How OpenAI, the Leading Startup of Our Time, is Structured: A Look into the Financial History of ChatGPT

In the past few months, people have been talking a lot about the technological innovation, ChatGPT, which is an AI-based chatbot. ChatGPT can engage in dialogue with users and answer complex questions on a variety of topics including economics and politics. However, ChatGPT’s capabilities go beyond just that – it has passed an MBA exam and even written a thesis for a student.

Let me tell you about the financial side of the famous chatbot and about OpenAI, the organization that created it:

  • Who owns ChatGPT: investors or no one?
  • How much will Microsoft earn from OpenAI and what are the profit restrictions?
  • How much is OpenAI worth and what connects it to offshore agreements of Chinese VIE companies?
  • What does OpenAI earn from and what can it earn from?

We will answer these questions, among other things, with the help of ChatGPT.

How OpenAI is structured

Sam Altman founded OpenAI as a nonprofit organization in 2015, with financial support from Elon Musk, the owner of Tesla and SpaceX, and Reid Hoffman, the founder of LinkedIn. They invested billions of dollars in the project.

As the former president of Y Combinator, a well-known startup incubator, Sam Altman is a significant figure in the tech world. Under his leadership, the incubator produced and nurtured successful companies such as Stripe, a fintech service, Instacart, a grocery delivery service, Coinbase, a cryptocurrency exchange, and many others.

In 2019, Microsoft donated approximately one billion dollars to OpenAI, mostly in the form of Azure credits, which granted the right to use Microsoft’s cloud service for free. OpenAI also utilized Microsoft’s cloud software to train its chatbot.

The OpenAI project published its charter in 2018, stating that its primary duty is to humanity and that its focus is to provide resources to the general public to help society create full artificial intelligence. The organization was conceived as a non-profit.

However, the financial requirements of OpenAI’s projects exceeded what could be obtained through donations alone. The project needed not only donors but also investors.

OpenAI required more funds to continue developing its operations. According to Altman, processing chatbot responses on average cost less than 10 US cents per chat, likely referring to the cost of processing an entire conversation. Unfortunately, Altman did not provide specific details about the exact cost in cents.

In 2022, OpenAI incurred losses of $500 million, despite the revenue it generated. The exact amount of losses incurred by OpenAI each year is unknown.

As a private non-public company, OpenAI is not obligated to disclose its financial information to the public. ChatGPT often reminds those who inquire about the company’s finances of this fact. While Altman disclosed the costs of processing chatbot responses on Twitter, he was not required to do so.

To raise funds from investors, OpenAI re-registered its donation fund as a special commercial unit called OpenAI LP in 2019. However, only a small portion of OpenAI is owned by the organization, with the majority belonging to investors. As of March 2023, there are two separate entities operating under the name OpenAI.

The “main” OpenAI

The primary OpenAI entity is a non-profit organization dedicated to AI research. It is an NGO and has a board of nine directors, including key figures of OpenAI and early founding investors such as Altman and Hoffman. There are no owners of the organization, as it is structured as a non-profit.

The Board of Directors of the non-profit organization OpenAI consists of:

  • Greg Brockman, Chairman and President
  • Ilya Sutskever, Chief Scientist
  • Sam Altman, CEO
  • Adam D’Angelo, Internet entrepreneur known for founding the Q&A platform Quora
  • Reid Hoffman, entrepreneur and co-founder of LinkedIn
  • Will Hurd, American politician and former CIA officer
  • Tasha McCauley, American scientist and entrepreneur, wife of actor Joseph Gordon-Levitt
  • Helen Toner, American scientist
  • Shivon Zilis, Canadian entrepreneur, manager at Neuralink of Elon Musk and mother of two of his children.

The composition of the OpenAI Board of Directors may change.

OpenAI LP

OpenAI LP, which is a commercial limited partnership, is responsible for conducting and formalizing the commercial operations of the OpenAI non-profit organization, such as selling goods and services, and receiving funds from investors. The shares in OpenAI LP are owned by both investors and the non-profit OpenAI organization.

