More Tales From Nvidia: The Furious Race to Get IPOs Done and My Objections to ARR (Issue #242!)
It looks like Broadcom ($AVGO) is already back for another fix.
It really makes me wonder how there is enough savings in the world to do this, although the Treasury market is telling you there isn’t:
Broadcom CDS Explodes As It Seeks Up To $100 Billion In Massive Off-Balance Sheet Debt Deal
At any rate, it is interesting that both Anthropic and OpenAI seem to be furiously racing to get their IPOs done. It almost sounds desperate — from both them, and their mouthpieces. What is the rush? Is the rush perhaps that they see things starting to get really ugly and they want to get public before that happens? It seems to me neither company is in any shape to go public.
The top level picture is ugly for the industry. Slowing token growth and a massive decline in token pricing.
The details of OpenAIs last quarter, as reported by the WSJ, were incredibly ugly. A sharp deceleration in revenue growth, they lost $12.3 billion and they lost almost $2 for every $1 of revenue. More executive turnover and turmoil lately. From my perspective, no way this is a public vehicle. And I also thought it was a non-profit? Maybe losing almost $2 for every $1 of revenue is what non-profit means?
Although Anthropic’s last hyped quarter seemed to be OK on the surface — that was peak tokenmaxxing and they benefited from what appeared to be a 1x subsidy from SpaceX ($SPCX). But now, things seem to be slowing for them too. There is this data showing their ARR from coding flattening all of the sudden, after a sharp increase:
Others are now claiming their total ARR missed expectations recently, which would make sense in light of the overall macro trend of crashing token prices and slowing growth:
Why the rush, and how can these companies go public in the face of this?
I think I know why the rush, but I am not sure how they can go public while at the same time knowing how they can go public.
They go public with sell-side models that are about as good as all the models Cathie Wood and ARK made up for Tesla ($TSLA). There will be models with completely made up numbers going 10-20 years into the future that are completely unforecastable, and a giant multiple will be put on those numbers as well. The way the models will be made start with this query from the boss: “pull numbers out of your ass, back into a model that can be used to justify the IPO price.”
The IPO is the ultimate end of the shell game that has been played to this point. These things are not investments, they are just part of a shell game. The valuation is walked up round to round by the venture guys and their own customers, just because. It is all a bet on the notion that these businesses can be taken public at any price. Whatever the last rounds private valuation was, Wall Street will be able to make up a number for the IPO that is substantially higher. Then they just hope the public mania continues, and they can get out.
But the challenge is obviously getting harder. The amount of money/savings in the world is growing short. SpaceX didn’t work out great. The theoretical market caps of these things are all gigantic. The unlocks are huge. It is very challenging for the stocks to be the rocket ships people want them to be. It is especially challenging in the face of what appear to be sharply decelerating fundamentals.
They don’t belong public, yet the rush to go public is on. They also want and need the money to stay alive at the rate they burn it. While pre-mature IPOs may extend the shell game for awhile longer, ultimately they may be the beginning of the end. Both Anthropic and OpenAI have had the benefit of operating under the cover of darkness. Once they go public, the sunshine will be let in for everyone to see.
Side note, good and funny:
A few more thoughts:
* Big picture, the whole AI trade and industry, after seemingly starting to stall out about 20 months ago for a variety of reasons, engaged on its big second leg which re-ignited reflexive momentum and revenue growth in the space for two primary reasons: It became clear the U.S. government was going all in on the sector (from a stock perspective, following the government might be more important than following the Fed) and then the wave of increasing stock prices and unheard of levels of circular financing emerged, which goosed the whole space from a revenue and stock price perspective. It really was a reflexive investment and stock price momentum boom.
Now, political pressure is going the other way. It is clear where Democrats stand. And now, even Republicans are going in the opposite direction, and hard in the opposite direction. The general population, globally, hates AI.
