In 2019, Microsoft took a gamble by investing a billion dollars in OpenAI, a startup that was then little known to the general public but very promising, specializing in artificial intelligence. The software giant did not enter the capital, but offered a privileged partnership with a potentially revolutionary nugget thanks to its natural language processing technologies and its researchers among the best in their fields. OpenAI would not only train its compute-intensive artificial intelligence models on Azure, Microsoft’s infrastructure, but would also give privileged commercial access to its technologies.
Three years later, OpenAI surpassed all of Microsoft’s expectations by releasing three impressive tools in quick succession: Codex, an AI capable of translating natural language into computer code; DALL-E 2, an image generator from a phrase given by the user; and ChatGPT, a chatbot capable of answering a wide variety of questions, following instructions or even writing entire texts. Technologically impressive, the three tools conquered the general public at the end of last year, opening up new perspectives for artificial intelligence.
Delighted with these results, Microsoft is now looking to integrate OpenAI tools into its software. He is therefore preparing to reinvest to strengthen the existing partnership, this time in the order of 10 billion dollars, according to Semaphore. At the end of a complex financial agreement, the technology giant would obtain 49% of the startup’s capital.
Copilot, the first success of the Microsoft OpenAI alliance
With no visible results for almost three years, the partnership between Microsoft and OpenAI brought in 2021 a first product, called Copilot, integrated into the GitHub interface (one of Microsoft’s subsidiaries, acquired in 2018) a year later. This wizard for IT developers suggests lines of code for natural language commands. For example, if the user writes “ find the images and color their edges green “, Copilot will offer code to implement the command, which the user can immediately integrate. GithHub claims that Copilot writes up to 40% of its clients’ code – the percentage varies depending on the programming language used. Even better, the company expects its AI-powered products to generate around 80% of developer code by 2027. challenging player.
Thanks in part to Copilot’s million subscribers, GitHub has passed the $1 billion mark in recurring revenue this year. Its leader, Thomas Dohmke, has recently had ambitions in the gallery that the “new age” of AI-enabled programming would make GitHub a mainstay of the Microsoft group, both in terms of operation and revenue.
Can ChatGPT breathe new life into Microsoft’s Bing search engine?
According The information and the street newspaper, Microsoft would like to repeat this success in its other software, with AIs integrated into its Microsoft 365 office suite (Word, Powerpoint, Excel, Outlook, etc.). In Outlook messages, the integration of OpenAI tools would make it easier to find emails during a search, offer response templates to save time or even suggest alternative phrases. The possibilities are countless, and professionals are already appropriating these uses of ChatGPT. At the moment, the tool is not available via API (i.e., there is still no direct gateway to connect ChatGPT to its software), but OpenAI is already thinking about its profitability, in order to hope to exceed its goal of one thousand millions of dollars in short-term turnover.
At the same time, Microsoft is also considering integrating the powerful chatbot into its search engine Bing, which is struggling to exist against giant Google (3% market share versus 92.6%). ChatGPT would provide answers at the top of searches instead of links with few lines. But that integration would require tweaks: The model was trained on a fixed database — billions of lines of text pulled from the web between 2020 and 2021 — and therefore doesn’t update. This lack of dynamism causes him to sometimes give incorrect or outdated answers.
Furthermore, the chatbot only reproduces shifts and reasoning that it has studied: if the source material is incorrect, the tool’s response will be incorrect, which means that ChatGPT is not to be trusted. That is, not only should a possible integration with Bing include a continuous update of ChatGPT, but also care in the presentation of search results. In any case, this possibility worries Google so much that it is already organizing its response. The threat is even greater because Microsoft has already begun to integrate DALL-E 2 into a unique Bing image builder.
Microsoft, future shareholder of 49% of ChatGPT
Even before the launch of ChatGPT in December 2022, Microsoft and OpenAI had already started discussions about a new investment. The billion dollars injected in 2019 would only be used as fuel for five years. According to the media Semaphore, Microsoft would therefore be ready to put another US$ 10 billion on the table, in the form of an unusual agreement: the technology giant would recover 75% of OpenAI’s profits until the amount was refunded, then the startup’s structure would change. Microsoft would then have 49% of the company’s shares, the other investors with 49% as well, while the controller of the commercial arm of OpenAI, non-profit, would have 2% of the capital. as pointed out Semaphorethe contours of this operation are not yet defined.
At the same time, OpenAI is currently trying to sell part of its capital (including shares held by its employees), up to 300 million dollars, which would value the company at more than 29 billion dollars according to the street newspaper, that is, twice as much as in 2021. This performance would be exceptional in a context of reduced investments, although it generates very little revenue. Among potential buyers would be Founders Fund (AirBnB, Stripe, SpaceX…) and Thrive Capital (Robinhood…). In recent months, OpenAI CEO Sam Altman has repeated that his company does not plan on being acquired or going public. He has also taken steps to limit investors’ returns to 20 times the initial outlay, in hopes they are committed for the long term.
Is OpenAI deviating from its original promise?
Launched in December 2015, OpenAI had the ambition to eventually create an artificial intelligence capable of reproducing the functioning of the human brain – better known by the acronym “AGI” – in order to address important issues such as the fight against global warming. This start-up like no other has therefore been funded by a billion dollars since its creation by major investors in Silicon Valley, including Elon Musk (Tesla, SpaceX…), Peter Thiel (PayPal, Palantir… ), Reid Hoffman (LinkedIn) or even Sam Altman (Y Combinator), the latter assuming leadership of the project. OpenAI also had the support of Yoshua Bengio, one of the fathers of artificial intelligence, and recruited the best talent in the market with exorbitant salaries.
To frame its research, the startup adopted a charter in 2018, which contains its main ethical principles. In particular, he undertakes to ensure that his research works for the common good of humanity, to encourage cooperation with other actors in the field, and to publish the results of his research in open code (freely accessible to everyone, editor’s note). To ensure the application of its principles, the founders of OpenAI opted for non-profit organization status.
Problem: Lack of a business model quickly turned OpenAI into a money-burning machine, despite excellent search results. That’s why, in 2019, CEO Sam Altman created a second company, OpenAI LP, this time for profit. This new entity remains under the control of the previous organization, but aims to more easily raise funds and commercialize its “pre-AGI” technologies. It is therefore in this profitable branch that Microsoft is investing, with the aim of maintaining the advantage over its competitors thanks to the nugget’s artificial intelligence models. It now remains to be seen whether the quest for profitability will not lead OpenAI to gradually move away from its initial objective.