However, the non-profit organization that focuses on AI projects is not owned by investors. OpenAI LP plays a purely technical role by collecting all the funds earned by OpenAI, but it is the main non-profit OpenAI organization that decides how to allocate the funds for research and development in the field of AI, which is the main goal of the non-profit.

Members of the OpenAI LP board of directors are appointed by the non-profit OpenAI organization. Additionally, the non-profit organization has veto power over decisions made by the OpenAI LP leadership that relate to the values of the non-profit organization, licensing of technology, and the security of its transfer.

OpenAI LP’s broad wording grants it full control. For instance, if OpenAI LP decides to offer commercial services in Russia, OpenAI nonprofit organization may use security concerns or OpenAI’s values as pretexts to ban such work.

This structure exposes investors in OpenAI LP to considerable risks as the company owns nothing. Currently, Microsoft is the primary source of funding for OpenAI, but other companies may offer more money. It is unclear if investors have hedged against this risk. If OpenAI plans an IPO, shareholders will lose everything if the company maintains its current structure.

A stake in OpenAI LP is like owning shares in Chinese VIE companies, which exist solely for stock market listing and have no assets. Even ChatGPT agrees with this comparison.

Let’s talk to ChatGPT about VIE analogies

Who has invested in OpenAI?

Investments in OpenAI, both already made and announced, amount to a total of $12 billion, with the lion’s share of $11 billion coming from Microsoft.

Here are other investors in OpenAI:

  • Venture capital funds Khosla Ventures and Founders Fund.
  • LinkedIn co-founder Jeff Weiner.
  • Singapore sovereign wealth fund.
  • Amazon’s cloud division – AWS.
  • Technology consulting company Infosys and many others.

OpenAI has not disclosed how much private individuals and organizations have invested, including the full list of investors.

Microsoft announced its $10 billion investment in January 2023, which should make it the largest investor in OpenAI and the recipient of the bulk of the money.

Under the terms of the deal, Microsoft invested $10 billion and acquired a 49% ownership stake in OpenAI LP. Microsoft will receive 75% of OpenAI LP’s profits until it recoups its investment.

OpenAI is not required to disclose how much Microsoft owned prior to this deal. After reaching the $10 billion threshold, OpenAI LP will receive 100% of the profits. Other investors, such as Andreessen Horowitz and Kosla Ventures, own the remaining 49% of OpenAI LP, while the main OpenAI nonprofit organization holds an additional 2%.

Once OpenAI LP reaches profitability and resolves the “Microsoft $10 billion” issue, profits will be distributed proportionally among all owners. The OpenAI nonprofit organization will oversee the distribution of profits, and Microsoft and other investors do not have a stake in it.

OpenAI has set a cap of 20× on the potential return on investment for current investors in the investment contract. Investors will receive a maximum return of 20× on their investment, with any excess going toward funding the OpenAI nonprofit organization. This means that not only employees and investors but “humanity as a whole” will benefit from the company’s progress.

For those who invested in OpenAI in 2019, the maximum return is capped at 100×. This means that the maximum return Microsoft can expect from its investment is $300 billion: $100 billion for the 2019 investment and $200 billion for the 2023 investment.

How OpenAI earns money

During an investor meeting, Altman stated that OpenAI will generate billions of dollars in revenue annually by charging users and businesses for its services.

Unofficially confirmed data suggests that the company currently earns tens of millions of dollars per year by selling access to its software (PO) to third-party developers.

Based on social media activity, the most productive application of ChatGPT appears to be in marketing and search engines. Microsoft has already integrated OpenAI solutions into its Bing search engine and cloud service.

Currently, the most popular uses for ChatGPT are writing emails and code snippets, generating gift ideas, and summarizing meetings.

ChatGPT provided a conservative answer to my question about the commercial application of its technology, stating that the service sees itself primarily in the field of chatbots and content creation.

This is understandable, as the service is officially limited to a database that only includes data up to September 2021. As a result, it is not up-to-date on recent developments in the world.

