In mid-Sep 2024, OpenAI CEO Sam Altman announced at a company-wide meeting that OpenAI’s nonprofit structure was going to change, likely in 2025. Altman expressed his belief that OpenAI had outgrown its nonprofit-controlled structure. It is a shift that has been rumored to be brewing over the past few months.

After the conversion, the nonprofit Delaware entity would continue to exist and retain a minority stake in the new for-profit benefit corporation. (OpenAI rivals Anthropic and xAI are also structured as benefit corporations.) According to OpenAI, it will continue to focus on “building AI that benefits everyone.” Because a nonprofit is an independent entity with a societal obligation under the law – i.e. not owned by its founders or donors – its assets can’t be readily extracted. The full value of the assets needs to stay with the nonprofit, which makes for a complex transaction in which the assets would probably need to be sold to the for-profit.

This week, OpenAI closed a $6.6B funding round at a $157B post-money valuation (up from $86B earlier this year). The funding was in the form of convertible debt. Investors reportedly have the right to withdraw their equity investment (converting into debt at 9% interest) or renegotiate the valuation if OpenAI doesn’t convert to a for-profit within 2 years and remove the caps on returns.

The recent funding round was backed by Thrive Capital ($1.3B, with an option for another $1B later), Microsoft ($750M-$1B), SoftBank ($500M), Tiger Global ($350M), Altimeter Capital ($250M), ARK Venture Fund ($250M), Coatue Management ($250M), Nvidia ($100M), Khosla Ventures ($500M), Fidelity, new UAE AI/chip vehicle MGX, Quiet Capital, and CalPERS.

The round came alongside a new $4B credit facility (with an option for another $2B) from JPMorgan Chase, Citi, Goldman Sachs, Morgan Stanley, Santander, Wells Fargo, SMBC, UBS, and HSBC, among others, for a total war-chest of $10.6B. The credit line can be tapped over 3 years, at an interest rate of SOFR (Secured Overnight Financing Rate) + 1%.

Notably, OpenAI has asked investors not to invest in its rivals, including Anthropic (which recently hired OpenAI co-founder John Schulman and is backed by Amazon and Google), Perplexity, Elon Musk’s xAI, ex-OpenAI chief scientist Ilya Sutskever’s Safe Superintelligence, and Glean Technologies. The ask for exclusivity is reportedly nonbinding – ARK Venture Fund, for instance, is already invested in xAI and Anthropic, and Fidelity is invested in xAI.

As a reminder of the history of OpenAI: OpenAI Inc – also known as OpenAI Nonprofit – was founded in Dec 2015 as a nonprofit research institution by Elon Musk and Sam Altman. Its mission was to ensure that “artificial general intelligence” (AGI) benefits all of humanity. The nonprofit was backed by $1B+ in commitments from Musk (who ended up putting in about $45M), Altman, Greg Brockman, Reid Hoffman, Jessica Livingston, Peter Thiel, Amazon (AWS), Infosys, and YC Research. A later TechCrunch analysis added up $133M in actual donations to OpenAI’s nonprofit from 2015-2021, which roughly matches what OpenAI has said itself. Elon Musk left the board in Feb 2018 – reportedly on reasonably good terms then – due to disagreements on direction and to avoid conflicts with Tesla.

In Mar 2019, just after the GPT-2 announcement (which drew attention for its versatility and staged release), OpenAI Inc launched a “capped profit” subsidiary called OpenAI LP to attract the billions needed to fund its compute-driven research (and pay $1M+ salaries to AI talent). OpenAI LP – known as OpenAI – was controlled by the nonprofit board and obliged to advance the aims of the nonprofit’s charter. (The structure has become even more convoluted since, with separated economic and control rights.) Investment in the for-profit arm could be quite lucrative – returns were “capped” at 100x investment for the first investors (with the multiple to be reduced in later rounds). Altman says he has not taken equity or profit participation in the for-profit arm.

In Jul 2019, Microsoft invested $1B in OpenAI in cash and Azure credits. (OpenAI’s large-scale experiments had been running on Azure since 2016.) Microsoft became OpenAI’s “preferred commercialization partner” and Azure its “exclusive cloud provider.” While somewhat controversial, a close partnership with a major cloud player was necessary for OpenAI. With compute requirements doubling every few months, it needed access to scale compute and funding. The relationship also meant that OpenAI could commercialize its technology without being distracted from its research.

Under Microsoft’s capped-profit deal structure, after OpenAI’s first set of investors are paid back on a 100x capped-return basis ($150B), Microsoft gets 75% of OpenAI’s profits until its investment is paid back ($13B) and then 49% of profits up to a hard cap ($92B). Microsoft owns 49% of the equity, and other investors/employees own the other 49%. OpenAI initially retained 2% of the equity, with the remaining equity reverting back to OpenAI once investors got their return. The deal was reportedly later changed to increase the cap by 20% per year starting in 2025. By Jan 2023, Microsoft had put in a total of $13B into OpenAI (in large part, by footing its cloud bill), which now becomes nearly $14B with the latest funding round.

