Nvidia invests in suppliers like photonics companies and Corning. These companies need the cash to scale up while Nvidia likely gets majority of the supply and some equity.
They invested in Anthropic and OpenAI in order to keep them invested in Nvidia's ecosystem. If Google and Amazon were throwing deals at OpenAI and Anthropic to get them to build on their hardware, Nvidia had to counter.
Lastly, Nvidia invests in Neoclouds because big tech is building their own chips to compete against Nvidia. Why shouldn't Nvidia invest in Neoclouds and ensure that they don't get crushed by hyperscalers? They tried to compete directly with hyperscalers with DGX Cloud in 2023 but decided to not compete against their customers by 2025. Instead, they're backing Neoclouds. Makes more sense.
They have the cash. Instead of returning the cash to shareholders in buybacks and dividends, it's smart that they're putting it to use strategically.
If they just returned all the cash to shareholders, 5 years from now, people might complain why didn't Nvidia strengthen their grip on the market with their cash?
I think this. China is still in the optimist mode - life is still getting better, tech is here to improve daily life. In the US, everyone is skeptical of everything. Trust is low.
Yeah there's definitely something to this. Do you think it's mainly a trust issue? Is it also hard to imagine potential benefits in areas like Health/Education vs quite easy to imagine risks such as jobs displacement? Do you think people would be more focussed on potential benefits if the trust was there?
I would not invest in SaaS companies right now that are not naturally able to switch to serving agents.
I think the seat-based SaaS company is in huge trouble in terms of public market valuation. They will not command 80+ P/E anymore. In fact, I expect them to be on the lower end of P/E industries, something below 20.
The article is right that there are only 2 paths but I'll name the two paths in simpler terms:
1. You have great ideas for growth and this newfound productivity allows you to achieve those ideas. You don't have to do layoffs.
2. You layoff most of your company, put the company on a lower-growth path but your margins increase substantially due to AI productivity.
Doesn't Amazon engineering culture have a very engineer-led product culture? Meaning, devs are often responsible for the UX and flow.
I remember many years ago we hired a junior developer who just finished his internship at AWS and he showed me the dashboard he shipped all by himself in the summer with no product or designer help. It looked horrible.
Some devs have a good product/UX sense but the vast majority are horrendously bad at UX.
My point is that maybe it was intentional, but just bad UX culture.
Some background. I work at an Amazon sub. This is a good UI for the way we work. We don't spin up a single machine pretty much ever unless it's a cloud dev machine, at which point the price is listed at startup on a custom internal UI. They should consider putting that UI in the ec2 console.
When I spin up machines I pick an instance class by looking through specs and the price chart and set it via AI into a cdk construct. Usually pick a relatively normal machine type digging through all the ilvarious enterprise discounts (which are not reflectedin the prices in the console). Then as I roll out or when I get resource limit alarms on the fleet I adjust the instance types. Or when accounting asks me about price. In those cases I usually look if it's worth it to optimize.
The enterprise discounts are a big consideration. Every year new hires make bad decisions because they don't know about the discounts. They wildly affect total cost. Some things are more expensive (lambda first few years), and others are very cheap so we dog food. The console price in no way reflects reality.
In 15 years we've had about 1k services stood up, around 700 are active. 2000 or total counting tutorials and tests. That means out of an eng org of 500, we've made those decisions maybe 10k times total.
That's how Amazon thinks about it as well. So yeah I agree that the UI isn't meant to be like one where your spinning up a host. I haven't spun up a single host in like 5 years, but I've made many clusters.
But that doesn't mean it shouldn't be better to work for a wider audience. Customer obsession and all
AWS UX isn't bad because engineers are bad at UX. It's because inside AWS it's every man for himself, and every team for itself. They don't collaborate, they don't talk, they compete to ship everything as quickly and cheaply as possible - quality, usability, and common sense be damned.
FWIW, my team in AWS had help from UI designers who were cool people that impressed me with their work. We definitely had to push through some needless organizational friction, e.g., they were in a different org and frequently got left out of meetings, whereas we should really have been acting as one team. I don't think we saw it as everyone for themselves, we really tried to make it work and had a good, trusting relationship.
