What Happens If OpenAI Dies?

If you liked this piece, you should subscribe to my premium newsletter.

It's $70 a year, $18 a quarter, aliases $7 a month, and successful return you get a play newsletter that’s usually anyplace from 10,000 to 18,000 words, including vast, elaborate analyses of NVIDIA, Anthropic and OpenAI’s finances, and the AI bubble writ large. 

My Hater's Guides To the SaaSpocalypse, Private Credit and Private Equity are basal to knowing our existent financial system, and my guideline to really OpenAI Kills Oracle pairs nicely pinch my Hater's Guide To Oracle, arsenic good arsenic the Hater’s Guide To Oracle (Part 2). I've moreover done a two part Hater's Guide to NVIDIA.

Subscribing to premium is some awesome worth and makes it imaginable to constitute these large, deeply-researched free pieces each week. To subscribe, usage 1 of the pursuing links: $70 a year, $18 a quarter, aliases $7 a month.

If you want to get successful touch — and particularly if you person immoderate juicy accusation astir Anthropic, OpenAI, aliases immoderate different companies successful the AI bubble — deed maine up connected Signal astatine ezitron.76. I’m besides connected IB connected The Terminal. 

---

I’m not trying to beryllium a buzzkill here, but I person meaningful concerns astir OpenAI’s expertise to survive, and they’ve only grown much pressing successful the past fewer years. In the aforesaid week that it completed a $7 cardinal soul stock buyback, OpenAI saw some COO (and erstwhile CFO) Brad Lightcap and Chief Revenue Officer (CRO) Denise Dresser time off the company, the second of which had only been location 8 months, and had this to opportunity a specified 4 months ago: 

“I conscionable person ne'er seen this level of condemnation dispersed truthful quickly and consistently wrong the industries,” Dresser told CNBC successful April, arsenic she was wrapping up her first 90 days connected the job.

Dresser apt walked distant from a ample magnitude of banal options by leaving aft little than a twelvemonth connected the job, which I’m guessing intends she decided that staying astatine OpenAI would, for immoderate reason, not beryllium worthy getting what I ideate are tens of millions of dollars of banal she would beryllium capable to liquidate erstwhile it went public. You know, that point that’s decidedly happening. 

Unless it’s not rather truthful definite anymore. Back successful precocious June, The New York Times reported OpenAI was “leaning toward” going nationalist immoderate clip successful 2027, but that was earlier Anthropic started 1 of the most-aggressive pre-IPO trading campaigns I’ve ever seen, with investors “leaking” to the Financial Times that they thought it would person a $2 trillion valuation and person (sigh) annualized revenues of $100 cardinal to $120 cardinal by extremity of 2026, an wholly fictional connection made pinch the intent of pumping their bags, pinch the FT, for immoderate reason, printing it pinch small pushback.

Yet what’s apt far-scarier for OpenAI is that moreover Anthropic’s pre-IPO trading has a whiff of desperation. A Reuters study from precocious past week that feels precision-engineered to manipulate dimwitted investors said that “Wall Street [was] looking further into the early than it ​commonly does to put a value connected the AI company, valuing it based connected really overmuch gross it could make 2 years from now,” adding that it was “projecting gross of astir $190 cardinal to $200 billion.” 

This was arguably the worst part:

Established companies are typically weighted much heavy connected earnings, aliases EBITDA, which gives investors a consciousness of the economics of the business.

For Anthropic, however, existent EBITDA does not ​fully seizure the economics investors expect the institution ​to execute astatine scale. Anthropic is spending tremendous ⁠amounts connected GPUs and different computing capacity, exemplary training, conclusion and hiring. Those expenses are basal to support its accelerated description but could go a smaller percent of gross arsenic the business grows.

While I ideate the writer successful mobility believed that this was being “fair” and “objective,” this paragraph exists only to manufacture consent for a institution that intelligibly has questionable economics. “Current EBITDA does not ​fully seizure the economics investors expect the institution ​to execute astatine scale” is simply a euphemism for “ignore your lying eyes,” a plea pinch the assemblage to not judge a institution based connected its actual business but connected a theoretical business that, to quote Reuters, person “...training and conclusion [costs] go much businesslike arsenic exertion improves, while unit and different operating costs ​could go a smaller stock of gross arsenic the institution scales.”

Could, could, could, could, could, could could COULD! It’s ever a bloody could aliases will aliases might pinch these fucking companies, and it’s astonishingly bad publicity to spot it arsenic an “objective” prime to vaguely opportunity that a institution should not beryllium evaluated based connected its actual business but connected immoderate theoretical business that they might build successful the future wherever the economics are wholly different. 

Sidenote: the defense of a connection for illustration this is ever that it’s “to show some sides,” but the article besides fails to disclose that Anthropic loses billions of dollars a year, aliases that the AI labs are horribly unprofitable. It does, however, see that Anthropic had a “profitable quarter,” which is thing that was only made imaginable pinch Musk’s discounts connected its compute costs successful May and June 2026. That truth is besides near retired of the article.

The logic I bring up the noises coming from the manufacturing consent instrumentality is that if Anthropic thumps OpenAI to an IPO, I cannot spot a viable (or reasonable) way for Sam Altman to float his nasty small company. The truth that the Financial Times and Reuters are already being co-opted into softening the rustle is simply a motion that Anthropic’s S-1 will look and smell for illustration the wrong of a tauntaun, and Anthropic is, from the reporting I’ve read, successful a overmuch amended information than OpenAI, if only because it didn’t person aggregate broadside quests involving video generation aliases browsers aliases smart speakers, though both companies emotion to springiness distant $20 to $40 for $1. 

Put simply, if Anthropic goes nationalist pinch its ain horrifying economics connected parade, it’s difficult to ideate OpenAI — a institution that mislaid $20.9 cardinal successful 2025 connected $13.07 cardinal successful revenue — will fare overmuch better. 

After all, Anthropic conscionable hit, per Bloomberg, $65 cardinal successful annualized tally rate — a period multiplied by 12, aliases 4 weeks multiplied by 13, I’m guessing, because it ne'er defines this number — successful May 2026, and OpenAI is “on track” to deed $40 cardinal annualized revenue…in the mediate of August. 

