So "facts" now are "I uploaded a pdf to a LLM, and asked about it". But if you know how to read a balance, income statement and cashflow, your skills are poor.
2,723 karma · joined June 22, 2018
Can we be sure of this? The second they tried to hike the prices to be profitable, they lost a lot of clients, or a lot of clients lowered they token expenditure. In any market margins tend to go to zero. I am using frontier models, but only because they are dirty cheap if not free. If they raise the price, I just switch to a cheaper model, start exploring the option of renting a GPU-VPS or buying my own card. Or reducing my AI usage, that currently includes stupid useless things like "translate this cooking recipe from french... for free".
> But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
They are being priced as the "ifs" are certainties, and not only a possibility. If anything goes a millimetre below perfect for them, their finances collapse. High margins attract competitors, and unless you have a huge moat (and their only moat is expensive training), your high margins vanish.
https://www.sec.gov/Archives/edgar/data/2130357/000162828026...
Income at page F-7 (about 350 overall):
(units in Kilo-millions)
Revenue in first 2026 9-months 33.2
Compute and infra expenses 18.8
Personel 4.8
Sales 5.5
Administrative expenses 5.3
Operating earnings (loss) (1.2)
Q-3 only, this is calculated substracting first half from 9 first months:
Revenue in Q3 16.9
Compute and infra expenses 9.9
Personel 1.7
Sales 2.3
Administrative expenses 3.6
Operating earnings (loss (0.6)
If VC money gets tight, they would have to issue debt, that pay interests. With a current negative EBITDA, their bond will be junk at IPO, paying higher interests and having a lot less market than AAA debt (e.g. Google has access to that market). Both things combined translate to "running out of cash".
But this is a bet we can settle right now: https://www.sec.gov/Archives/edgar/data/2130357/000162828026... , the ammended S-1 with Q3 data (that for some reason they omited from the S-1 that made big news), page 23:
(units in Kilo-millions)
Revenue in first 2026 9-months 33.2
Compute and infra expenses 18.8
Personel 4.8
Sales 5.5
Administrative expenses 5.3
Operating loss 1.2 (they are still losing money before ITDA)
Revenue in first 2026 6-months 16.3
Compute and infra expenses 8.9
Personel 3.1
Sales 3.2
Administrative expenses 1.7
Operating loss 0.7
First 9 months minus first half:
Revenue in Q3 16.9
Compute and infra expenses 9.9
Personel 1.7
Sales 2.3
Administrative expenses 3.6
Operating loss ~0.6 (Remember, this is EBITDA)
But the reality is that the burning rate of ICE cars is a lot higher than electric cars (only electric, that don't carry a heat source along the battery), by far: https://www.motor.com/2026/07/ev-fires-vs-ice-fires-myth-vs-...
Same tune for diesel generators: they catch fire, fumes must be correctly ventilated or you get a flammable atmosphere, rooms where they are installed must be seriously protected against fire hazards (https://global-generators.com/blog/diesel-generator-fire-saf...). But when one catches fire, only local news report it, nobody outside town ever get notice of it, and people tend to think they are 100% fireproof. Meanwhile, a big lithium battery catches fire in California, and the news might even go global.
Anthropic filling shows negative EBITDA, how is that profitable? Revenue < (computing + personal + selling expenses). That is not a business, that is a money burning furnace.
That before even taking into account the huge amount of money they promised to datacenters (i.e. "build that for us, and we would rent it for 5 years", non cancellable).
Or datacenters are the old mainframes, and when they stop building them and squeezing the market, GPU prices go back to what they were before the frenzy or even lower, so anyone really interested can buy a decent GPU for a fair amount of money, and run models locally for peanuts. Meanwhile, dataframe owners go bankrupt.
Valid: if someone is hit by a train, he dies. Someone was hit by a train, then he is dead.
Invalid: if someone is hit by a train, he dies. Someone is found death, thus he must have been hit by a train.
If that was a plane crash, rarely pilots are to take all the blame and even if they are the main cause, the recommendations would be to add more training, more off time, redundant measures to help the pilot or overcome their mistakes and they usually do deep investigations that end in plane redesign / retrofit, changes in routes, adding safety equipment, etc. It never goes "human error", and close the report.
With the current news this was not a manteinance issue (that might be a problem anyway), but a failure in "control tower" that put two planes in the same runway, with one of them stopped in the middle. It is beyond logic how two trains can be in the same railway closer than X distance, even with human allowing it, instead of having a TCAS-like system.
Audited, revenue in 2025 is $4.6 with a computing cost of $7.3 (i.e. heavily subsidizing tokens). Unaudited, revenue in 2026 (first six months) is $16.3, computing cost $8.9 (1.83:1). But after adding personnel, sales and administrative costs, they are still spending $17.0 to get that $16.3 revenue (still subsidizing tokens, but less).