In the end, our leadership changed what we were building so often that all of the UI work was scrapped long before we shipped. We ended up launching a janky console, quickly assembled by SDEs who were racing against deadlines. We skipped virtually all operational readiness work to meet the launch deadline. After claiming the launch win, the director, two managers, and the pm promptly left for other orgs.
That's not just AWS. That's Amazon generally. All Amazon orgs I've worked for have been like this, and due to the nature of my work, I (and my teams) have been treated like pariahs for daring to suggest that there ought to be even a minimal amount collaboration, shared standards, and cross-pollination on ideas between teams.
> My point is that maybe it was intentional, but just bad UX culture.
This may be valid, but even if it is someone (or a group of people) at Amazon are violating one of their core leadership principles - Customer Obsession
A useful (and hopefully delightful) UX is key to showing customer obsession.
That being said, I personally feel the UX at Amazon sucks overall, not just for pricing/packaging but even getting basic shit done. So perhaps Amazon (or at least AWS) doesn't think a good UX is a key ingredient to demonstrating Customer Obsession.
Everywhere on Amazon.com has bad UI/UX. For one example, the flow on checking out as a non-Prime member (not sure about Prime members) is janky and feels straight out of 2005. Like it reloads the page, taking ten+ seconds, every time you enter new data (address, credit card, personal info, etc.). I would be laughed out of the room if I tried to deliver this at work, but Amazon delivers it for millions of people.
So no, they care zero about their customers, except maybe for getting as much money as possible out of it.
Personally I think the UI flow is geared towards the idea that engineers don't really see the costs, they just build stuff and then management pays at the end of the month.
Often I see something that's supposed to be leaner - like Fargate is leaner than renting a whole server to run docker, right?
So it's cheaper as well? - Well, no.
Also if you reach any appreciable level of complexity, you should move to IaC - configuring all that stuff on the UI, and getting it right is torture.
They're not but if they don't talk to the pricing team, and most devs don't want to talk to business people, they'd never coordinate on where it makes sense to show pricing to customers.
You didn’t read the comment I replied to, did you? The premise was :
> the UI flow is geared towards the idea that engineers don't really see the costs, they just build stuff and then management pays at the end of the month.
So this is about the engineers consuming AWS, not the ones who designed and implemented AWS
Right - nobody who’s had a formal education in engineering would think that way, because cost considerations are part of the curriculum from the start.
I don't think a lot of formal education places teach AWS's resource pricing structure, which can be incredibly confusing, but can be boiled down to: if you want to be as cheap as possible, just use EC2 for everything and maybe S3 for storage.
I'm very surprised you expect any formal education to teach any specific pricing structure. You teach how to evaluate solutions for their price impact. No one was claiming any curriculum includes AWS's resource pricing structure.
I can't recall cost ever coming up as a consideration during my years of formal computer science studies in school. Big-O efficiency, sure, but the cost of compute, storage, bandwidth, nope, not once.
It was absolutely hammered into me in the years of working for startups that followed, though.
I would argue that the intentions don't matter at all, the end result is all that matters both for the buyer and seller. In systems design, it is often said that The Purpose of a System Is What It Does. Good intentions can produce very bad systems with bad outcomes, and neutral/bad intentions can create good systems that benefit everyone.
I think that applies both to Amazon's dev system and pricing system. From what I hear about the insides, alignment is chaotic neutral inside of Amazon, but that shouldn't affect how we judge the system itself.
> Some devs have a good product/UX sense but the vast majority are horrendously bad at UX.
I think the problem is that nobody understands the size of the problem.
For most tasks, the accomplishment is getting something to work. That takes 90% of the time. But the UI requires polish, working things out, backing out and trying again, and takes the OTHER 90% of the time.
I remember talking to a friend who worked with apple to port some dvd authoring software. And steve jobs started with the UI, and said "this is what you do". I think it was just a blank screen and you drag your video onto it. the software they were porting was a bunch of windows type confusing nonsense, and they had big changes to make.
