I really like this for CodeAct, but like with other similar tools it's unclear how to implement data pipelining to leverage, like, lockstep batching to remote providers, or paged attention-like optimisations. Basically, let's say I want to run agent for every row in the table, I would probably want to batch most calls...
It's something, I think, missing from smolagents ecosystem anyway!
wall street analysts are starting to realise that software companies shouldnt trade on a P/E of 300.
DocuSign is currently valued at 30 times its annual earnings. Adobe is currently 16. Amazon is 28 -- has been as high as 50 recently. NVDA is 44.
Investors are basically starting to realise that enterprise are not going to subscribe to software like DocuSign for 50 years. They'll probably just move to odoo or zohosign or something and save a lot of money. So its probably a better bet to put that capital into something like Nvidia or Tesla or whatever. it also looks like the US Fed isn't going to cut rates, so capital is getting more expensive.
Of course, if you are a CEO its great to blame all this on AI and then tell your investors you are increasing AI in your business (see: salesforce whose stock price is down 42% in a year and is now trading at 25x earnings)
> software companies shouldnt trade on a P/E of 300
You are playing pretty fast and loose with your definition of a "software company" when you include Amazon and NVIDIA in your list. Amazon is many things but it is not a "software company" and neither is "NVIDIA".
I actually don’t even think it’s a IaaS company. I mean it is, but Amazon seems to always excel at operations: whether it’s organizing a retail business, operating a logistics company, or managing a hyperscaler, it just seems like its real secret sauce is running ops.
(Which makes sense because all of their end user products suck)
AWS is not a software company either, it is a capital intensive computing infrastructure company. NVIDIA as well is a capital intensive computer hardware company.
Just because software plays a role, doesn't make them software companies. It is like saying all companies are "electrical companies" because they require electricity to operate.
Argh, we first ignored the fact that most of its income comes from being the lead e-commerce retailer to just focus on AWS and then we need to discount the fact that the majority of AWS is CapEx towards hardware / datacenter (expected CapEx this coming year is $200B), to just leave the software. Whatever.
One thing is for sure: if the seal builds and manages hundreds of billions of dollars worth of computer infrastructure across seven continents, it isn’t just a software company.
My understanding on this may be spotty (and I appreciate it if someone corrects me), but CUDA is not the software layer that allows you to use NVIDIA GPUs for AI processing?
AMD may develop their own software layer, but a lot of things already work on CUDA, and the job to port this to a different platform may be non-trivial (or even possible depending on the level of feature parity).
CUDA is a GPGPU software layer that is very mature, and integrates with C,C++, Python, Fortran very well. AMD just never really got the same quality of GPGPU software in the last 20 years. 99% of scientific computing that uses GPUs (which is a lot since they are so much faster than CPUs for linear algebra) have gone to Nvidia because of this. All of the big AI libraries (Tensor Flow, PyTorch) basically ended up writing around CUDA, so they just didn't write things for AMD. Plus if you go and look at a job for signal processing or whatever at say Lockheed Martin or Raytheon, they specific CUDA.
> I don’t suppose you know a good “for dummies” explanation of why CUDA is such an insurmountable moat for them?
Theoretically the moat isn’t insurmountable and AMD has made some inroads thanks to the open source community but in practice a generic CUDA layer requires a ton of R&D that AMD hasn’t been able to afford since the ATI acquisition. It’s been fighting for its existence for most of that time and just never had the money to invest in catching up to NVIDIA beyond the hardware. Even something as seemingly simple as porting the BLAS library to CUDA is a significant undertaking that has to validate numerical codes while dealing with floating point subtleties. The CPU versions of these libraries are so foundational and hard to get right that they’re still written in FORTRAN and haven’t changed much in decades. Everything built on top of those libraries then requires having customers who can help you test and profile real code in use. When people say that software isn’t a moat they’re talking about basic CRUD over a business domain where all it takes is a competent developer and someone with experience in the industry to replicate. CUDA is about as far from that as you can get in software without stepping on Mentor Graphics’ or Dassault’s toes.
There’s a second factor which is that hardware companies tend to have horrible software cultures, especially when silicon is the center of gravity. The hardware guys in leadership discount the value of software and that philosophy works itself down the hierarchy. In this respect NVIDIA is very much an outlier and it shows in CUDA. Their moat isn’t just the software but the organization that allowed it to flourish in a hardware company, which predates their success in AI (NVIDIA has worked with game developers for decades to optimize individual games).
Lots of other versions exist including reputable ones like Intel’s MKL. The hard part isn’t reimplementing it, it’s validating the output across a massive corpus of scientific work.