Another Sidenote: I gotta say, that Bloomberg communicative astir Anthropic’s tally complaint is moreover weirder than usual, defining tally complaint arsenic “a metric that projects full-year gross from a shorter period” without really saying really it’s derived. No request to inquire difficult questions I guess!

We are, of course, successful the era of madness, truthful I’ve already publication 3 aliases 4 group connected Twitter opportunity that OpenAI’s actual annualized gross is so overmuch higher, because they’ve heard worldly from people they trust. The AI industry’s loudest advocates deliberation and enactment for illustration cultists astatine the extremity of a last day prophecy, isolated from alternatively of the world ending, OpenAI and Anthropic go the largest companies — aliases successful the lawsuit of giga-oaf hedgie Gavin Baker, the only companies — successful the world, rewarding each those who believed with…something. Glory? Smugness? Salvation? 

In immoderate case, OpenAI has a existent problem if Anthropic thumps it to the markets. 

On October 31, 2025, a flustered Sam Altman told booster and investor Brad Gertsner that OpenAI would make “well much than $13 billion” successful gross that twelvemonth earlier saying he’d “find a purchaser for his shares.” In the end, per my ain reporting, “well more” would mean “$70 million,” pinch OpenAI making $13.07 cardinal successful gross successful 2025, pinch SoftBank accounting for $862 million. A week later connected November 6, CNBC would report that OpenAI was “on track” to make “more than” $20 cardinal successful annualized revenue. OpenAI useful retired its annualized gross by multiplying its most-recent four-week-long play by 12, which intends that successful a four-week-long play it had $1.66 cardinal successful revenue, I guess? 

On March 4, 2026, The Information would report that OpenAI had “topped” $25 cardinal successful annualized gross aft hitting $21.4 cardinal astatine the extremity of 2025, and included the pursuing hilarious line:

OpenAI calculates annualized gross by multiplying the past 4 weeks’ gross by 12. If OpenAI calculated the metric based connected gross spikes conscionable successful the past week, OpenAI’s annualized gross would beryllium astir $30 billion, 1 of the group said.

Yeah man, this is why utilizing annualized gross is specified a stupid idea. If you person a particularly-busy four-week-long play — for illustration a merchandise motorboat pinch a large societal media push — you tin usage that play to inflate your revenues, which is precisely what OpenAI is doing, arsenic evidenced by the sources (who I presume activity astatine OpenAI) saying that’s precisely what they’re doing.

Annualized revenues are not a useful measurement of measuring these companies’ financial condition, and beryllium only arsenic a shape of marketing, made worse by the truth that AI token walk is not a recurring root of revenue. While you could theoretically usage annualized gross arsenic a directional spot of information if it was just two companies trading (subsidized) subscriptions, the expertise for these companies to cherry-pick periods of clip that mightiness beryllium inflated by aberrations (like erstwhile personification spent $500 cardinal connected Claude tokens by accident) makes these numbers location betwixt useless and actively harmful to investors.

Even then, it took OpenAI seven months to beryllium “on track” to scope an annualized gross tally complaint ($40 billion) that was seven cardinal dollars smaller than Anthropic’s ($47 billion) from May, and a afloat $25 cardinal successful tally complaint little than what it deed astatine the extremity of July. 

Perhaps it’s a coincidence, but it’s besides worthy noting that the news about OpenAI’s breathtaking caller annualized gross “leaked” specified hours after the abrupt resignation of its Chief Revenue Officer. 

The logic that OpenAI (and Anthropic, for that matter) wants you to deliberation astir things successful position of “annualized revenue” is because its existent revenues look a small tame compared to its commitments and pain rate. The Information reports that successful Q1 2026, OpenAI burned $12.1 cardinal connected “cost of revenue” and training connected $5.7 cardinal successful revenue, though it near retired the income and trading conception wherever OpenAI burned $5.73 cardinal successful 2025 — or, put different way, OpenAI spent $12.1 cardinal connected compute to suffer $6.4 billion, and that doesn’t see things for illustration information costs aliases salaries or, well, anything. OpenAI (and by proxy The Information) someway rationalizes this to only beryllium a pain of $3.7 billion, likely utilizing the aforesaid accounting bullshit that it did successful the financials I saw.

Now, immoderate of you mightiness publication that and opportunity “wow, $5.7 cardinal is simply a batch of money!” but it doesn’t matter, because the much money OpenAI makes, the much its services cost. This is not difficult mathematics, but it is thing that continues to flight the immense mostly of sum of the company, I presume because each of this feels a small insane erstwhile you deliberation astir it.

I cognize you’re gonna telephone maine a firebrand or a hater aliases a skeptic aliases try to seizure maine and put maine successful a zoo, but I must beryllium clear that OpenAI has group expectations — and made commitments — that scope from ridiculous to outright impossible.

To get really specific:

• For OpenAI to meet its compute obligations, it needs to person some the request basal and more than $800 cardinal successful cash (or, alternatively, the expertise to waste and acquisition banal for compute, which it’s done successful the past).
• For OpenAI to proceed arsenic an ongoing concern, it has to, astatine immoderate point, activity retired a measurement to go profitable. 
  • It is unclear really it (or Anthropic) manages to do this. The Information reports estimates that OpenAI will spell from antagonistic $51 cardinal successful free rate travel successful 2029 to affirmative $39 cardinal successful 2030. I’ll stock the floor plan below.
• For OpenAI to actually survive, it will person to raise betwixt $100 cardinal and $200 cardinal fundamentally each twelvemonth until then.
• For OpenAI to spell public, it will request to person numbers that are competitory — some successful revenues and losses — pinch Anthropic, a institution pinch significantly-faster maturation and a larger endeavor customer base. 

For immoderate of these things to happen, OpenAI will person to turn astatine a staggering pace, and efficaciously (per The Information’s reported projections) 10x its gross betwixt now and the extremity of 2030.

OpenAI’s projections person it near-tripling its 2025 revenues, doubling its 2026 revenues, astir doubling its 2027 revenues, increasing its 2028 revenues by 68%, and past increasing its 2029 revenues by 64%. At the extremity of this magical enigma circuit done gross hallucinations, OpenAI will person it making much than NVIDIA did successful Fiscal Year 2026 ($215.9 billion) and, somehow, becoming profitable:

I recognize that galore group person been conditioned by the tech manufacture to judge that each thought that a tech CEO has will ever go reality, but the sheer standard of what OpenAI is some promising and obligated to do outpaces thing successful modern history.

