OpenAI's Survival Dilemma: The Triple Threat of Capital, Technology, and Business Model—The Most Prominent May Not Be the One That Lasts Longest

Original Author: Ed Zitron

Original Translation: Deep Wave TechFlow

Introduction: OpenAI is trading its core technology to Microsoft for cloud computing resources while burning through $5 billion a year. Can this star company, valued at over $100 billion, really survive the next two years? This article calmly dissects OpenAI's existential dilemmas — under the triple pressure of funding, technology, and business model, which one could truly kill it?

How will OpenAI survive?

Over the past year, I've written in detail about the tech bubble — how grifters raise money and consolidate power, how big tech executives desperately maintain growth myths, and how hollow the latest hype cycle is — but ultimately, these are all just companies, which raises a simple question: Can the biggest and most visible company in this hype cycle actually survive?

Of course, I'm talking about OpenAI. Regular readers know I'm deeply skeptical of OpenAI's products, business model, and sustainability. Without repeating the arguments from my previous articles and podcasts, let me just state the core contradiction: Generative AI simply doesn't have mass-market necessity — at least not at the scale of truly revolutionary technologies like cloud computing and smartphones — yet it costs a fortune to build and run.

These two points cast serious doubt on OpenAI's ability to survive in the medium to long term, especially when — I'll be bold and say, should I say when — the faucet of investment and cloud computing resources gets turned off.

I don't have all the answers. I don't know every detail of every deal. I'm neither an engineer nor an economist, and I don't have insider information. But I can read public data, evaluate independent reporting from seasoned journalists and opinions from experts and scholars, and then draw conclusions.

My assumption is: For OpenAI to survive another two years, it must (in no particular order):

Fix its complicated and burdensome relationship with Microsoft, which is both a lifeline and a direct competitor

Raise an astronomical amount of capital that no startup in history has ever raised, and do so at a pace unprecedented in funding history

Achieve a major technological breakthrough that reduces the cost of building and running GPT — or its successor — by thousands of percentage points

Achieve a major technological breakthrough that enables GPT to handle entirely new use cases that no AI researcher has even thought of yet

These new use cases must both create new jobs and fully automate existing ones, thereby justifying the massive capital expenditures and infrastructure investments

I ultimately believe that OpenAI in its current form cannot survive. There's no path to profitability, it's burning cash too quickly, the energy grid that generative AI technology requires simply can't handle it, training these models is equally unsustainable, and it faces both legal issues (due to pirated training data) and the need for massive amounts of training data to develop further.

Simply put, any technology that requires hundreds of billions of dollars to prove itself has an inherently flawed architecture. What OpenAI needs has no historical precedent. No one has ever raised the amount of money it needs, and no technology has ever required this much financial and systemic force — like rebuilding the U.S. power grid — just to survive, let alone prove it's worth the investment.

To be clear, this article focuses on OpenAI rather than generative AI technology itself — although I believe OpenAI's continued existence is a necessary condition for keeping companies interested/invested in this industry. OpenAI has raised the most funding ($11.3 billion), arguably has the highest media attention, and has both popularized and created the large language model business model that allows the current crop of "AI-driven" startups to exist.

I'm not saying OpenAI will definitely go bankrupt, or that generative AI will definitely fail. I've discussed the industry's problems in detail before, and I won't repeat those arguments here — just to note that I believe there's deep instability in the tech ecosystem. What I'm doing here is calmly explaining why OpenAI, in its current form, cannot survive more than a few years without a series of astonishing technological breakthroughs and financial maneuvers — some of which might be achievable, but most of which are unprecedented.

Let's take a look, shall we?

The Microsoft (and Valuation) Problem

Before I continue, I should note that OpenAI and Microsoft have not publicly disclosed the terms of their deal, and there may be things I don't know that would change this assessment.

But regardless, the OpenAI-Microsoft relationship is very strange. It began in 2019 with a $1 billion investment, described as a "multi-year exclusive computing partnership" to "migrate OpenAI's existing services to Azure" (Microsoft's cloud computing product), making Microsoft the "preferred partner for OpenAI." OpenAI CTO Greg Brockman added at the time that this was a "cash investment," but mentioned that OpenAI "plans to be a major Azure customer."

One thing particularly worth noting: Brockman said Microsoft would receive rights to sell "pre-AGI products" based on OpenAI research to its own customers, and the accompanying blog post also said Microsoft and OpenAI were "jointly developing new Azure AI supercomputing technology."

"Pre-AGI" here refers to anything OpenAI has developed, since it hasn't developed AGI yet and hasn't even gotten past the "chatbot" level on its 5-level system for evaluating artificial intelligence.