That said, AWS might be a dark pattern. Remember the cable companies that didn't WANT to show the hidden fees? because $29.99 a month was really $71.41?
Prior to AWS at Amazon hosts were provisioned as “host classes” and typically operated on in that way. We were encouraged to make them “touchless”, which meant the infrastructure team could replace that host without contacting the team first. The deployment tool deployed to host classes (though you could put an individual server there if you wanted). EC2 wasn’t quite the same, but not very foreign either. We didn’t originally even use the AWS interface (at the team level). They were managed by a team working on the transition.
It just goes to show if you are the first big player in a space, you can have whatever UX. No UX will override that first mover advantage. I could cite countless examples, where the first commonly known company rules the space and no newcomer with a flashy UX can come close, no matter how hard they try.
Honestly UX became irrelevant in the last years (infra as code) and even more in the last year (coding agents). What you need is well structured API and a CLI that does not limit you. You can call it UX if you want, but the skillset is different.
When I started my latest project my first rule was: I never have to login to AWS console. I didn’t achieve ‘never’ but I am pretty close and the experience is a lot better
Translating a project that includes a good test suite from one language to another is known to be a great case where LLMs work well.
When you’re starting with a complete codebase to use as an example and a test suite to check everything it’s much easier to iterate toward the desired goal. The LLM can already see what the goals are and how they’ve been implemented once already, which is a much easier problem than starting from a spec.
I fail to think of a successful Rust rewrite, so far what I've seen is just programmers who aren't sufficiently experienced, who decide to pick Rust and rewrite something in it, and then (this is the bad part) claim it's better for that reason only. It never is. It's always worse, because rewrites fundamentally end up with a worse product first.
Sure, but, given that, does it not seem like the conclusion is: if you have something that could in principle be reverse engineered by a competitor with more compute, they can and will steal it, because the only constraint is roi.
Unclear. Very good products tend to be about doing one or a few things very well; not about doing tons of stuff. So far, all I see is “Man, Im a 10x engineer now!”, shipping more code but without clear direction and taste. At this point, most of LLM-based work is just noise.
You could have said the same thing about steam power or electricity. And it’s not just an analogy: The magic of these things is in being universal information engines. You spend capital to build them, using well-understood, scalable techniques, plug them into electricity, and out comes value.
My point is, there’s no chance of a “haves and have nots” emerging, any more than electricity turned out that way in the modern world.
Electricity might be a good analogy - but for the other side of this argument.
In the US, (nearly) full electrification wasn't achieved until the late 1940's/early 1950's - a process of nearly a century. (A moment of personal trivia, my great grandfather worked on crews electrifying rural areas of the midwest.)
We already have SOTA local inference devices in everyone’s pocket, which also provide high bandwidth access to SOTA data center inference at what is rapidly becoming commodity pricing.
Nah. These agents are getting easier and easier to run local. Have you tried Qwen 3.6 27b? It’s insane what it can do compared to its size. Like 100% vibe small projects if you manage context properly.
These models are a race to the bottom just like compute.
My guess is it won’t be worth it to focus specifically on coding models once local small models work just as well or within range. That will naturally close the gap even more
It costs several times what it would cost a small team of engineers, even assuming you gave the engineers more time to do it. I'm guessing (wildly) this was around 0.5M USD in compute time. You do get the result quicker, though.
> I'm guessing (wildly) this was around 0.5M USD in compute time.
That seems like an especially wild guess. If you take e.g. Opus 4.7 prices, and make the assumption that you are consuming roughly $30 for every million tokens of output (this comes from just summing the $25 per million tokens of output and $5 per million tokens of input and assuming that caching basically makes all that work out), and assume an output rate of 80 tokens per second (which seems like a high estimate based on online searching), it would take you about 2411 days of non-stop Opus 4.7 usage to hit 500k in API spend.
The only way you could possibly run that amount of usage in 6 days is if you were running ~400 instances in parallel. From personal experience, that seems crazy high for this project.
