It’s great to have lots of cash (debt?) on the balance sheet, but ultimately it will come down to ROI ! It remains to be seen.

    • Wolf Richter says:

      You missed the point or didn’t read the article.

      What it says it that their spending might not slow when they have trouble raising new funds. They sit on this huge cash cushion that they can spend even if markets turn against them.

      The whole point of the article is to lean against the theory that if they slow their fund-raising, it will immediately slow their spending and tank the economy. But that’s not the sequence because they will keep spending their cash cushion, and their cash cushion is still growing at this point.

      • JW says:

        Yes, I understand your point and did read the article. The long term value these behemoth AI tech companies are worth will ultimately be determined by their ROI from CAPEX, which obviously includes cash, debt, etc . Yes, they have a large cash cushion IF their ability to raise funds deteriorates, but then their stock valuations would logically continue to decline, which the shareholders don’t want to see!!

  18. Jeff says:

    My question is the debt. Who are the lenders? If it’s not banks, is it non-bank bank-like entities, insurance companies, pension funds? Who owns the debt? I could ask AI, of course, but I don’t trust the answer!

    • Wolf Richter says:

      It’s all kinds of investors buying the bonds and their newly issued shares. It’s bond funds and pension funds around the world, retail investors, it’s everyone BUT the banks. The banks that issued loans to some of these companies are selling them or are securitizing them.

      • Jeff says:

        Yikes, now I’m sorry I asked AI! But I am pleased that the life insurers that hold our current annuities do not seem to be on the AI’s list (they are mutual, not PE owned).
        AI Overview
        Life insurance and annuity subsidiaries owned by alternative asset managers are the primary entities buying and holding AI-related private credit and infrastructure debt.
        Insurance and financial firms heavily involved in channeling policyholder and annuity funds into AI data centers, graphics processing units (GPUs), and digital infrastructure debt include:
        • Athene (Apollo Global Management): Apollo utilizes its life insurance arm, Athene, to hold large-scale private credit and infrastructure packages—such as backing GPU acquisitions for xAI and participating in massive tech and data center financings (including Meta and Broadcom-related structures).
        • Blackstone-affiliated / Managed Insurance Assets: Blackstone manages extensive pools of insurance assets and has established direct financing partnerships to mobilize capital for AI and digital infrastructure buildouts.
        • Brookfield-owned Insurance Operations: Brookfield owns and operates insurance businesses that channel long-dated policyholder reserves into private credit and infrastructure debt.
        • KKR-owned Insurance Operations: KKR similarly integrates its owned insurance platforms with private market originations targeting AI-related capital expenditure.
        • Other Institutional Debt Originators & Fund Managers: Firms like Nvidia-linked financing partners (including Goldman Sachs and BlackRock), along with credit players like Pimco (though Pimco has actively trimmed portions of its massive Meta and Oracle AI data center debt stakes), have structured or anchored multi-billion-dollar AI debt packages that flow into institutional and insurance-backed portfolios.
        • Credit Risk Insurers: Traditional non-trade credit risk and commercial insurers—such as participants surveyed by Willis Towers Watson and major brokers like Aon (which manages specialized data center insurance facilities)—are heavily exposed to covering default and credit risks on billions in data center and AI project finance, alongside banks like Barclays seeking risk-transfer coverage.
        Key Risks & Regulatory Concerns
        • Private Credit Exposure: US life insurers collectively hold hundreds of billions in private credit and private-label/private-letter-rated bonds, a significant portion of which is tied to data center and AI compute infrastructure.
        • Related-Party Origination: Critics and regulators highlight conflicts of interest where alternative asset managers own both the private equity/credit originators creating the AI debt and the insurance companies buying it with retirement and annuity savings.
        • Regulatory Tracking: According to reports from the National Association of Insurance Commissioners (NAIC) and financial watchdogs, tracking the exact risk profile inside these confidential, private-placement debt piles remains challenging.

  19. Narmageddon says:

    About Alphabet: GOOG holds 265,098M (265B) total in securities, but that includes $75.2B of SpaceX stock (p15 footnote of 10Q from 2026-09-30).

    If SpaceX stock drops, Alphabet’s ability to spend on AI datacenters will be directly impacted, and also the SpaceX stock still has some short-term restrictions on selling.

    See also:

    https://wolfstreet.com/2026/10/29/tech-makes-a-mess-as-wrinkle-start-to-show-markets-are-edgy-now-arent-they/#comment-696073

    • Wolf Richter says:

      Yes, but those stocks are not included in the cash holdings I posted here.

      • Narmageddon says:

        I double-checked GOOG 10Q from 2026-09-30: In p15 table + footnote it is explicit that the 265B includes the $75.2B of SPaceX stock ownership. So at least in the case of Alphabet/GOOG, the “cash” listed in this article is not all cash.

        Whether any of the other companies in this artile’s table also incklude stock ownership (equity) I do not now.

  20. THEWILLMAN says:

    So after IPO Anthropic will have more than double everyone else listed combined.

    • Reticent Herd Animal says:

      Huh?

      Are you maybe confusing cash gained from selling IPO shares with the projected market value of the entire company? Last estimate I read was for about $100 billion cash from the offering. That’s a lot less than $1.13 trillion.

      • THEWILLMAN says:

        Yes. Momentary lapse of reason on my part. Numbers are so crazy these days that it didn’t hit my “wait that number is off” detector before hitting submit.

        • Reticent Herd Animal says:

          Agreed. There’s a lot of extra zeros being conjured in front of decimal points these days. Sign of the times.