What you call "dumpster fire", I call "steps towards uv", maybe even better tools in the future.
According to https://www.theguardian.com/technology/2026/oct/06/google-da..., they had to halt the construction because they failed to assess the environmental impact. The same article you say it doesn't refute that the leveling was illegal says that at least four areas that must be protected are now leveled. It sounds to me that Google is playing the card "it is done already, just fine us and let's move on" thanks to their deep pockets.
If you feel that your idea is worth IP protection, just protect it as a trade secret, not as a patent or copyright. Very famously Disney have a copyright over drawing three close circles.
Germany still uses coal, but I wouldn't say they are heavily dependent. They consume ~100TWh per year of coal (20% of total), down from ~300TWh in 2015 (50% of total). They plan to close them all by 2038, and they are on the way to do so even before that.
But coal is a political problem because the are producers, thus a good chunk of voters work in that industry. Political parties that promise to keep coal running could still get many votes.
If any of us did a tenth of what they did, we would be already in jail.
Did we? Where are those fusion plants installed?
> Not compared to Nuclear, Fossil Fuel, or PV R&D + subsidies
You are probably referring to this document: https://www.eia.gov/analysis/requests/subsidy/pdf/subsidy.pd... . A huge share of the $183.3 billion are tax expenditures, which is to say they reduce taxes on activities related with energy (e.g. nuclear decommissioning is tax free). That is not the same as throwing money at those technologies R&D, but it counts as subsides because they can operate more cheaply. Only $11 billion of the $183.3 is true expending. But they mix R&D with other direct expending, like guaranteed loans. In Figure 11 we can see that R&D for all renewable energies is less than $300 million.
Fusion energy don't want tax expenditures, they want cash and lots of it.
We don't know this. There are a number of investments that fail and the investors need to let go or you get into the phallacy of sunken costs. E.g. wave energy was a failure, thermo-solar under delivered, algae based fuels never scaled up, and ethanol fuels from waste proved very difficult to engineer, syngas was a ecological nightmare... All of them received significant investment, and they were abandoned, pivoted or relegated to niche applications. I personally would throw fusion in that same group of failed investments.
Luckily for us, solar PV (paired with batteries) is a huge win that justifies all the above failures: for decades it was a dream for the 25th century, but some people make it true in the 21st.
Not every spending/investment ends up in success! Fusion is one hurdle after another, yet it advocates still believe that is just a throw-more-money-at-it problem. I am not so convinced. Of all the failed (for now) energy sources, fusion is the biggest money sink. France spent 25 billion, plus 7 billion in private funds. It is not like they are working with peanuts, at those quantities you have to show some serious results, maybe facing scaling problems rather than basic development problems (e.g. ethanol failed at scaling, when it already proved viable on smaller scale. Fusion is not even in "we have this toy reactor that works we are fighting to scale" phase).
E.g. the -80°C's in our lab are all naturally air gapped: the monitoring system is send-only.
That is a bad thing to do. We diversified bets, and we end up finding a couple of pretty good energy sources, plus making some of the existing ones cleaner and safer.
Had we put all our money in fusion, today we might be drowning in cheap clean energy... or still living in a dirty coal-fuel environment, with a few old and dangerous fision reactors for lack of investment in development. A very risky bet.
In today's standards, or at least what I am used to, Franklin and Wilkins should be co-authors of the "Molecular Structure of Nucleic Acids: A Structure for Deoxyribose Nucleic Acid" Watson & Crick paper. But they only got this note at the end: "We have also been stimulated by a knowledge of the general nature of the unpublished experimental results and ideas of Dr. M. H. F. Wilkins, Dr. R. E. Franklin", not even talking about the image but giving the impression that they had an unimportant talk while having an informal tea.
But they (W & C) include this text: "The previously published X-ray data on deoxyribose nucleic acid are insufficient for a rigorous test of our structure. So far as we can tell, it is roughly compatible with the experimental data, but it must be regarded as unproved until it has been checked against more exact result". They suck for this: they could ask Wilkins & Franklin to include the X-ray that is sufficient to base their claims, add them as authors, and make a better article. But they choose to "be stimulated" by the perfect picture 51 and claim ignorance afterwards.
Kobo had its store on the home page, and while it was unobstrusive, the Pocketbook has nothing at all. Worth a look, if you are interested.
Bought, yes. Installed and in use? We don't know, but probably on a warehouse waiting for a place to be installed, for an energy source to be build.
Put the GPUs in another business, sell it to investors, and watch it unfold.