There is a great deal of angst in the Tesla community around xAI and it boils down to three questions:
1. Is xAI likely to become profitable – aren’t there too many AI companies already?
2. Why did Elon not simply start an AI project within Tesla, just like Optimus?
3. Should Tesla become an investor in xAI?
I’ve given this a great deal of thought in the past week, and I’m going to present my conclusions here. They are my opinions, and I’m sure that there are many other, equally valid thoughts out there. Feel free to give yours.
The answers, IMHO, are somewhat entangled, one with the other, but I’ll try to separate them as best I can.
Let’s begin by talking about the business proposition. Is xAI likely to be successful?
LLMs are highly varied; one is not like the other, even from the same vendor. Each has a unique architecture, and maybe trained on a different set of data selections, with the result that each has unique attributes.
But in my view, that isn't the point: LLMs are not likely to be the ultimate product of AI companies, they will instead be the foundation (or an essential component) of new and unique architectures and applications to be built on the LLM and associated/related AI platforms.
Combining LLMs in unique ways with a wide range of AI and Coding techniques is going to produce a Cambrian explosion of many new platforms that are outgrowths of LLMs, often driving specialization and differentiation of the LLMs themselves.
These requirements may ultimately be as important as the data, reflecting the target use of the platform.
This brings us to xAI. We can imagine that a large component of the differentiation will be that the LLM will largely be based on
X.com data and will be in support of X (particularly as X expands its role into "The Everything App"). This helps clarify the equity involvement of X in xAI.
Beyond the obvious role on X, Grok will no doubt have enormous roles in other Musk companies. Obviously, Grok will provide information and in-vehicle utility to passengers of Robotaxis and Tesla vehicles (including those in the Boring co. tunnels), language and general knowledge to Optimus, and will also play many roles for SpaceX and Neuralink.
I can expand at great lengths about the use of AI in each of these companies, but I'm sure readers can conjure these up for themselves.
Now to the claim that Tesla has AI experts and that they could "just do AI’.
We begin with the proposition that building a general, symbolic-based model AI versus building FSD, or “real world AI”, requires an entirely new, dedicated hardware platform and team.
Why a dedicated hardware platform? Because creating a foundation model requires a functionally limitless amount of training (as does FSD). In an excellent series of posts on July 26th,
@_LouiePeters details the extraordinary amount of training required by the leading foundation models, and points to the training clusters already installed and planned by xAI. He estimates $4b for the current xAI 100k H100 training cluster (of which 32k have already been installed, the rest being installed this summer), and “
@elonmusk noted plans for a 300k, B200 cluster next summer… ~$15bn Capex”
In summary, the hardware platform, in the next year or so, will require $19b of Capex, before the cost of energy, general operating expenses, and AI Engineers.
Having established the size and scope of the dedicated AI hardware platform, we consider the people part of the equation.
We know that these are currently the most sought-after engineers in the world today; we also know that their focus is not as much the salary they will command (which will be high), but the ability to profit directly from their work. As Elon has pointed out at length, they would not be attracted to equity in Tesla, because their earnings would be intermingled with those of the other Tesla businesses. While these may be good, even stellar, they would not be fully reflective of and geared to the direct AI bet.
Parenthetically, we may ask why has Tesla attracted such great talent to its FSD team, if not able to attract AI engineers to a Foundation Model team? The answer is simple and clear – because in the immediate term, a huge component of Tesla’s earnings (and therefore share price potential) is geared directly to the success of FSD.
But returning to the question of xAI – we have outlined that there is need for $19b of Capex in the next year or so, as well as the need to hire a team of the leading AI experts in the world. In short, an investment of at least around $20b, and probably significantly more until revenue and profit.
Is this a bet that Tesla wishes to make right now, or indeed can make, given Tesla’s own ambitious capex plans for its current projects? These include the multiple new vehicle models, FSD development, Robotaxi, the Semi, Megapack Factories, and above all, Optimus? Tesla has a Capex plan that exceeds $10b in 2024 and be between $8 to $10b in each of the following two fiscal years, or a total of around $28b, compared to the approximately $30b of cash on hand (see Q2 10k filing, page 28.)
Even more important than the question of the use of cash is that of near-term profitability. Until the maturing of the many projects detailed above, Robotaxi above all, Tesla’s current earnings and cash generation will remain constrained.
Is this the time to launch a new, $30b startup project? This is a project:
(1) On which Tesla’s other projects, while they may benefit, are not critically dependent
(2) Would consume all of Tesla’s current cash reserves
(3) The success of which, at best, is several years distant
IMHO the answer to this question, given the facts, is “No, Tesla should not commit $30b to this venture right now."
On the other hand, would it make sense for Tesla, like X, to make a judicious investment, say $5b, in xAI at this time?
IMHO the answer to this question, given the facts, is “Yes”.
It would give Tesla a meaningful stake in one of Elon's biggest bets, and one which could benefit Tesla particularly, and all his companies in general, and which could be of great value in a future.
By taking a minority stake we are betting a relatively small proportion of our overall cash position, a reasonable bet given our financial position and risk profile.
This is why I voted "Yes" in Elon’s poll, and why I didn't think Tesla should be the major, or even sole shareholder in Xai. Apparently the 1m or so voters who joined me overwhelmingly agreed. Now let’s get the board to get it done! We need
@TeslaBoomerMama to get on the case!
Here are the posts from
@_LouiePeters:
nitter.app/_LouiePeters/sta…
and
nitter.app/_LouiePeters/sta…