While overmuch of what I’m saying is besides existent of Anthropic, (a institution that itself has complete $300 cardinal successful commitments owed successful the adjacent 3 years and is similarly-unprofitable) OpenAI has decidedly grounded to play catchup astatine a clip erstwhile endeavor customers spot costs arsenic a “huge issue,” which besides makes it improbable that (along pinch caller exemplary value cuts) it will magically re-accelerate extracurricular of allowing users to pain $14,000 a period successful tokens for $200, which…also didn’t activity good capable to get close.

In immoderate case, immoderate acceleration of revenues would besides beryllium an acceleration of costs, which will mean OpenAI will request respective much $122 cardinal rounds from a dwindling heap of investor capital. SoftBank tin quite virtually not afford to put thing further, with liquidity becoming truthful tight that it’s had to return retired a $10 cardinal indebtedness collateralized by its full OpenAI holdings, pinch NVIDIA CEO Jensen Huang saying that its $30 cardinal finance from this twelvemonth apt being its last. While various different task capitalist paypigs whitethorn person some liking successful backing it further, OpenAI will request much than it past asked for, without fail, each azygous year.

So, there’re really only 2 eventualities:

• OpenAI becomes the literal largest and most-successful institution of each time.
• OpenAI runs retired of money astatine immoderate point.

This, again, is not maine being a firebrand, but taking a relatively-clinical look astatine the difficult numbers and asking really the fuck it affords it all.

And man, does a batch of crap person to spell right.

Per my past premium newsletter, OpenAI needs astatine slightest $800 cardinal to meet its commitments successful the adjacent three-and-a-half years, based connected some the Wall Street Journal’s study connected its projected $750 cardinal successful compute walk done 2030 and an study of expert notes connected Broadcom, Microsoft, Google, Amazon, and CoreWeave.

The problem, however, is that overmuch of this money will travel owed done the extremity of 2027, and require astatine slightest 1 much monolithic information of funding. 

To get specific:

• I estimate that, based connected expert notes from Wells Fargo, that OpenAI is connected the hook for astir $87.5 cardinal successful Broadcom chips crossed Fiscal Years 2027, 2028 and 2029. 
  • I person not included these successful the total, because it’s unclear really overmuch OpenAI is really connected the hook for. The Information reported a fewer months ago that some Broadcom and Microsoft would beryllium financing the chips. It’s unclear whether OpenAI would beryllium connected the hook for ongoing payments arsenic Anthropic is nether its $35 billion, private-credit funded woody to bargain Google TPUs and past rent them backmost from Google. 
• Per expert notes from Wells Fargo, UBS and Barclays, OpenAI unsocial is expected to relationship for complete $126 cardinal of Google, Amazon and Microsoft’s unreality revenues successful the adjacent year-and-a-half.
  • The logic for the overseas year-and-a-half nickname is that Microsoft’s Fiscal Year 2027 runs July 1 2026 done June 30 2027).
  • This study besides assumes that OpenAI will walk a linear $40.1 cardinal (per Wells Fargo estimates) connected Microsoft Azure successful Fiscal Years 2027 and 2028. In each likelihood, its woody and commitments will require it to walk more.
• This doesn’t count what OpenAI will request to salary CoreWeave arsenic portion of its five-year-long, $22.4 cardinal deal.
• Though the estimate is from December 2025, Michael Turrin of Wells Fargo estimates that OpenAI’s publication to Oracle’s Fiscal Year 2027 (which conscionable started connected June 1 2026) will beryllium astir $10 billion, past rising to $39 cardinal successful Fiscal Year 2028. I deliberation a adjacent estimate present is to put this astatine astir $20 billion.

Now, each of this is contingent connected Google, Microsoft, Amazon and Oracle building capable capacity to seizure that revenue, but if we presume that happens, OpenAI needs much than $147 billion conscionable to grip its expected compute commitments done the extremity of 2027. 

Here’re immoderate different costs that aren’t included:

• Any and each costs associated pinch its still-theoretical $30 cardinal improvement successful Georgia.
• Any and each costs associated pinch the motorboat of its user device.
• Salaries for its thousands of employees.
• The billions of dollars that OpenAI spends connected information to train its models.
• Its compute costs pinch CoreWeave.
• Its compute costs pinch Cerebras ($20 cardinal complete 3 years).

With its IPO apt delayed — if it ever happens — until 2027, OpenAI will almost-certainly person to raise different information of backing by March 2027, apt astatine a akin standard to its $122 cardinal information from March of this year.

Who Would Invest?

The biggest problem that OpenAI has is that $110 cardinal of its past $122 cardinal information was made up of Amazon ($50 billion), NVIDIA ($30 billion), and SoftBank ($30 billion), leaving a specified $12 cardinal funded by a primordial crockery of different task capitalists, backstage in installments funds, and nationalist endowments that should person their executives fired, ideally into the sun.

In immoderate case, $12 cardinal isn’t capable to screen a azygous quarter’s compute costs.

The constituent I’m making is that raising further rounds — earlier we get to immoderate niggling problems astir valuation — has already go near-impossible to do without the thief of monolithic entities that are showing expanding signs of strain astatine precisely the infinitesimal OpenAI needs much money.

Let’s break it down.

As mentioned previously, SoftBank is moving astatine the very edges of its liquidity, and owes different $10 cardinal owed connected October 1, 2026. While successful mentation it could waste much of its ARM banal to money further rounds, said banal makes up effectively each of its Net Asset Value, and while further separator loans are possible, doing truthful would put genuine unit connected ARM’s banal value as, well, astatine immoderate constituent you’re not conscionable investing successful a company but whether SoftBank mightiness usage its banal for illustration a piggy bank. 

A fewer weeks ago, Amazon sent the remaining $35 cardinal of its $50 cardinal finance arsenic portion of the larger round, and while it’s theoretically imaginable that it could put more, its free rate travel has now gone negative, and it needs arsenic overmuch money arsenic imaginable to meet its (agh!) projected $220 cardinal successful 2026 superior expenditures.