Governor Abbott Directs Comprehensive Data Center Audit
Governor Ron DeSantis Signs Law to Protect Floridians from Subsidizing Data Centers
The circular financing that was part of the reflexive feedback loop cannot accelerate from the levels it is at and I doubt it can sustain the levels it is at. The big spending public companies ($GOOGL, $META, $AMZN, $MSFT, etc.) are now different combinations of cash flow negative, full of debt, or full of debt plus off-balance sheet commitments. The amount of market cap ascribed to the space now is just massive and might be getting bigger depending on what happens with OpenAI and Anthropic. There continues to be substantial equity issuance (including insider unlocks) and debt issuance. And the economics of the industry remain highly negative. On top of this, the world is short savings to finance all of this stuff.
Just like the political cycle is turning, it feels like the funding cycle has at best peaked as well. Said another way, the second derivative of all of this is likely to turn negative.
* This is interesting, one more potential risk on top of what I just cited along with open source and the overall lack of economics in the industry. A potential move to much cheaper and much more efficient small language models:
If this is true, the hyperscalers are toast
* I have a huge issue with how the frontier model businesses (and their hypesters) like OpenAI and Anthropic position their revenue. They refer to it as ARR, as in Annualized RECURRING Revenue. I do not know they can be allowed to do that as part of an IPO, and how their auditors and bankers can stand behind the use of any term other than “REVENUE.”
ARR is capitalized much differently than normal period revenue (ergo revenue Walmart ($WMT) reports for selling food and general merchandise for example). ARR is viewed as more valuable revenue, because it is perceived as annuity revenue that will not go away. Annuity revenue is much more valuable than period revenue for that reason.
Which is why companies try to build their revenue models around ARR, and then seek to capitalize themselves on that basis as well.
But, ARR really is an overhyped concept. I understand why companies build their revenue models around it, and why they try to capitalize themselves on that basis, but very few things really are ARR. Nearly all the companies that sell software on this basis are doing it with a combination of either one-year contracts, or 3-5 year contracts that come with a discount.
However, that too can also go away, and people know it. That is why investors sold the stocks off hard recently during the AI meltup. At least, in the case of those businesses, companies and auditors had a reason to refer to it as ARR, because they had a reasonable expectation it would be annuity like revenue, even though there was no guarantee, as is now being born out.
In my view, in the case of the frontier models, I do not understand how they can they consider their revenue to be anything close to ARR and anything different than normal revenue. I can all go poof in seconds, for a variety of reasons:
* We know about the shift to open source models. Even AT&T ($T) is now doing this. And they are as big and dumb and low risk of a company as you can think of. You would think they are exactly the type of company that would overpay for the frontier models, kind of like the old notion of buying from IBM ($IBM). But nope, even they are shifting to much cheaper open source. Think about what more tech savvy companies are doing and those that are smaller and more nimble as well:
AT&T is Using Open Source Models to Curb Anthropic Bills
* Customers can shift between models, even the frontier models, whenever they want. They have already done this. There are no barriers to entry or lock in or switching costs. Whatever model is the best at any given point in time, they can switch to. Business seems to flip flop all over the place. It is a commodity product.
* Then there remain to be all sorts of future technology risks we do not even know about. A potential move to small language models, as linked to above. Or new technologies and approaches that emerge, like neuro symbolic AI or world models, or who knows what else.
My view, the revenue the frontier models have is no different than the revenue Netscape, Yahoo, WeWork, Nike, Kodak, Zerox, Polaroid, Mikes Buggy Whip Company or Pete’s Vinyl Record Company had. It is not ARR and should not be quoted that way. Nor should the auditors or bankers allow for that, in my opinion.
Further, I have argued (above) that these companies do not belong public and should not be public but are rushing to go public because they (and their investors including the circular ones) have all of the exact same concerns I do. I ask again, what is the rush? I think their own behavior in this regard speaks volumes.
Relatedly, these are two good substacks on the issues I have cited this morning:
ARR vs ARR. Watch out for this one sly trick.
BREAKING: More bad news for the frontier AI companies
Position: None