BLAS is an example though, it’s the tip of an iceberg.
the first problem is a whole generation of people learned to code ai applications by fiddling around with the gpu in their gaming pc 10 years ago. so an entire generation of talent grew up with cuda
the second problem is that so many libraries and existing software is cuda only. even some obscure hardware stuff. i discovered the hard way that some AMD thinkpads dont support thunderbolt transfer speeds on their usb-c ports, whereas nvidia ones do
the third problem is that the cost to develop a cuda equivalent is so great that its cheaper for companies like google to make TPU and amazon to make Trainium. its literally cheaper to make an entire new chipset than it is to fix AMd. i dont see companies like apple/amzn/goog etc fixing AMDs chips
i mean its cheaper from an enterprise customer perspective. if a company is training an LLM, writing their training programs to use AMDs hardware instead of just using CUDA is so expensive and time consuming that it is cheaper to pay four times the price and use nvidia hardware. in this space its important to move fast, although that economic will shift over time
which is why nvidia hardware trades at a 4x premium to AMD
its not necessarily cheaper to make chips at cost either. nobody is making them, only designing them. so first you have to design your new chip, then you have to get a minimum order in with the chip fab so big it competes on unit economics, and then finally you have to get your dev team to write a CUDA equivalent software that is a problem so hard its only really been solved by apple, google, intel, and nvidia
only companies with big fab orders can get priority too.. if a company did all of the above and was ready to go, they probably wouldn't get fab capacity until 2030
youre right, i should say tech company. but at least my flawed epistemology reveals my humanity
although one could argue disingenuously that nvda is a software company because the product they ultimately manufacture is a bunch of blueprints they email to tsmc or samsung who then actually make the chips
I've always found it confusing how run of the mill SaaS trades at multiples assuming decades of doing business. The amount of change in software businesses has been massive and being able to run a successful software business even for 15 years from 2010-2025 requires a great deal of strategy and foresight and more likely than not that's not enough. Considering how these dynamics have been accelerating as technology accelerates it just seemed so off that the market was landing on a 20-30x multiples for software businesses that don't have much moat (e.g. swathes of B2B CRUD apps).
Investor analyst looks at earnings growth and determines Customer Acquisition Cost (CAC) and Customer Acquisition Cost Payback Period (CACPP). They determine that ABC Software Corporation has no marginal manufacturing cost because it makes software that it sells online, so if it invested 90% of its profit margin into marketing it could grow its ARR by 140% a year. Then they extrapolate that for 30 years and say ok the NPV of 30 years of 140% ARR on current CAC, etc etc...
If everyone in the industry benchmarks on more or less the same multiples, it becomes a good idea to buy any b2b crud saas trading at 10x earnings because if the big boys see it they'll probably bid it up to 30x
the other classic move is to take a business which really isn't even a new technology, like revolut, and call it a tech business. now suddenly a bank can trade on a 50x earnings multiple instead of 15x like say a bank. many such cases~
we are only examining the valuation metrics here, not comparing the companies themselves as investments.
you could decide that if you are a very large company, building software internally to replace a SaaS product is a path forward. Or replacing a premium software like DocuSign with a cheap one like Zoho sign. or just building your own proprietary electronic signature app
It is however impractical for big company to start manufacturing cars or designing competitive GPUs
so the earnings of tesla and nvidia is theoretically more 'stable' than a software application company like salesforce, adobe, etc.
this analysis ignores both the size of the company, and what it does, or whether or not any one of them is a good investment
Every tech company assumed they would be the benefactors, not victims, of AI. And investors now see that without the alleged AI growth, these companies at best look like stable utilities, not high growth stocks. At worse companies look like they make highly replaceable software as software stops being a moat.
Moreover they look like large, inefficient organizations with a lot of human veto points that prevent innovation (requiring more human coordination is an anti moat now)
Was software ever a moat? Software typically only gave companies a small window of opportunity to turn a fleeting software advantage into a more resilient moat (network effects, switching costs etc.)
Yes, I would argue good (stable, fast, easy to use) software was somewhat of a moat and much harder before coding agents.
Stripe, Square, Shopify, Google, all thrived in some part because their services take a hard problem and make it easier to use. Now more people can take a hard problem and make it easier to use.
All you have to do is look around (esp 5+ years ago) and see the many many BAD, unstable, hard to use, slow, etc versions of these companies
Windows' moat was not the operating system code, but that they were able to get distribution via IBM, and then grow an ecosystem of applications that were targeted at Windows, which created a snowball effect for further applications.
I have read that numerous places and it seems plausible but it is beyond my investing experience.
I think the new nominated Fed Chair is also a hard money advocate and is spooking USD alternatives (gold, silver, BTC, etc.) But hard money can be quite hard on the economy, so that could limit growth.
Investment is all about belief. When the root cause is the willies, then we hallucinate reasons together.
Crypto and memes have demonstrated us a lot about the drives of individual investors.
Unfortunately it seems that professional investors are not that much more rational (from my limited personal experience with a small hedge fund, and from my years of looking at markets).