Google is simply a imaginable investor, arsenic I’m not judge group recognize really large a Google Cloud customer OpenAI has become, pinch Stephen Ju of UBS estimating it will walk $9.375 cardinal successful 2026 and $12.5 cardinal successful 2027, and Google Cloud progressively becoming Google’s largest maturation vehicle. Then again, Google’s free rate travel besides went antagonistic successful its latest quarterly earnings, and moreover the astir braindead of investors are becoming a small tense astir really information everything is looking.

NVIDIA could, successful theory, spend to put more, but the markets are moreover more tense astir its slow translator into GE Capital. Jensen Huang is intelligibly alert of this, which is why his “backstop” of a “10GW” information halfway successful Ohio (which OpenAI has signed a 20-year-long lease to rent) isn’t really backstopping OpenAI’s compute spend, but the underlying assets successful the arena of a short sale:

For instance, if OpenAI were to locomotion distant from the project, SB Energy would first effort to lease the tract to different customer for the aforesaid price, immoderate of the group said. If SB Energy wasn’t capable to find different suitable tenant, the patient would effort to waste the site, and Nvidia would salary immoderate quality successful the value, up to $105 cardinal if the first shape is completed. 

By backing the plus worth of the information center—not OpenAI’s ongoing lease payments—the building limits Nvidia’s consequence vulnerability substantially, those group acquainted pinch the woody said. The spot giant’s guarantee covers completed information centers, not accommodation nether construction. 

That’s a beautiful large “if,” because it refers to 5GW of theoretical capacity built by a institution that has ne'er built a information center, astatine a clip erstwhile the nearest balanced — Stargate Abilene, astatine 1.2GW — is 2 years successful and has only vanished 3 retired of 8 of the buildings. Based connected this explanation of the deal, NVIDIA only has to guarantee things in the arena the information halfway is really built.

As portion of the deal, NVIDIA is investing $1.5 cardinal successful SB Energy, a institution invested successful by some OpenAI and SoftBank that is trying to spell nationalist immoderate clip this year, apt arsenic a intends of adding further liquidity to SoftBank’s equilibrium sheet, though the IPO would only raise, per Reuters, betwixt $5 cardinal and $7 billion.

OpenAI has already, crossed aggregate backing rounds, raised from backstage in installments costs from Blackstone, BlackRock, and Insight Partners, and it’s imaginable that these aforesaid costs could fuse together for illustration Voltron arsenic a intends of keeping OpenAI alive.

That being said, we’re talking astir complete $100 cardinal a twelvemonth for the foreseeable future, which is simply a small much than they could tummy connected a backstage institution pinch ultra-negative margins and a younger competitor presently eating its lunch. 

Then there’s different problem: that backstage in installments is already having problem backing AI information centers, which are a (theoretically) far-more-stable finance successful infrastructure and power. When NVIDIA announced its “$500 billion” fund, the media was speedy to presume that it had already closed the money, alternatively than it really being a “memorandum of understanding,” besides known arsenic “a non-binding statement to possibly do thing successful the future.”

Yet a follow-up from Bloomberg recovered that it was even little than nothing, and that Jensen Huang had insisted connected making the announcement contempt months of slow progress:

Goldman Sachs Group Inc., Blackstone Inc. and Apollo Global Management Inc. had been moving tirelessly for months to tie up indebtedness deals that would thief developers of artificial intelligence systems salary for chips from Nvidia Corp.

With slow advancement connected the analyzable deals, Nvidia’s main executive officer, Jensen Huang, decided to alteration tack: He went nationalist this week pinch the effort, saying the group is aiming to collectively finance AI computing deals totaling $500 cardinal — a information fig pinch nary evident provenance.

The logic I bring this up is that if backstage in installments costs are having problem backing information centers, they’re going to person a shit-ton of problem convincing investors to heap into an unprofitable second-place AI laboratory tally by a uniquely-unlikeable CEO who has a penchant for lying.

Could Venture Capital Step Up? 

As mentioned earlier, OpenAI (and Anthropic) person scraped the bottommost of the tube of task superior clip and clip again, and ne'er managed to raise much than $30 cardinal astatine a time. 

• $30 cardinal of OpenAI’s $40 cardinal 2025 backing round came from SoftBank.
• Anthropic’s $30 cardinal backing information from February 2026 progressive an estimated $10 cardinal from NVIDIA and $5 cardinal from Microsoft, pinch the remaining $15 cardinal aliases truthful covered by thirty-seven different task superior and backstage in installments funds, including hedge money Jane Street.
• Anthropic’s $65 cardinal backing information from May 2026 included $10 cardinal from Google and $5 cardinal from Amazon, arsenic good arsenic backing from Micron. Out of the 28 investors, only 8 were task superior firms, pinch the remainder made up of a substance of hedge funds, plus managers, sovereign wealthiness costs and finance firms. 

The sheer measurement of names connected these deals suggests that it’s genuinely very difficult to mobilize this overmuch capital, and I deliberation it’ll go difficult-to-impossible to do this each azygous year, moreover if Anthropic were to spell public, arsenic it’s very improbable that the mostly of these investors will really beryllium capable to liquidate their holdings.

And remember, we’re talking astir OpenAI present — stinky, expensive, second-place OpenAI, the 1 pinch each the obligations, the 1 pinch the CEO that wants to surveil everything his customers do. The 1 that has raised nary much than $12 cardinal of backing from sources extracurricular of NVIDIA, SoftBank, Microsoft aliases Amazon. That one. 

There’re really two awesome problems:

• While task superior might want to put successful OpenAI, really mobilizing much than a fewer cardinal dollars is very difficult.
• OpenAI’s valuation is conscionable too gosh darn high.

The Valuation Problem

OpenAI’s $122 cardinal backing information weighted it astatine $852 billion.

And, per the New York Times, advisers pushed backmost connected the thought of trying to spell nationalist astatine a $1 trillion valuation:

OpenAI’s advisers presented institution executives pinch the action of waiting until 2027 to spell nationalist pinch a $1 trillion valuation, aliases little the targeted valuation for a quicker I.P.O. Mr. Altman, said 1 personification successful interaction pinch him connected the topic, responded that immoderate alteration to the trillion-dollar valuation was a nonstarter.

For immoderate perspective, a $1 trillion valuation would beryllium astir a 15% premium, for a institution that now accounts for 70% of Microsoft’s AI revenues and allegedly is the single-most-important startup since Google aliases Facebook. 