However, some users have managed to obtain up-to-date information from the service using various tricks and errors in the bot. Unfortunately, I am not experienced enough to deceive AI and bypass the “September 2021” restrictions.

OpenAI recently released a new commercial solution, ChatGPT Plus, which is an upgraded version of its chatbot. For a fee of $20 per month, users will have 24/7 access to the service, which is a significant improvement as during peak hours, the service becomes inaccessible to many users due to high demand.

Paid subscribers will also enjoy faster responses and “priority access to new features and fixes.” While faster responses are self-explanatory, it is unclear how the “priority” feature will operate. Based on the wording, it appears that ChatGPT Plus users will receive new features and bug fixes ahead of those using the free version, rather than exclusive functions. However, until the paid service becomes available to the general public, we cannot be certain.

Selected OpenAI clients are currently receiving invitations to the service, and those who are interested in joining can sign up for a waiting list. The company may soon send an invitation to join, but they are not disclosing the number of clients at this time.

OpenAI is still in the early stages of commercializing its developments, but they may explore other ways of generating revenue in the future. Recently, they opened paid access to the chatbot’s application programming interface (API), allowing those who pay to integrate it into their own developments. Another source of monetization may appear in the near future, possibly within a few weeks of this article’s publication.

How much is the company worth?

As of 2021, OpenAI’s value was estimated to be approximately $14 billion. However, in early January 2023, reports surfaced indicating that the company’s management was seeking to sell shares to venture capital funds Thrive Capital and Founders for $300 million, which would more than double its valuation to $29 billion.

Based on the information available, it appears that Microsoft’s $10 billion investment in January 2023 did not impact OpenAI’s valuation, which remains around $29 billion.

There are two concerning factors to consider with this valuation.

Expensiveness

Altman’s prediction suggests that if OpenAI were to generate one billion dollars in revenue per year, its P/S ratio would be 29, which is unusually high even for a promising startup.

However, the reality is that the company has not yet reached the billion-dollar revenue mark. In fact, according to Reuters, who received information from “knowledgeable sources,” OpenAI’s revenue for 2022 is estimated to be around $80 million. OpenAI has not confirmed this information.

OpenAI investors hold a token that represents a share in OpenAI LP, which doesn’t have any assets of its own.

In theory, it’s possible for the non-profit organization OpenAI to cut ties with OpenAI LP, leaving investors with nothing. It’s worth noting that ChatGPT also agrees that OpenAI LP shareholders are taking on a risky investment.

Dialogue with ChatGPT about investor risks in OpenAI LP

Business application of ChatGPT

I attempted to use ChatGPT to write an article for me, but unfortunately, I didn’t receive any useful output that I could use in my material. It seems that on complex economic topics, the service is unable to provide meaningful responses. For instance, when I asked why English actors are cheaper than American ones, ChatGPT gave a generic response stating that “it can be different.”

Brian Chen from The New York Times conducted a test with ChatGPT, along with two other chatbots, Wordtune Spices and Rytr, to see if they could write articles. The outcome showed that these virtual assistants were only capable of editing pre-written texts, and were not able to think, search for or analyze information. While it’s possible that this could change with time, it’s not a certainty.

Codeword, a marketing company, “employed” AI interns to perform basic and mundane tasks related to design and text editing. However, there is no information available about the outcome of this initiative.

In theory, programs such as ChatGPT have the potential to be utilized in the financial sector and stock market. They can quickly sort through massive amounts of data, including financial reports and social media data, to make investment or lending decisions based on their analysis. However, this is only a theoretical possibility.

Game developers are also exploring the use of chatbots like ChatGPT to create dialogues for non-playable characters (NPCs) during gameplay, which could lead to unique responses based on user interactions. While these plans seem promising, it remains to be seen how effective this technology will be in real-world conditions. Extensive implementation and empirical data will be required to determine what tasks AI can perform better than humans and what it cannot.

But we can only assume that a more extensive implementation of this technology will reveal its results in the field. With a large volume of empirical data, we will be able to accurately determine what this technology can do better than humans, and what it cannot.

Responses