Essentially, the terms of this funding round allowed OpenAI to hand over its research results to Microsoft, permitting Microsoft to sell OpenAI technology as its own under the Azure brand. Moreover, Microsoft can see OpenAI's "pre-AGI product research," meaning it can see how it works, allowing Microsoft to both sell the technology and directly compete with it. Microsoft is already doing this — The Information reported in May that Microsoft was preparing its own "MAI-1" generative model, led by DeepMind co-founder Mustafa Suleyman (Google acquired DeepMind in 2014), who left in 2022 to do VC. Suleyman later founded Inflection, a transformer-based chatbot company, which was half-acquired by Microsoft in March after Microsoft had previously invested in it.

The Information also mentioned that OpenAI was cooperating to "help Microsoft get ahead of rivals." I'm still speculating, but just the terms of this deal are extremely unfavorable to OpenAI.

Now it gets a bit confusing.

Apparently, Microsoft made some kind of investment in OpenAI in 2021 (mentioned in a 2023 blog post about another funding round, which I'll get to in a moment). Strangely, I can't find specific details about this round. Crunchbase mentions a secondary market transaction in 2021 with an undisclosed amount (meaning existing OpenAI shareholders could sell stock to cash out), but says the money came from Tiger Global Management, Sequoia Capital, Bedrock, and a16z, at a $14 billion valuation.

It's unclear whether this is the same round, or whether Microsoft invested through other means. The Information mentioned it in an early 2023 article, but somehow nobody reported it at the time. If I'm wrong, I'd love to see how it actually invested.

Regardless, in early 2023, Microsoft invested $10 billion in OpenAI — but most importantly, the deal terms and how it was structured. Although the terms weren't disclosed, reports say Microsoft could take up to 75% of OpenAI's profits until it "recoups its investment" (in other words, earns back that $10 billion), plus a 49% stake in the company, even though OpenAI's weird nonprofit-for-profit hybrid structure itself is unusual.

Semafor reported that, at least as of November 2023, OpenAI had only received "a small portion" of the $10 billion investment, which was being distributed in tranches, and that a "considerable portion" was in cloud computing credits, meaning Microsoft's investment was primarily in the supposed value of a currency that can only be used on its own services. For those who don't understand how strange this is, it's like an airline investing in a company but not giving cash — giving miles instead. You can still fly, but you're locked into A) one airline and B) their interpretation of what a "mile" is worth.

Moreover, saying this "investment" might — again, we don't know the deal terms — have allowed OpenAI to inflate its valuation in the process is extremely weird.

Semafor also (vaguely) added that Microsoft has "certain rights" to OpenAI's intellectual property, and even if the relationship broke down, it could "still run OpenAI's current models on [its own] servers."

Strangely, several journalists have said Microsoft invested "$13 billion" in OpenAI, but I can't find that extra $2 billion anywhere. Was it invested in 2021? Was the amount $2 billion? Was it cash or credits? This number has been repeated for so long that it's odd so many people just accepted it.

I think this matters because, by any startup standard, $2 billion is a lot of money.

For example, Snowflake, a very successful enterprise computing company, raised a total of $2 billion, mostly before its IPO (although it sold another $621.5 million in stock after its 2022 IPO). And Snowflake lost $316 million last quarter.

Furthermore, this deal was completely unannounced and unreported, happening in the same year (2021) that Microsoft and OpenAI announced the Azure OpenAI service, which didn't publicly launch until 2023. Although I'm speculating — if I'm wrong, please email me — it looks like Microsoft gave OpenAI another $2 billion in 2021, while OpenAI raised an undisclosed amount from other sources. I'm guessing this deal might also have been a mix of cash and cloud credits, even though ChatGPT wasn't public until November 2022.

It's also possible that those reporting Microsoft's "$13 billion" investment are simply wrong, but the error is so widespread that Microsoft or OpenAI would likely have quickly denied it.

The reason I raise all these questions is that Microsoft effectively already owns OpenAI at this point. Microsoft CEO Satya Nadella was instrumental in Altman's return after his firing in November 2023, and Nadella had already planned to hire him if he didn't go back. In many ways, this doesn't matter much. Due to the nature of Microsoft's deal with OpenAI, it effectively owns — or at least has access to — all the intellectual property behind OpenAI's products, plus all the research, and the ability to license it at will.

OpenAI is tied to Microsoft. It's locked into Azure, Microsoft's cloud platform, through both its agreements and the fact that most of its funding is in credits that can only be used on Microsoft Azure. Microsoft sells GPT access through Azure, directly competes with it using its own upcoming models, and takes three-quarters of the (theoretical) profits from OpenAI's services.