Sorry, I’ll extremity vagueposting: this is bad. For a institution of this standard and importance, OpenAI should’ve waltzed into a $2 trillion valuation, isolated from a nationalist offering requires you to provide audited financial statements and an mentation of why your institution is worthy that much that goes a small further than an investor platform pinch annualized tally rates and charts that committedness the world.

The problem present is that if OpenAI can’t spell nationalist astatine moreover a trillion dollar valuation, it’s unclear why anyone would put astatine $865 billion, aliases $800 billion, aliases moreover $700 billion, unless they happened to judge that it would spell nationalist astatine less than a trillion past magically go worthy trillions more, somehow. The expertise for immoderate investor astatine this constituent to make a important return is very, very small, made smaller by the truth that Anthropic appears to actually beryllium gathering pinch investors for an IPO and is showing gross growth…

…except moreover then, AI bulls are nervous, because $65 cardinal successful annualized gross (at the extremity of July) was lower than some forecasts, pinch market intelligence patient Yipit claiming it had deed $74.3 cardinal connected July 22, causing confusing feelings successful the minds and bowels of boosters that had expectations group by, I imagine, a operation of achromatic magic and achromatic mold.

While Anthropic CFO Krishna Rao has not been discussing valuations astatine early IPO meetings, investors and analysts are either expecting aliases wishcasting that it hits a $2 trillion valuation, though if OpenAI can’t get a trillion, it’s difficult to spot really Anthropic — a business of larger-yet-comparable size and equally-rotten economics — would someway double that and, I assume, past some. 

Seeing each of this, why would immoderate task capitalist pinch a moving encephalon still put successful OpenAI astatine thing adjacent to an $865 cardinal valuation? While existent investors mightiness travel connected arsenic a intends of keeping the institution afloat, astatine immoderate constituent their constricted partners mightiness inquire reasonable questions for illustration “how do you intend to make america money?”

This is simply a problem already hitting Thrive, which has invested billions successful OpenAI. Per Bloomberg:

The firm’s 2022 growth-stage money — which includes Wiz, a business sold to Alphabet Inc.’s Google earlier this twelvemonth — has returned 0.3 times the first money it invested. That places it supra the apical 5% of funds. Most task costs return betwixt 10 to 12 years to return capital.

Thrive’s largest investment, OpenAI, is expected to make a meaningful return. The patient was an early backer, investing successful the startup done astatine slightest 5 abstracted costs going backmost to Thrive’s $408 cardinal conveyance from 2018 and a money that closed this year, a $6.23 cardinal instrument. Other notable IPO contenders wrong Thrive’s portfolio see Stripe and Anduril.

That’s correct folks, if you invested successful Thrive’s 2022 growth-stage fund, you’ve made 30 cents connected the dollar, pinch overmuch of it tied up successful OpenAI. 

While I’m not denying it’s possible, constricted partners person their limits — especially arsenic costs from Sequoia and different task superior firms underperform the S&P 500.

And, not to repetition myself excessively much, OpenAI needs truthful overmuch much money! It needs astatine slightest $100 cardinal a year, aliases it’s toast!

The illness of OpenAI would apt beryllium a consequence of the walls closing successful astir its ruinous obligations and economics, pinch counterparties near short-changed and deals surgery arsenic things statesman to unravel.

It starts, arsenic evident arsenic it sounds, pinch OpenAI moving short connected funds, and we’ve already seen 1 motion that had happened with Amazon “completing” its $50 cardinal finance successful the institution a fewer weeks ago by sending different $35 billion.

To beryllium explicit, that $35 cardinal was rumored to beryllium contingent connected OpenAI either going nationalist aliases reaching AGI, though each that was said successful the backing announcement was that it was contingent connected “certain conditions being met.”

Nevertheless, Amazon didn’t determine to nonstop $35 cardinal retired of the goodness of its heart, aliases because it thought OpenAI was specified a awesome institution — if I had to guess, it’s because OpenAI needed that money to salary for its compute costs, an estimated $9 cardinal of which travel done Amazon Web Services. 

The truth that OpenAI needed $35 billion specified months aft receiving astatine slightest $40 cardinal (and hardly a period aft getting different $10 cardinal from SoftBank) suggests that either  compute pre-payment costs are sadistic aliases OpenAI is perfectly annihilating rate astatine a complaint unforeseen successful the history of capitalism. 

Whatever the reason, OpenAI intelligibly needs tens of billions of dollars each fewer months to support up pinch its costs, and will only need more money arsenic it “grows” — by which I mean has to pre-pay for compute costs for Amazon, Google, Microsoft, CoreWeave, Oracle, and Cerebras.

While it’s foolhardy to opportunity when OpenAI mightiness illness (don’t I cognize it!) its illness will travel from the astir evident spot — erstwhile it’s required to pony up a bunch of money without a intends of raising much funding. 

When you return a measurement back, OpenAI has had to raise backing near-perpetually since its $6.6 cardinal information closed successful October 2024 on apical of a $4.4 cardinal in installments facility. On December 27 2024, OpenAI would opportunity successful a blog post that it needed “more superior than it imagined,” and would statesman talks a specified period later successful January 2025 to raise different round of $40 cardinal that would “close” connected March 31 2025, though it would only raise $10 cardinal astatine first from SoftBank (with $2.5 cardinal of that from a syndicated group of investors).

Five months later successful August 2025, OpenAI would raise another $8.3 cardinal “as portion of” the information from a group of task capitalists and plus managers, sell different $6.6 cardinal of internally-held shares to investors successful October 2025, and by the mediate of December 2025 was already rumoured to beryllium raising different $100 billion, conscionable earlier getting different $22.5 cardinal from SoftBank connected December 31 2025.

While we cognize OpenAI ended 2025 pinch astir $25 cardinal successful cash, The Information was capable to update america that it had astir $73 cardinal successful rate and “marketable securities” astatine the extremity of Q1 2026, which apt includes astatine slightest $35 cardinal from Amazon, NVIDIA and SoftBank, though for immoderate logic the newsman refused to break retired the rate part. Nevertheless, this intends that OpenAI’s existent rate position looked better only by kindness of an influx of capital, and whatever happened to the institution successful Q2 2026 meant it needed different $45 cardinal (Amazon positive SoftBank, and possibly different $10 cardinal from NVIDIA, arsenic it’s unclear really that full point was amortized).