  21. Poor Like You says:

    Could mean weathering a DOTCOM bust like storm, or it could mean extending the bubble even longer.

  22. Paul Z says:

    Interesting position the companies are in. Unlimited cash but not enough to buy. Result is grossly inflated pricing. Inevitably Hardware prices will fall back to normal as supply increases. Collateral on all chip backed loans will disappear as much more powerful and energy efficient chips become available at a fraction of the cost. Massive power plants and data centers will be useless relics. My smartwatch has more powerful compute than a system 360 mainframe was but does not require massive power plants. AI will run on a typical desktop with a few terrabytes of RAM. Pricing will fall to whatever competition pushes the software toward, heavily influenced by Chinese model pricing.

    Lots of cash on hand buys time to react, but at some point it just results in a lot of wasted cash burn. Kodak had a massive cash position when digital cameras came out. The minicomputer industry was highly successful until PCs started selling.

    As other people have mentioned, destruction starts at the weakest points: construction funding for datacenters and nuclear powerplants far from funded will be cancelled. Some cloud service providers collapse as demand for their already built compute falls. What do they do with all that money when they realize that the projects they were planning to build no longer make sense or can be built for a tenth of what they were budgeting?

  23. danf-fifty-one says:

    Thanks Wolf. It’s always hard to get a clear picture of the size and strength of the Balance Sheets these companies have.

    They may indeed be able to keep spending.

    If they do, it will boil down to weather that spending produces the revenues they are expecting in the future.

    Anyone who claims to know with certainty, Is kidding themselves. I thank God that I am done working. Just imagine in my lifetime, Software Engineering went from great and fund career to nothing.

    the unintended consequences of all this in economy, society, politics, war and peace are completely unpredictable.

    We are on the roller coaster thats still clunk, clunk, clunking up to the summit of that first hill. The ride is just starting.

  24. Robert Tonkavich says:

    As always, nice article with intelligent information. An easy tool to find guidance.

  25. Kenny Logouts says:

    Will they really spend it though? Doesn’t that cash form part of the value of their stock?

    AI seems to have not really moved on for some time now, it’s just bigger LLMs eating their own data, any data, increasingly trashy data.
    What we have right now via agents seems pretty good, and not sure blowing another $1Tn is going to buy them a USP product that everyone must have at any cost.

    So will investors want them to spend that $1Tn, and keep buying their stock even as that cash dries up, at higher and higher prices, or an ROI increasingly unlikely to ever come.

    What are we at now? Every man woman and child spending $1,000 a year each for the next decade on AI to USA companies?

    GLWS.

    I’ve managed to use free Chinese models to do my work stuff using AI for the last 6 months, and it keeps getting better.
    I’ve used SOTA models too, paid, and they’re better, but only 6-12 mo ahead. That’s a LOT of cash burn to stay 6-12 months ahead of free products.

  26. Depth Charge says:

    I wonder how much they’d be sitting on if the FED hadn’t printed $10 trillion?

  27. JustAsking says:

    The of balance sheet debt, privately placed by AI companies…..is the new subprime mortgage paper

    “Across the AI complex, there is now more than $4 of off-balance sheet commitments for every $1 of debt and leases actually on the books…”

    Wolf, any comments?

    • Wolf Richter says:

      “commitments” with 30-day cancellation periods, 90-day cancellation periods, or just an MOUs? “Commitments” means nothing unless you can read the entire contract. And they’re not public.

  28. Observer says:

    The Big Tech bosses are sucking up to the president.
    The Big Tech bosses are secretly panicking about their likely incoming bankruptcies. They know that when they need to be rescued, he is da man.

    Any better take on the sucking?

  29. SoCalBeachDude says:

    ZITRON: ANTHROPIC JUNK…

  30. Eddie Money says:

    Well, I am a bit sorry to hear that the party may continue even after the borrowing stops. I wanta see the ____ hit the ___ before I turn senile and can no longer enjoy the fireworks.

  31. Kenny Logouts says:

    It’ll go pop irrespective of that cash pile. None of those contractors making stuff, local governments accepting builds, power companies investing in connections, the list must be huge… will accept the kind of circular hopium promise contracts once the shtf. Everyone will be worried they’ll be the one to not get paid further down the chain, a year, two years, whatever, and they’ll stop it imo (likely tons of break clauses in these deals for stuff not yet in motion)

    They want to keep building, but power connections are years out in some cases.

    Now if they built some power stations? I can see Tesla’s next wheeze being a power provider, but not at 6:1 or 10:1, noooooo, they’re an innovative techs power provider, so they trade at 1,000:1… and people will buy it because they’re fools.

  32. Jeff says:

    This has been a great discussion. I’m not the least concerned about an AI crash affecting the economy directly. The U.S. economy can easily repurpose it’s capital allocation toward other useful things. I am worried about indirect affects:
    1. a perceived wealth effect on consumption — that is, a reduction in consumer spending among folks who believed their AI-related tech assets were correctly valued as real wealth (!) and thus counted on them for spending power. If lots of consumer spending is currently based on this perceived wealth effect, and those guys cut way back after an asset value crash, that could cause a cumulative economic slowdown, and
    2. financial implications of non-payment of AI buildout debt filtering through to state governments or too big to fail private financial entities that would require a federal bailout. Or not. Bailouts or no, the implications and strains of debt non-payment on the economy could be significant.

    • Jeff says:

      Besides, while AI is restructuring the economy fairly dramatically, the impact of AI buildout is partially cancelled out by imports of computer equipment. The AI sector is crowding in as the biggest investment, but the imports associated with all that investment means it doesn’t increase GDP by as much as you’d think.

Comments are closed.