Collusion is a viable strategy, regardless of the market situation. They've been shown to apply price fixing in the past; presumably they still know each other's phone numbers.
"[lithium ion battery prices] declined by more than 99%. In 1991, lithium-ion batteries cost around $9,200 per kilowatt-hour — 33 years later, they cost just $78."
The scale we're talking here is not just a shack with a couple of batteries. Even with low prices of today, we're talking several millions of investment for 10 MWh.
> The price of battery storage is dropping exponentially
We're at the flat end of things. Even if this trend continues, there won't be another 99% reduction coming.
> Eventually, the price becomes pretty trivial.
Only if you can scale manufacturing and materials the same way. Hopefully not with Lithium, because that's just not abundant enough.
> The scale we're talking here is not just a shack with a couple of batteries. Even with low prices of today, we're talking several millions of investment for 10 MWh.
1.78 million USD at current exchange rate. Before sales tax, but if you're buying at this scale, you claim that tax back anyway:
> Only if you can scale manufacturing and materials the same way. Hopefully not with Lithium, because that's just not abundant enough.
1. Sodium is already cost-competitive with lithium, for batteries.
2. There's loads of lithium all over the place, the hard part of using it is that it's spread out everywhere rather than particularly concentrated. Like, we're talking 150 million tons of "reserves", 250 billion tons dissolved in the seawater, and plausibly 5 quadrillion tons in the bulk of the planet's rock.
You could make hamster wheels an attractive energy source with enough tax refunds. That's not an especially compelling argument when arguing about the global ramifications of a new technology. Tax breaks don't magically make alternative fuel sources better on any given axis. They shift incentives, but that power isn't infinite.
In the EU, all sales taxes are refunded to businesses, it's available for literally everywhere in the supply chain.
It's called "Value Added Tax" in English (UK has same rules despite Brexit), because it works out as the tax on the value you add: Business 1 digs €1 of stuff out of ground, charges their customer e.g. 15% VAT; business 2 buys it for €1.15, claims that €0.15 back always, processes the stuff into €2 of new stuff, charges their customer (in this case) €0.30; business 3 buys it for €2.30, claims that €0.30 back always; etc. all the way to the end customer or export market.
If you're spending millions on power management equipment, you're absolutely a business, and absolutely will get to claim this tax back, and you effectively owe tax only on the net benefit it brings. Electricity is also subject to VAT, though the rates are allowed to vary, and the e.g. UK electricity VAT rate is 5% while the general rate is 20%.
Look at the graph in the article. The trend is clearly log-linear at the tail end in 2024, which is to say, still dropping exponentially. People said the same thing about solar for years, of course. Exponential decreases in cost are apparently hard for people to reason about. As others say in adjacent comments, there's lots of lithium around, and lots of promising work on salt based chemistries for mass storage. Plenty of room for progress.
I'm not a fan of Jensen by any means, but he did start the CUDA program some eight years before it was actually useful (the alexnet moment). He called the need for high performance linear algebra for scientific computing, it just took a while for it to find its first killer app.
That was becoming a hot topic at the time prior to DL getting on the map but it was smart to recognize the latent demand for GPGPU. Was still going to lead to solid growth with AI was just not going to be the exponential growth we saw if it didn't happen.
CUDA was well-used in HPC and other compute-intensive spaces long before neural nets were in style.
If you look at the evolution of GPGPU computing, at least from an HPC perspective, CUDA is almost an obvious step. What’s perplexing is that AMD and Intel let it get so far ahead in the late 2000s and early 2010s.
HPC was a really small market for a long time. I meant obvious from a tech perspective; the business case was clearly not as cut and dry.
If you look at the evolution of the HPC space in the 2000s, you see a string of advances, from people hacking fixed-function pipelines (iirc, mipmaps were used very creatively) to writing assembler with special driver extensions, to using shaders, and eventually CUDA.
Still there were a lot of attempts. Intel seemed culturally stuck on amd64, out of which came the failed Xeon Phi. AMD seemed more focused on the gaming market, and had capital issues during that era. After its Xserve product, Apple stopped all HPC efforts and more or less didn’t have competitive GPUs until the M1, and calling that competitive with a discrete GPU might be a stretch.
There was an attempt at OpenCL in the early 2010s which failed for reasons that were never clear to me other than NVIDIA rejecting it as a competitor.