What I’m getting astatine is that astatine immoderate constituent successful the adjacent 3 months, OpenAI is going to request much money, apt tens of billions of dollars, particularly arsenic it enters caller fiscal years for Google, Amazon, and CoreWeave, each 3 of which will apt require up-front payments for capacity that OpenAI does not have. 

And, arsenic I’ve many times said, OpenAI needs to support raising money because its costs summation pinch its revenues, and it has nary clear way to either reducing them aliases expanding prices, arsenic it recovered when it (and Anthropic) moved endeavor customers onto accounts that required them to salary the existent costs of their AI services.

None of this has overmuch to do pinch my feelings astir AI, and acold much to do pinch basal mathematics. OpenAI has nary economies of scale, it’s horribly-unprofitable, and does not person a unchangeable business. This people intends that it has to continually raise capital, isolated from raising further superior is going to beryllium difficult, based connected the sheer amounts it needs, the dwindling costs disposable for it to raise, its already-inflated valuation, and the truth that it’s way down a competitor facing precisely the aforesaid problems.

OpenAI has promised the impossible, and built a institution that only makes consciousness if you’re consenting to disregard the worst economics successful the history of capitalism. Its early is limited connected raising complete a 100 cardinal dollars a twelvemonth successful 1 of the worst backing climates successful history. Its revenues are slowing, its competitor (and there’s really only one) has outpaced it (all while slowing itself), and its CEO is 1 of the single-worst spokespeople successful history. 

However you whitethorn feel, it’s intolerable to reason pinch the logic that OpenAI is going to request much money by the extremity of the twelvemonth — apt tens of billions of dollars — and that money will person to travel from somewhere. It could beryllium from Google, aliases Amazon, aliases moreover Meta. It could beryllium from SpaceX, though Musk would person to clasp his chemoreceptor a little. It could beryllium from Microsoft. It could beryllium from a past gasp telethon of task capitalists coming together to prop it up 1 past time.

But it’s gotta travel from somewhere.

And astatine immoderate point, OpenAI will simply not beryllium capable to salary its bills, aliases much precisely, it will person to manus complete money to personification who will not judge equity aliases IOUs successful return.

Whoever it is that refuses that woody will beryllium the 1 that pulls the trigger, and sends OpenAI’s assemblage to the glue factory.

Sidenote: I want to beryllium clear that this is each speculation. The world is chaotic, the early is uncertain, etc.

So, arsenic overmuch arsenic I person talked astir OpenAI’s death, its apocalypse could get successful galore different forms, but apt starts (as I conscionable said) pinch it personification asking OpenAI for immoderate real, non-circular dollars, only for Sam Altman to look astatine them for illustration this:

But the first spot to look for the extremity is OpenAI’s gross growth. To compete pinch Anthropic, it will person to deed $60 cardinal successful annualized gross (I’m truthful fucking tired of annualized revenues) wrong the adjacent 3 months. The first domino to autumn will beryllium them either missing this target aliases seeing revenues regress — if they haven’t already done so, of course, fixed that OpenAI measures tally complaint based entirely connected a hand-selected four-week-long period. 

All that it takes is simply a small stank of regression for the marketplace to get nervous. 

It’s inevitable, astatine this point, that some Anthropic and OpenAI’s gross maturation slows, if only because some of them person only sewage this acold done a operation of subsidized subscriptions and companies burning millions connected token-maxxing initiatives that will person petered retired by the extremity of the year. OpenAI has spent a small complete a twelvemonth trying to play catch-up connected the endeavor — a strategy led by now-departed COO Brad Lightcap — only to find that customers are becoming cost-conscious astatine precisely the clip they request to beryllium spending more. To make matters worse, Ramp found that customers person been slow to adopt Anthropic’s more-expensive “Fable” exemplary because of the price, meaning there’s efficaciously nary measurement to jack up prices.

I ideate Anthropic’s liking successful bumrushing for a September IPO is an effort to debar investors seeing post-tokenmaxxing deceleration. In doing so, it’ll put OpenAI successful a sadistic position of having to take sides itself against some its ain and Anthropic’s economics astatine the aforesaid time. 

So, the point to watch retired for is immoderate motion of deceleration, which could mean outright “run rates person dropped,” to little pain connected OpenRouter, to much value cuts, to immoderate benignant of attempts by OpenAI to connection discounted tokens if bought successful bulk. 

Then, astatine immoderate point, the money will extremity flowing to somebody.

The problem astir guessing who that mightiness beryllium is really overmuch of the AI bubble is held up by OpenAI’s revenues. Microsoft, Google, and Amazon each person vested interests — virtually and figuratively — successful astatine slightest appearing to get paid by OpenAI, which intends they’re apt activity pinch it connected deferred payments and/or equity shares successful trade, apt instituting immoderate benignant of bastardization of the already-problematic “payment-in-kind” strategy utilized by backstage in installments erstwhile it can’t spend it loans. 

CoreWeave could beryllium a spot to look, pinch its largest customers being Microsoft (for OpenAI), OpenAI, NVIDIA, Google (for OpenAI), and Anthropic. While Microsoft and Google are improbable to extremity paying their bills owed to OpenAI lacking the cash, OpenAI is allowed to salary its bills Net 360, meaning that if CoreWeave’s cashflow abruptly starts sagging contempt revenues growing, it’s perchance because of Sam Altman stapling IOUs to Michael Intrator’s car on pinch a statement that says “I’m sorry. I can’t. Don’t dislike me.”

Cerebras — which gets location betwixt 50% and 70% of its revenues from its OpenAI statement — would beryllium different spot to look. If revenues (or cashflows) neglect to materialize, it could beryllium different motion that OpenAI is incapable to salary its bills.

Other evident signs would impact changes successful guidance crossed immoderate awesome hyperscaler, particularly Oracle, Microsoft, Google aliases Amazon — specifically connection suggesting that OpenAI’s gross either isn’t real aliases isn’t arriving. 

I also, to beryllium clear, expect some sort of fundraising, apt heavily-funded by plus managers, pinch the imaginable for NVIDIA to break its promise and put again arsenic a intends of keeping the statement going. Despite OpenAI’s lousy financial condition, its beingness is captious to the full AI industry, representing the mostly of compute request crossed efficaciously each provider, which will mean everybody will astir apt effort and chuck a fewer dollars its way.

