[OpenAi](https://inleo.io/@leoglossary/leoglossary-openai) is not a [company](https://inleo.io/@leoglossary/leoglossary-company-business) I am optimistic about.  If it were publicly traded, I do not think it would be a good long term [investment](https://inleo.io/@leoglossary/leoglossary-investment).


Many feel this is the industry leader when it comes to frontier models.  This was the case, with OpenAI jumping out to an early lead.  GPT 3 was certainly a game changer, putting the company in a position to [leverage](https://inleo.io/@leoglossary/leoglossary-leverage) its early standing.


That evaporated over the last [year](https://inleo.io/@leoglossary/leoglossary-year).


Today, xAI, [Google](https://inleo.io/@leoglossary/leoglossary-google-company), and Meta all closed the gap.  They are operating at the GPT 4.o level, with more on the way.  xAI has a 100K GPU cluster training Grok3, with another 100K due to be delivered to the company in Januaray.


Meta did not disclose what is being used for Llama4 but it is believed they are also over 100K with their cluster.


That is a lot of horsepower, something OpenAI is lacking.


# OpenAI: Challenges Going Forward


Sam Altman admitted the failure to deliver the [applications](https://inleo.io/@leoglossary/leoglossary-application) the company showed is due to a lack of compute.  It seems that when it comes to Jensen Huang, [Elon Musk](https://inleo.io/@leoglossary/leoglossary-elon-musk) and [Mark Zuckerberg](https://inleo.io/@leoglossary/leoglossary-mark-zuckerberg) are ahead of Altman.  xAI was able to gain priority on the Blackwell chips that are rolling out in January.  Elon was able to utilize his ability to get these clusters online (along with a premium payment) to get what he needed.


Altman doesn't have the same pull.


Here is where success helps.  Both Musk and Zuckerberg have put clusters together.  Meta has a long [history](https://inleo.io/@leoglossary/leoglossary-history) of that while Musk's xAI team set a record for going from box to online with its Memphis facility.  Huang likes this because, in the [world](https://inleo.io/@leoglossary/leoglossary-world) of compute, clusters online mean more orders are placed.


The key is to cycling quickly.   Getting clusters online means more [time](https://inleo.io/@leoglossary/leoglossary-time) on the accelerated compute.  In addition to the training, inference also enters as the answers come out quicker.  This is what xAI is looking at.


We could see Grok3 in two [months](https://inleo.io/@leoglossary/leoglossary-month), Grok4 in 5 months, and Grok5 in 8 or 9 months.  If this is the case, OpenAI will be far behind.  It is estimated that GPT 5 may not come about until May of 2025.  


## Its The Data Dummy


OpenAI has another disadvantage.  


Meta, Google, and xAI have a steady stream of data flowing in each [day](https://inleo.io/@leoglossary/leoglossary-day).  These companies have [social media](https://inleo.io/@leoglossary/leoglossary-social-media) platforms, with hundreds of millions of users (billions for Meta).  Each time they tweet, post on [Instagram](https://inleo.io/@leoglossary/leoglossary-instagram), or upload a [video](https://inleo.io/@leoglossary/leoglossary-video), the [database](https://inleo.io/@leoglossary/leoglossary-database) of these companies grows.


As more compute is added, the datasets have to keep pace.  This is becoming increasingly more difficult.  It is a topic that we covered in detail, yet it keeps coming up.


Synthetic data is questionable with regards to the degradation that will be realized over time.  While it is acceptable for a couple runs through, after that it becomes unknown.


[Human](https://inleo.io/@leoglossary/leoglossary-human) data doesn't have this issue.  The additional data is added to the training.  With cluster size growing (i.e. the amount of compute available), social media platforms can help to appease the data appetite.


OpenAI is forced to enter into costly agreements with publishers to gather the date.  This is not only a slower process, it is costly.  The social media companies has the data handed to them by willing users.


Once trailing, it becomes difficult to catch up.  Google and Meta have enormous resources behind them.  Their [money](https://inleo.io/@leoglossary/leoglossary-money) generated from [advertising](https://inleo.io/@leoglossary/leoglossary-advertising) buys a lot of compute.  xAI has Musk, a guy who seemingly gets richer by the day.


Altman did well to raise funds for OpenAI.  However, if he is a generation or two behind the others by the middle of next year, [investors](https://inleo.io/@leoglossary/leoglossary-investor) might balk at coughing up funds from something that could end up being obsolete before it is rolled out.


Then we have this from X:






 <center> https://img.inleo.io/DQmXnxgMtX4zR7JQqdzCVSxwoUms8NUuaTyya7dBTJCqwtm/image.png 


[Source](https://x.com/jokowords/status/1864662061784908177) </center>


It is not good when the loss increases in size as revenues go up.  That is something Altman is going to have to fix, and quickly.


Meta, Google and xAI have platforms that can help in the [monetization](https://inleo.io/@leoglossary/leoglossary-monetization) of the applications tied to their models.  They can also integrate aspect into the platform, something that makes others aware of what they are doing.


For a company like xAI, which also goes through funding rounds, this is of benefit as investors can monitor what is taking place.  With OpenAI, the reverse might be true.


Will this be its downfall in the end?  We will have to see.  For the moment, OpenAI might be at a crossroads where the other major players start to assert themselves.


____


[What Is Hive](https://inleo.io/@leoglossary/leoglossary-what-is-hive)





Posted Using [InLeo Alpha](https://inleo.io/@taskmaster4450le/openais-problem-going-forward-6oh)