Um what? You seem confused, they didn’t measure a year with 372 or even 364 days. They measured 28 days. And it’s almost a certainty that taking the average spend over the last 28 days and multiplying by 13 is higher than taking the average spend over the last 31 days and multiplying by 12. Especially if we believe their revenue growth numbers.
> it’s almost a certainty that taking the average spend over the last 28 days and multiplying by 13 is higher than taking the average spend over the last 31 days and multiplying by 12
This is generally true of any growing time series. If you don't want this look at the audited GAAP figures.
I suppose one would end up with checks happening on a random second instead of a random minute - reducing the size of the spike by a factor of 60, but still spiky.
1. That's fair, I didn't realize people were expecting my comment to be perfectly-aligned to every aspect of the thread parent, including the "their own hardware" part, but now I at least see where the confusion came from. For the record, any issues at this level of the stack are also present in IaaS, not just on-prem.
2. I still don't see where you got that they are only talking about configurations versus messing up an implementation (I.e. in a paramater they coded incorrectly)...
But that aside, even on AWS you're still configuring your proxies, gateways, load balancers, etc. AWS gives you just as much rope to hang yourself with as a box in a colo network cage does. Companies are already doing this stuff themselves, on-prem is not more security work. IaaS is about scaling and maintenance load, not security being handled for you.
"If you just waited a year, would some other AI produce that same proof independently?"
I had a conversation about this very problem last November, with a visiting mathematician. It's essentially signaling a deflationary spiral in intellectual work: why bother doing the work now, when it will take half as much effort a year from now?
This is perhaps always a possible stance, but it's exacerbated when some proportion of your own agency is handed over to, well, agents.
This might hold up for theoretical intellectual work without any economical value, but if you are researching anything that can be applied to the industry, you will instead have 1 year of market ownership and first-mover advantage.
I think it says more about what the value of math as intellectual work is, than about intellectual work in general.
Like if an Egyptian historian would come and say "let's wait one year until the methods are good enough to decrypt this language", no one would be bothered or surprised.
You imply "worth" being equal to some known economic value.
You even imply, economic value was determined correctly in advance for arbitrarily long future intervals. And the applicability was known in advance to begin with.
Obviously, you're completely wrong.
Applying simplistic models to real life incurs a penalty. The affected scale multiplies with that penalty. When dealing with things affecting humanity as a whole, that makes careless over-generalizations a serious hazard.
First movers often lose though in the market. Relevant to today's news about DoorDash acquiring Instacart for a few billion dollars: WebVan came out a decade earlier (i.e. at the wrong time) and went bankrupt after 3 years.
A perhaps more relevant personal anecdote: About a year ago I told a non-technical friend that I'd help him migrate his small business website that was originally set up in WordPress (by a developer who is since long gone) to a format that would be much easier for him to manage with no dev support. I planned to move him over to one of the usual suspects like Wix or Squarespace, but this is one instance where my procrastination was a virtue. In the ~9 months since he originally had an issue with his site, AI capabilities had exploded. Turned out it was a lot easier (and cheaper) to use Claude to fully migrate his site and get it up on Cloudflare workers, and I set up everything so he could manage all the content himself. Plus, it took me less than a day to do the migration off WordPress, and most of that time was just me wanting to learn the ins and outs of Claude Code because I was new to it at the time.
“if you are researching anything that can be applied to the industry, you will instead have 1 year of market ownership and first-mover advantage.”
I would like to disagree with you on that point. Given the very popular rise of https://rea.tools/, it appears the time to reverse engineer anything software based (up to maybe 80% or more?) could be down to minutes. With a recent example, some indie video game developers are not going to show off their video game until release as a few have been flipped and made it to market before the original developers have. Point being, a few moths of market ownership could be possible if llms can’t reverse it in any way.
Someone is going to make the case against using LLMs to reverse engineer software solely for commercial gain being as illegal as using guns for extortion.
Unless you’re able to get a patent on day one, once an idea is public potentially within the hour someone can make more money off of it and it’s not just your idea. Seems like a valid case to lock it down.
What a waste. All the anticipation that used to be built up with previews and looks into the dev process, now flushed down the toilet so billionaires can flood the web with crap and make more billions.
It might also mean we can take more time, or prep for it more, or think of the implications. The deflation can be good, it also means the future could be way better off. It could be that civilizations destroy themselves through multiple conflicting levels of power (political, genetic, energy, material consumption, mass media, malaise).
I like this observation very much. Intellectual deflationary spiral is exactly how it feels at the moment and we might end up in a worse place because of it.
reply