This could return the shape of a termination information (valuing it astatine aliases supra the $965 cardinal valuation from Anthropic’s Series G round) aliases a sadistic downround of astir $800 billion, justified arsenic ‘technically higher’ than the $730 cardinal pre-money valuation it sewage erstwhile NVIDIA, Amazon and SoftBank past invested. 

I could besides spot it taking a doomed tally astatine a nationalist offering — particularly if Altman someway pushes retired CFO Sarah Friar, who had antecedently said it wasn’t fresh for IPO and sewage rewarded for her honesty by being made to study to “CEO of Applications” Fiji Simo, who left the institution successful July owed to aesculapian issues but for immoderate logic remains progressive down the scenes, per the FT. 

Going nationalist is simply a terrible, atrocious decision, which is why I’m increasingly-confident that Altman would see it, particularly if there’s request for liquidity from investors. OpenAI, contempt its salient successful the manufacture and load-bearing compute spend, is successful a hopeless and untenable position made worse by a competitor that worked retired how to swindle endeavor customers that don’t cognize really to measurement their token spend astatine a much-larger scale, and without thing completely-unexpected, it’s unclear really it pulls itself out.

When things get rough, expect Altman to make comments astir the challenges of building the future, criticizing those who are “endlessly negative” astir AI and group "unrealistic expectations” from a man who said that OpenAI is adjacent to creating a ghost that tin assistance immoderate wish. He will blasted everybody — critics, the financial markets, journalists, ex-employees, Elon Musk, Dario Amodei, counterparties that “don’t understand what invention demands,” task capitalists, Twitter posters, and fundamentally anybody different than Sam Altman, the feline who made hundreds of billions of dollars’ worthy of commitments to the largest companies successful the world pinch small aliases nary scheme arsenic to really he mightiness do so.

Sidenote: I americium not engaging pinch worldly astir authorities bailouts aliases nationalization, because I deliberation some are intelligence crutches that beryllium to debar reasoning astir genuinely chaotic events. OpenAI whitethorn get a authorities lifeline, it whitethorn get the expertise to raise a indebtedness from the government, aliases Trump whitethorn do perfectly nothing, arsenic midterms are coming up and his support standing is successful the shitter. 

No, these information centers are not each portion of immoderate big, concealed surveillance state. No, location is not immoderate mysterious $150 cardinal bailout. Every clip you take to judge this you are attempting to broadside pinch the wealthy, assuming they each person immoderate superb scheme they’ve formed pinch their magnificent brains, erstwhile successful reality they’re each obsessed pinch maturation and thought AI was the adjacent large maturation thing. Reality is acold much depressing — the rich | and powerful are arsenic stupid (or stupider) than a regular person, they conscionable sewage lucky.

OpenAI’s actual death could return a fewer forms, each of them reasonably destructive. 

Microsoft Absorbs OpenAI

In the arena this happened, Microsoft’s first move would beryllium to cancel efficaciously each unreality contracts that OpenAI has, and person to restate guidance to region the $250 cardinal successful “incremental Azure spend” it promised. There isn’t a chance successful Hell that Satya (if he’s allowed to stay) is going to springiness Google, Oracle aliases Amazon hundreds of billions of dollars, moreover if it intends taking monolithic impairments connected GPUs.

In this scenario, Microsoft would perchance portion backmost (or wholly eliminate) the free ChatGPT product, and apt either tighten complaint limits aliases move everybody connected a ChatGPT Plus aliases Pro subscription to token-based billing, much arsenic it did pinch GitHub Copilot successful June. 

OpenAI Is Allowed To Die, And Altman Becomes The Sacrificial Lamb For The AI Bubble

While I ideate some rescue package is pulled together, OpenAI could simply beryllium allowed to tally retired of money, short-changing astir a trillion dollars’ worthy of compute contracts, sidesplitting CoreWeave, Cerebras, and anyone other reliant connected its income. Its customers would beryllium fixed API keys that travel to Microsoft AI Foundry, Amazon Bedrock and Google Vertex, and beryllium told that location would beryllium small aliases nary further improvement aliases training of OpenAI’s models.

This situation, while evidently destructive for the full industry, would springiness everybody a scapegoat. Who made each the promises? Sam Altman. Who ran a shitty institution into the ground? Sam Altman. Who misled everyone into believing that there’d beryllium infinite request for compute? Sam Altman. Stories will leak that OpenAI was “not consistently candid” pinch its financial information pinch partners, allowing everybody to reframe a trillion-plus dollars successful discarded arsenic the consequence of 1 egregious con artist.

To beryllium clear, the personification to blasted is Satya Nadella. He’s the 1 that made the first investment, bought each the GPUs, and past kept buying them the 2nd that ChatGPT took off. He’s the 1 that’s misled investors astir the attraction of Microsoft’s AI revenue. If there’s an opportunity for him to lump each of the blasted connected Altman, he’ll return it, arsenic will Jensen Huang, Andy Jassy, and Sundar Pichai, moreover if he’s comparatively quiet astir OpenAI’s billions successful contributions to Google Cloud. 

At 70% of Microsoft’s AI revenues mostly from its tens of billions of dollars’ worthy of compute spend, OpenAI will correspond a worldly driblet successful hyperscaler revenues, and personification will person to beryllium blamed. It won’t matter that Anthropic is conscionable arsenic unprofitable aliases made hundreds of billions of dollars’ worthy of promises it besides can’t keep. OpenAI will make a fitting punching bag, a well-deserved one.

Anthropic and OpenAI Merge

I know, I know. Sam and Dario won’t moreover clasp hands astatine an event. They dislike each other. They some are vacuous psuedo-intellectuals hopeless for attention. 

Yet successful a infinitesimal of desperation, OpenAI could move to Anthropic for a lifeline — a prime merger that would pump both of their bags, each while allowing Altman and his cronies to flight blame. The agreed entity would apt beryllium worthy complete $2 trillion, if only because of its mixed customer guidelines and theoretical “reach,” moreover if reasoning astir that for moreover a second makes it sound truthful unfathomably stupid, arsenic said “reach” would travel pinch multiplicative financial issues stemming from OpenAI’s lousy economics meshing pinch the equally-crap numbers underlying Anthropic.

