AI Newsround – OpenAI & Google

Open AI vs. Musk – Cop out.

• The trial of the year has ended with a whimper, with Musk losing on a technicality, but this means that OpenAI can refocus on its business and that the way is now clear for an IPO in 2026.
• After just 2 hours of deliberations, the jury decided that Musk’s case fell outside the statute of limitations, which means that too much time had passed between the alleged event occurring and Mr Musk filing the complaint.
• This is unfortunate as it means that there is no resolution when it comes to the issue that was being litigated, namely that OpenAI abused its position as a non-profit entity.
• I think that in the court of public opinion, this has already been litigated and lost by OpenAI, but to be fair to the company, I am not convinced that it acted in bad faith.
• When it was founded, there was no way to know about the gigantic financial opportunity that it would go on to enable on the way to creating artificial superintelligence, which it is still very far from doing.
• However, the issue is now moot, and what really matters is that the way is now open for OpenAI to focus back on its business, which is critically important as lack of focus and distractions have been large risks for some time.
• It also clears the road for OpenAI to go public, which, if it is going to do so, it needs to do it as soon as possible, while demand for AI paper is lower than supply.
• I still don’t think that OpenAI is any financial or structural shape to go public, but that won’t stop a frenzy if it can offer its paper before Anthropic does.
• I still think that there is a risk that the market declines to pay the valuation that OpenAI is looking for, and now, with the court case out of the way, it is likely that we are going to find out.

Google & Blackstone – All about economics

• Google has entered into a venture with Blacksone where Google’s silicon will be offered for rent, but the success of the venture will depend entirely on the economics it can generate.
• Blackstone is putting in $5bn and will be the majority owner, with Google putting in a smaller amount and the remainder coming from other financings, which I presume means debt.
• This is where I start to get nervous as the entire venture hopes to raise around $25bn from which it will bring 500MW of capacity online in 2027.
• If I assume that the entire $25bn will be spent on 500MW of compute, this puts the venture broadly in line with the cost of building the same compute capacity using Blackwell.
• This looks expensive to me as the $50bn / GW estimate for Blackwell includes Nvidia’s gross margins of 70%+, and so this should be possible at a lower cost, assuming lower gross margins.
• This also means that unless the venture can generate significantly more than $10bn / GW, then it is likely to struggle to earn a decent return for its investors.
• I also have doubts whether Google and Blackstone can stand a data centre up that quickly, and so unless they are taking over something that is already half built, I suspect it will be more like 2028.
• Hence, with this venture in compute, just like everyone else, everything will hinge on the revenue per GW that the company can generate.
• At the moment, somewhere around $25bn / GW is what everyone should realistically be shooting for, which is very far from the standard $10bn / GW made today.