Microsoft hasn't really sacrificed anything in these deals either. Even if we assume the earlier funding rounds — theoretically $3 billion — were all cash, and the 2023 round was 25% cash and 75% credits, that's only $5.5 billion, with the latter half paid in installments on an uncertain schedule. Assuming (again, I don't have the exact terms) it really is $5.5 billion in cash, that's nothing to Microsoft, a company that earned over $21 billion in profit in its most recent fiscal quarter.

As part of this deal, Microsoft effectively bought the rights to OpenAI's "pre-AGI" technology, licensing all its technology in a way that goes beyond any partnership or future deal I can imagine. Microsoft also "invested" cloud credits at an uncertain valuation, covering both how OpenAI is valued and the credits themselves.

Ask yourself, what is a dollar of "cloud computing credit," and what can you use it for? Microsoft Azure cloud has many, many products, and it's unclear whether OpenAI gets a discount, what products it would use, and on what terms. Microsoft effectively created its own currency to invest in OpenAI, then OpenAI uses that currency to pay Microsoft, which in turn books it as revenue.

In many ways, OpenAI's continued existence serves as the R&D arm of Microsoft's generative AI business, with the dice loaded in Microsoft's favor. If OpenAI collapses, OpenAI's technology will still run on Microsoft servers, and Microsoft will still have access to OpenAI's intellectual property and products, and can sell them. If OpenAI thrives, and future generations of GPT are extremely profitable and successful, Microsoft harvests billions in profits while retaining access and licensing rights to any research or products used to achieve that. Even Microsoft's $100 billion supercomputer project, according to The Information's sources, is tied to whether Altman and OpenAI can "significantly improve" AI capabilities.

I'm clearly not sure and can't verify this, but don't you think the "pre-AGI" technology and research includes SearchGPT, OpenAI's recently announced competitor to Google Search? Isn't it worth considering for a second that Microsoft benefits whether OpenAI lives or dies?

This is a deal with the devil — the kind you only sign when you're burning cash so fast that you must find a wealthy backer who can repeatedly save you while you consume billions of dollars a year.

Sadly, this might be the truth about OpenAI.

The Funding Problem

Last week, The Information reported that OpenAI could burn through $5 billion in 2024, based on "previously undisclosed internal financial data and industry insiders."

That article made several well-founded estimates (I'd recommend subscribing to The Information if only for that article), and I'll build my analysis on top of it. While these estimates could be wrong, or the underlying data could be misleading or erroneous, I trust the rigor of The Information's analysis and reporting:

"As of March 2024, OpenAI expects to spend nearly $4 billion this year renting Microsoft's servers to power ChatGPT and its underlying large language models," according to "people with direct knowledge of the spending."

"OpenAI's training costs — including data payments — could swell to $3 billion this year." To be clear, training costs aren't just about acquiring data; they also include cleaning and preparing the data — a laborious task — and then using massive cloud computing resources to train the models.

The Information "guesses" that OpenAI's team of about 1,500 people (and growing) could cost around $1.5 billion annually. While this sounds a bit high — especially considering it works out to $1 million per person — it's actually quite reasonable. Top AI talent is extremely expensive, and seven-figure salaries are far from rare. Then you have to factor in office space, payroll taxes, equipment, and other operating costs.

The Information also estimates OpenAI's revenue at between $3.5 billion and $4.5 billion, including ChatGPT and fees charged to developers for using OpenAI's API to integrate generative features.

The Information speculates that OpenAI therefore has an annual operating loss of $5 billion. This also assumes OpenAI's revenue is at the higher end, and it could potentially inflate to $6 billion or more.

While we don't have direct knowledge, OpenAI's operating costs have continued to grow rapidly over time. An estimate from early 2023 showed that running ChatGPT cost $700,000 per day at the time, when ChatGPT was popular but not as much as it is today. That would put ChatGPT's cost alone at approximately $235.3 million per year. I'd also speculate they're much larger because OpenAI has raised over $13 billion in the past five years, with most of that capital (and credit) raised between 2021 and 2023.

According to reports, OpenAI has historically failed to make its models more efficient, failing to deliver the more efficient "Arrakis" model to Microsoft by the end of 2023. While the recently released GPT-4o mini model has been touted as an "efficiency" move, it seems to only be more efficient and cost-effective for people developing with OpenAI tools. While some might argue this shows OpenAI has found more efficient/cost-effective models and therefore cut prices, it hasn't proactively provided this information to confirm it.

Assuming everything exists in a vacuum, OpenAI needs at least $5 billion in new capital annually just to survive. That would require it to raise more money than any startup in history, possibly permanently, which in turn requires it to access capital at a scale for which I can find no comparable in business history.

WeWork — Silicon Valley's fallen prodigal son — raised a total of $22.2 billion, nearly half of which (over $10 billion) was debt financing (loans with varying terms) raised