From: allin
OpenAI, initially a non-profit organization co-founded by Elon Musk, has undergone significant changes in its structure and market approach, leading to intense competition in the AI space.
Financials and Structural Changes
OpenAI recently announced a 157 billion [00:57:53]. The company is projecting $3.7 billion in revenue for the current year [00:57:59], with its revenue stream increasingly resembling a SaaS business model rather than just API token pricing [01:28:28]. A notable aspect of their recent deal is a “poison pill” clause, requiring OpenAI to convert to a for-profit entity within the next two years, or investors can demand their money back [00:58:05].
This corporate restructuring has faced internal and external challenges. The resignation of CTO Mira Murati and two other top researchers prior to the funding announcement was widely seen as a protest [00:58:18]. Furthermore, Elon Musk is suing the company to stop the for-profit conversion [00:58:26], and Meta has joined him, sending a letter to California’s attorney general to halt the process [00:58:44].
Shifting Market Share
A chart analyzing recent trends indicates a significant shift in OpenAI’s market share, which has decreased from approximately 50% to about 33% in the last year [00:59:46]. This decline coincides with gains by competitors:
- Anthropic has reportedly doubled its market share [00:59:53].
- Meta has maintained its position [00:59:56].
- Google is “picking up steam” [00:59:58].
Competitive Dynamics
The AI market is characterized by an “arms race” primarily driven by hardware (GPUs), data access, and model quality [01:01:24].
Hardware and Capital War
Companies like Google, Amazon, Microsoft, and Meta, along with ventures backed by individuals like Elon Musk (e.g., xAI), possess the ability to attract “effectively infinite capital” at a low cost, giving them an advantage in the hardware war [01:01:31]. xAI, for instance, has demonstrated the ability to scale up to 100,000 Nvidia GPUs in a single contiguous system, with plans to reach a million within the next year [01:00:50].
Data and Experience
While there’s a perceived “terminal asymptote” in basic model quality [01:02:03], companies are now focusing on user experience and the integration of unique data sets. For example, the vast corpus of data on X (formerly Twitter) and kinetic data from Tesla, both controlled by Elon Musk, could provide an additive information pool for training models [01:02:26]. Google also has a significant advantage with its access to YouTube’s vast video data, estimated to be a billion times more than text data [01:30:02].
Model Promiscuity and Commoditization
Businesses are becoming “completely promiscuous” in their use of AI models, selecting from 30-50 different models based on cost, quality, and specific task requirements [01:02:53]. This trend suggests that AI models are rapidly becoming commoditized [01:03:41].
Open-source initiatives, particularly from Meta with models like LLaMA, serve as a “counterbalance,” limiting how much companies can charge for their hosted models [01:06:30]. This dynamic drives the price of AI tokens toward the cost of compute, plus a small margin [01:04:55].
Outlook for OpenAI
The speakers predict that OpenAI will face increasing challenges. One perspective suggests that OpenAI will eventually become the third, fourth, or even fifth player in the market, rather than holding a leading position [01:34:22]. This is attributed to the “compounding effects” of larger companies with inherent data, infrastructure, and personnel advantages, such as Google, which has “woken up” and is on “full assault” with its Gemini models [01:28:35].
Despite the intense competition, the overall market for software and AI-powered services is expected to expand dramatically, creating new categories and automating jobs, potentially offsetting the deflation in legacy software systems [01:14:45]. The focus for AI companies will be on building differentiated value on top of the core AI models [01:09:10].