My favourite term has always been "taking profits" which is generated by the technical analysis (I loath that term) of looking at the prices: taking an effect and publishing a cause (trying to sound smart).
We are deeply irrational beings; often the more you go up a professional ladder the more rationalisation you see.
During unstable periods, we see lots of weird side-effects and there is a lot that doesn't seem to make sense.
Edit: a better meme could be "The use of AI by funds is destabilising markets"
Disclaimer: I am not a professional investor. I am a cynic.
Speculation involves belief, sure. However, there is hard math involved in markets as well, and the yen carry trade equation exists whether you believe in it or not.
Not saying you're wrong, necessarily, but as I type this the Dow is at 49,700. Would you expect the collapse of the yen carry trade to cause the Dow to collapse as well? Or is the Dow not high-growth enough for people to put yen-carry money into it?
> Would you expect the collapse of the yen carry trade to cause the Dow to collapse as well?
USD devalued ~10% last year, so some of the losses are already priced into the DJI. When you account for that and sprinkle in ~3% inflation, it has lost value despite being up ~11% in the last year.
Anecdotally, read somewhere that delivery people are going idle. The Orange One wrecked everything with tariffs, and the ripple of destruction is slowly taking its course. That's before we take into account massive outflows of cheap labor and fired government workers.
In addition to the security issues, they would have to deal with non-negligible transaction costs every time they wanted to convert it to actual money so that they could purchase something. If they were using it as an investment, they had to deal with the opportunity cost of underperforming $SPY.
It really comes down to the burglar's expectations. If most crypto holders used geographically separated multi-sig, these attacks wouldn't be worth the effort anymore.
It’s the same logic as iPhones bricking themselves after being stolen. Even if your specific phone isn't an iPhone, the fact that most phones are now useless to thieves discourages the crime across the board.
Our knowledge is constantly expanding, allowing us to build things differently than we used to. Modern cryptography, which makes things like multi-sig possible, is only a few decades old; it didn't even exist when the current banking industry was being established.
This isn’t just a problem in the Netherlands or a thing of the past. 2025 actually saw the highest number of attacks ever recorded [0].
There are ways to prevent this. Like using multi-sig with geographical separation (so you can't move funds alone) or setting up forced time-delays. Ultimately, being your own bank is a massive responsibility, and I think too many people take that reality too lightly.
Antigravity is not a windsurf reskin, at least not today; it introduces many concepts and optimisations that you wouldn't find anywhere else, and in my workflows Gemini 3 Flash in Antigravity also happens to outmatch Claude Code with Opus 4.5 on some really gruesomely complicated tasks (i.e. involving compiler/decompiler work.)
The Gemini family seems to be better at high-level, generalizing
The Claude family seems to be better at localized coding tasks
I'm a big fan of Claude Code based prompts with Gemini 3 Flash in my coding agent. I'm unwilling to use any new Google products at this point in time. Used to be a stan, they have pushed me away and I'll never be a stan again
yeah looks like someone is either a hyper tailscale fan or had extremely bad experience with it, I also run several dozens of machines (and tablets and phones) on it. never had a single moment of downtime since I started.
Yeah, Headscale people don't hide that it's a toy. I didn't get a homelab full of datacentre-grade equipment because I want to use toy, nonscaling solutions with vastly incomplete feature sets, but for the exact opposite reason.
On a different note; the HN obsession with SQLite these days is getting a bit tiresome.
There is night and day difference in generation quality between even something like 8-bit and "heavily quantized" versions. Why not quantize to 1-bit anyway? Would that qualify as "running the model?" Food for thought. Don't get me wrong: there's plenty of stuff you can actually run on 96 GB Mac studio (let alone on 128/256 GB ones) but 1T-class models are not in that category, unfortunately. Unless you put four of them in a rack or something.
True, although the Mac Studio M3 Ultra does go up to 512GB (@ ~$10K) so models of this size are not too far out of reach (although I've no idea how useful Kimi K2.5 is compared to SOTA).
Kimi K2.5 is a MOE model with 384 "experts" and an active parameter count of only 32GB, although that doesn't really help reduce RAM requirements since you'd be swapping out that 32GB on every token. I wonder if it would be viable to come up with an MOE variant where consecutive sequences of tokens got routed to individual experts, which would change the memory thrashing from per-token to per-token-sequence, perhaps making it tolerable ?
Well to be fair Hackernews posts can get flagged too by the community itself where people then later talk about how or why a particular post gets flagged and discussion starts moving about the moderation/flag issues in HN.
(But this isn't to say that the fault's within the moderation community of HN which are great but just the issue which to me is imo that if many users flag a post, it can get flagged and the friction of getting it back is hard or a post typically ends up dying usually if it gets flagged in general imho)
It's something, I think, missing from smolagents ecosystem anyway!
reply