The OpenAI team has been pushing out products at a rapid pace. In Jun 2020, one month after revealing GPT-3, OpenAI released its first commercial product in private beta – the general-purpose OpenAI API based on GPT-3. It then released the text-to-image DALL-E in Jan 2021, the text-to-code Codex in Aug 2021, DALL-E 2 in Apr 2022, and the DALL-E API in Nov 2022.

An inflection point happened in Dec 2022 when OpenAI released the more accessible ChatGPT, which went viral. It was followed by the ChatGPT Plus paid subscription in Feb 2023, GPT-4 and ChatGPT plugins in Mar 2023, fine-tuning via API in Aug 2023, multimodal features in Sep 2023, DALL-E 3 in Oct 2023, the text-to-video Sora in Feb 2024, the more natural GPT-4o in May 2024, the SearchGPT prototype in Jul 2024, the o1 reasoning models just last month, a speech engine for 3rd-party apps and Canvas collaborative interface just this week, and AI agents are also on the way.

OpenAI’s revenue has grown rapidly over the past 2 years. It went from $28M in revenue in 2022 to $700M in 2023 to a projected $3.7B in revenue for 2024. (As of Aug 2024, its monthly revenue was up 17x since the start of 2023, and its annualized run rate recently hit $4B.) OpenAI is projecting $11.6B in revenue in 2025 and $100B by 2029.

More than 70% of OpenAI’s $3.7B in revenue comes from ChatGPT ($2.7B) and the rest ($1B) comes from enterprise services. ChatGPT now has 250M+ weekly users (350M+ monthly active users), 11M paid ChatGPT Plus users, and 1M+ higher-paying business users. Part of the expected revenue growth will come from planned price increases – ratcheting up from $20/month to $22/month by the end of 2024, and to $44/month within 5 years. OpenAI is still not profitable – this year it will lose a hefty $5B.

It’s been a chaotic year for OpenAI – not even considering Altman’s brief ouster and return in Nov 2023. Last Wednesday, 3 OpenAI executives announced their departures – CTO Mira Murati (who served as interim CEO during Altman’s ouster), research chief Bob McGrew, and research VP Barret Zoph. This year has also seen the departures of OpenAI co-founders Ilya Sutskever, John Schulman and Andrej Karpathy; Superalignment/safety leader Jan Leike (and nearly half of the safety researchers); and product VP Peter Deng. Of the original 11 co-founders, only Altman, OpenAI president Greg Brockman, and Wojciech Zaremba remain, and Brockman – who has a reportedly demanding leadership style – is on an extended sabbatical through the end of the year.

While OpenAI has lost more than 20 researchers and execs over the past year, it has also more than doubled the size of its team during that period to near 1,800 employees. Insiders describe a shift from a research culture to a very different “product-first” culture. Some are pointing to the pressure associated with releasing products resulting in premature releases and less rigorous safety testing. OpenAI’s ex-Superalignment lead Jan Leike, after he left, called out the challenges he faced in getting enough compute to do safety research. In his words, “[O]ver the past years, safety culture and processes have taken a backseat to shiny products.”

At the same time, OpenAI and its leaders are facing a lawsuit from Elon Musk for allegedly breaching the nonprofit’s original mission. Musk has been particularly infuriated by reports that Altman might get a 7% equity stake, which would equate to $11B in compensation. OpenAI has said there have been discussions on compensating Altman with equity but denied that specific figures have been mentioned. Altman himself has said it’s “ludicrous” and “just not true” that there are current plans to give him a “giant equity stake.”

OpenAI is on the road to becoming a big tech firm. It continues to need significant capital given its continued sizable losses, and Altman reportedly wants to keep Microsoft’s stake to less than 50%. OpenAI needs to be a for-profit in order to access the deep liquidity available in public markets. While Altman has said in the past that he doesn’t want OpenAI to go public, an investor in the recent round described an IPO as “the logical next step,” remarking that a valuation of 10x projected revenue “isn’t exorbitant for a company about to go public.”

Altman is thinking on a large scale, with an ambitious multi-year vision that could involve as much as $5T-$7T in funding from sovereign wealth funds, credit investors, and governments, as well as support from energy giants and industry players like TSMC to run the factories. OpenAI itself is working on projects on the scale of “hundreds of billions.”

In Feb 2023, we said in a brief: “Standing on this precipice, we should perhaps consider ourselves lucky that OpenAI has nonprofit governance and a capped-profit structure – given that AI will be lucrative and can be used as a force for evil as well as for good.” Now even these apparently flimsy guardrails are going away. Given the pace at which the world is moving down this path, it will be very hard to predict what will happen in the decade ahead – the only logical response is “situational awareness.”

Sep 20 2024 (3 Shifts): OpenAI's o1 models can reason and solve complex problems

May 17 2024 (3 Shifts): OpenAI’s drive towards more natural interactions and scale