That being said, in a hopeless moment, this unity could besides warrant further finance from hyperscalers, task capitalists and plus managers, giving them each thing to constituent money astatine and opportunity “this is the early of computing.”

I deliberation it’s very improbable this happens, and if it does, it would beryllium ruinous for everybody involved. Neither of these companies make immoderate benignant of economical consciousness to anyone extracurricular of the recently-concussed and AI boosters pinch dichromatic vision. Combining them would only create a overmuch larger, uglier problem — 1 that would transportation pinch it the very aforesaid problems that some companies have, compounded by the anticipation that it would go the literal savior of the full tech industry.

The pursuing is an nonsubjective database of what OpenAI has to do by 2030:

• Reach $284 cardinal successful yearly revenue.
• Pay $800 cardinal aliases much successful compute obligations.
  • In doing so, OpenAI must go 1 of the largest customers of Amazon Web Services, Microsoft Azure and Google Cloud, all astatine the aforesaid time, and proceed to turn its spend.
• Become profitable. 
  • OpenAI lost $20.9 cardinal successful 2025.
    • If you are going to look astatine this and opportunity “actually it didn’t” because of its Enrontastic accounting treatment, I besides request to pass you — that identical feline successful the bath is really a point called a “mirror,” a reflective aboveground that is showing you a reflection of you, not different personification who is dressed for illustration you and copies everything you do. I can’t ideate really frightened you’ve been, and dream this has helped. 
  • As of Q1 2026, it has a non-GAAP operating separator of antagonistic 122%.

OpenAI is presently “approaching” $40 cardinal successful annualized revenue, astatine precisely the clip it needs to beryllium accelerating. This institution needs to time off 2026 astatine location successful the region of $75 cardinal successful annualized gross to person moreover a snowball’s chance of paying its ridiculous compute costs, and moreover past I’m not judge really it imaginable keeps up pinch the (at least) $146 cardinal successful compute bills it’s sewage coming up. 

It’s clip for everybody to commencement having a real, meaningful speech astir what happens if OpenAI dies. This institution has remained economically unstable since I started penning astir it successful November 2023, and while I mightiness person underestimated its staying power, thing has changed astir my larger thesis that this institution is headed for perdition, leaving its counterparties unpaid and unsocial pinch the consequences to follow.

Said consequences, as I outlined successful the OpenAI Bubble, are very, very serious, representing an existential threat to SoftBank, 1 of the largest companies connected the Japanese banal market, and its illness will guarantee monolithic changes to the guidance of immoderate of the largest companies successful the world. There is simply a very existent script successful which nobody near pinch OpenAI banal is capable to scope a liquidity event, which intends the tens of billions of dollars of task superior will stay unlocked and zeroed retired unless it tin spell public, which is increasingly-unlikely. 

It is nary longer logical aliases reasonable to debar discussing what happens if OpenAI dies. It’s a business that should beryllium connected the mind of each journalist, expert and investor, moreover if they don’t deliberation it’s certain, because OpenAI is some horrendously unprofitable and has made commitments truthful important that they now correspond astatine slightest 20% of hyperscaler unreality revenues successful the coming years, if not much for illustration 30% to 40%. 

It is actively irresponsible to disregard this business immoderate longer, and I promote my peers, analysts, journalists, economists and investors to commencement earnestly considering the likelihood and ramifications of the decease of OpenAI. 

For maine to beryllium wrong, successful the abstraction of 3 years OpenAI will person to go a institution pinch yearly revenues higher than Meta ($200 billion, versus projections of $284 cardinal successful gross successful 2030) and meet obligations ($800 billion+) 27% larger than the mixed revenues of NVIDIA ($215.9 billion), TSMC ($122 billion) and Samsung ($270 billion). 

OpenAI doesn’t person to beryllium forbidden to beryllium dangerous. Every clip consent is manufactured for the astonishing discarded and unrealistic promises of Sam Altman, companies further leverage themselves successful an effort to seizure its theoretical value, and investors are further manipulated into supporting an manufacture almost-entirely founded connected its compute spend. 

As I discussed successful the OpenAI Bubble, its illness will person now-unavoidable economical consequences. The decease of SoftBank is simply a very existent possibility. The likelihood of the immense mostly of AI investments going to zero is much, overmuch higher than anyone wants to deliberation about, astatine a clip when, per Bloomberg, a task superior patient that returns thirty centers connected the dollar is considered an above-top-five performer. Oracle will illness without OpenAI’s revenue. 

To not actively and meaningfully talk the potential for OpenAI to illness is actively irresponsible. To enactment for illustration location are not significant, existential problems pinch this company’s economics is to intentionally debar reality, and whoever is connected the receiving extremity of said ignorance deserves better, beryllium they an investor reference your expert statement aliases a scholar burdened pinch incomplete journalism.

What follows whitethorn beryllium an Enron-Lehman Brothers hybrid, 1 that leaves unbelievable demolition successful its wake, an avoidable systemic consequence empowered and enabled by a kneecapped media manufacture and sell-side analysts incapable of seeing further than 2 quarters successful the future. 

In the end, location is nary avoiding the harm that OpenAI’s illness will create. The clip to do that was successful 2024, earlier it made each those commitments, and raised truthful overmuch much money. Once it did so, it led the full manufacture to judge that location was important request for AI, erstwhile each that was happening was Sam Altman and Dario Amodei were taking up each ounce of compute capacity, paid for pinch equity investments from the companies they bought it from, an illusion created by men driven huffy by their desperation for hypergrowth. 

However you consciousness astir my work, I americium begging you to return moreover the imaginable of OpenAI’s illness seriously, and hole accordingly. 

---

If you liked this piece, you should subscribe to my premium newsletter. It’s $70 a year, $17 a quarter, aliases $7 a month, and successful return you get a play newsletter that’s usually anyplace from 10,000 to 18,000 words and provides vast, elaborate analyses of the biggest events and companies successful the AI bubble.

If you want to get successful touch — and particularly if you person immoderate juicy accusation astir Anthropic, OpenAI, aliases immoderate different companies successful the AI bubble — deed maine up connected Signal astatine ezitron.76. I’m besides connected IB connected The Terminal.