How OpenAI became a $150 billion company—and lost its way
On its path to $150 billion valuation, OpenAI transformed itself and the industry without securing future success.
OpenAI has just raised $6.6 billion in funding at a $157 billion valuation to become one of the most valuable private companies in the world. The company now has a good chunk of cash to spend on training and shipping better models.
But all this money has not come for free and it has transformed not only OpenAI but also set the tone for the AI community and industry.
Fractured direction
The first direct consequence of the money being poured into OpenAI is the change that the company itself has gone through. Funny enough, this was never an organization that was supposed to be talked about in terms of valuation and revenue. It was founded as a non-profit that was going to develop artificial general intelligence (AGI) for the greater good.
Down the road, they became too focused on large language models (LLMs) and realized that they would need more cash. They changed to a “capped profit” company and started raising money from investors in exchange for up to 100X returns. The non-profit was supposed to remain in control and Sam Altman, the CEO, was not supposed to hold any equity.
The change in structure allowed them to continue their research. But they were now beholden to the whims of their financial backers and had to develop models that could be productized and turn in profits.
Ahead of the new funding round, news broke that OpenAI is “working on a plan to restructure its core business into a for-profit benefit corporation that will no longer be controlled by its non-profit board.” Oh, and Altman will get to own equity that will potentially raise his net worth to billions of dollars.
Balancing between scientific research and commercial goals was already a tough needle to thread even before the restructuring. Now it will become even harder for them to avoid becoming too embroiled in selling products. Their only chance at achieving their original goal is that scaling LLMs will actually get us to AGI.
On its way to the $150 billion valuation, OpenAI has also left a trail of drama, including a feud between Sam and co-founder Elon Musk; the exodus of other co-founders, early members, and key personnel, including Ilya Sutskever, John Schulman, Andrej Karpathy, and Mira Murati; the founding of Anthropic AI and Safe Super Intelligence by disgruntled former employees; and a slew of lawsuits around copyright infringements.
An obscure industry
Another important consequence of OpenAI’s rise is the effect it has had on scientific research. In the span of a few years, the AI community has mostly gone from sharing and collaborating to hiding and obscurity.
Leading AI firms no longer release the weights, code, training data, or even the architecture of their models. The models only exist behind APIs. Detailed papers have gradually turned into “technical reports” that don’t contain any information that would allow other researchers to reproduce the results or even catch hints at the techniques used in the models. More recently, even technical reports have been ditched for short blog posts and model cards.
OpenAI’s latest model, o1, doesn’t even share its reasoning trace for fear of tipping off competitors. To be fair, there is still a large part of the industry that is committed to open source and open research, with Meta leading the charge. But unfortunately, openness is gradually becoming the exception, not the norm.
And with so much money being poured into training models, it is only natural that companies like OpenAI would want to keep their discoveries secret to maintain their competitive advantage. This will enable them to secure more funding in the future as they are still burning through cash.
But the lack of transparency is slowing progress as it trades collaboration for secrecy. So much money is being spent on parallel efforts that are leading to the same results. In fact, the success of OpenAI and LLMs has sucked the air out of the room at the expense of other research directions that can yield impressive progress in the long run.
Uncertain future
Despite its success and latest round of funding, it is not clear whether OpenAI will be able to cement its position and survive the wave of AI hype. OpenAI is by far the most successful AI startup with a projected $3.7 billion revenue in 2024. It has the largest share of the market for both consumer and enterprise LLMs. But it is also burning through $5 billion per year.
Expenses will continue to rise as models grow bigger and data becomes more expensive. At the same time, other companies will stay on OpenAI’s heels, preventing it from cornering the market. LLMs have already become commoditized. For most applications, there are plenty of suitable closed and open models. OpenAI still has a brand advantage, but it does not have a moat that allows it to charge a premium above other companies. And for many enterprises, as their LLM applications mature, they move from closed platforms to open source models that cost less and are more controllable.
I don’t see any reason why one of the tech giants doesn’t manage to outspend and outpace OpenAI in the long run. Google had a rough start, but I think they’re on track to take the lead. It remains to be seen what will become of OpenAI. But for the time being, they have made history.
I think the points raised in the article are not that crucial for OpenAI, meaning the commercialization, non-transparency, if it will lead to AGI.
Reality is more mundane. OpenAI must show it can deliver an assistant that people can use to reliably automate some boring work, and pay up for that.
Surely Google can take the lead. But companies like Apple, Amazon, etc., view Google as a competitor, and given the choice would license or integrate OpenAI's tech instead.
There's enough space for a competitor to Google. The question is what will happen to Anthropic, Mistral, and Zuck's efforts.