I think you are off by at least an order of magnitude (potentially even 2 depending on how the person is managing agents, but I could see something like dozens of agents 24/7, so I'm way less confident in 2, but I think it's still more likely to be closer to 10-20k in API spend).
From the leaked internal prompts, Opus 4.7 vs 4.6 recomputes several times over before returning the result. For heavy use like this, it costs Anthropic far more than you're paying as a consumer. They rely on the light users to offset the whales, and they're still at a significant net loss. If you tried this as an end-user, they might cut you off (though I understand their data centers are underutilized, so that wouldn't be for logistic reasons). Being part of the company and directly sanctioned, the author has unlimited access.
~7x overcompute * ~7x real cost to Anthropic * your 10-20k estimate for consumer use is my thought for actual total cost. If the honeymoon period runs out and they're still in business, this is what everyone will pay.
But usually companies are much more careful before even spending that half a million. (And most companies don't have that money sitting around.) They would do small PoCs, do comprehensive benchmarks and evaluations of those PoCs, and decide whether to actually go ahead, and, more importantly, stick to it.
Being able to afford half a million doesn't mean you do it on a whim, or just throw all of that away if things don't go well.
But what do I know. I am nothing compared to our AI overlords like Anthropic.
45 million lines would get to ~$1.125 mil for the linux kernel.
950k lines for Bun would get to $23,750
use whatever math you like ofc.
Does an Anthropic/employee pay that, no. Even if it's at a loss in terms of company revenue, it's worth burning the private capital for all kinds of other reasons.
The saving grace here is a rewrite of a project with a good test suite is the sweet spot: LLMs are great at translation and do great with verifiable goals.
I agree it’s still mind blowing compared to before times, though.
It probably wouldn't take a single person who knew what they were doing more than a year to re-implement Bun in basically anything, by hand and from scratch, i.e. not even looking at source. Writing the code for something you already understand and have built before is incredibly fast.
I'm sure they'll market what you said, but it's so ridiculous that I would hope people would see through this stuff.
The majority of Bun was written by one guy in less than a year. In what world would a rewrite take hundreds of engineers more than a year to do? The hyperbole is getting ridiculous.
In the sense that the incremental improvements in capabilities that we've been seeing in recent models seem to taking exponentially growing amounts of compute to achieve.
Compute doesn't necessarily linerarly follow parameters. And with how many active parameters Mythos vs Opus gets its effectivenes from? Is it 1x or 2x? We don't know. We don't even know the parameters (it's more of rumor than confirmed 10T iirc).
But even more so, who said the improvements are "exponential"? Mozilla's single metric, that doesn't even prove anything of the sort?
I know parameters don’t translate directly like that (and that linear and exponential aren’t the only types of growth) but a doubling as a go-to example of “not exponential growth” is pretty funny.
Ah yes, the marketing model that's ostensibly so powerful us mere mortals aren't allowed to use it. It's certainly led to exponential hype and speculation.
Thankfully their leadership is leading the way in ethics since inception, so I am confident that no such shenanigans will ever take place. I may even bet on this.
> These things are dangerous. Someone who can take AWS down such as an employee can place a bet.
Imagine if the betting website itself shuts down because AWS is down. (half joking I suppose though)
> These bets aren’t as innocent as they seem because the bettors can often influence or change the outcome.
Overall I agree with your statement that these betting markets also are able to incentivize a lot of insider trading and one can say negative scenarios as this has given them an incentive to capitalize on that.
Anyways, it doesn't matter if it's high NA or low NA when it comes to capacity. What matters is how many total EUV machines and fabs. As of right now, a Google search says TSMC has 2x the number of fabs in construction as Intel.
They invested in Anthropic and OpenAI in order to keep them invested in Nvidia's ecosystem. If Google and Amazon were throwing deals at OpenAI and Anthropic to get them to build on their hardware, Nvidia had to counter.
Lastly, Nvidia invests in Neoclouds because big tech is building their own chips to compete against Nvidia. Why shouldn't Nvidia invest in Neoclouds and ensure that they don't get crushed by hyperscalers? They tried to compete directly with hyperscalers with DGX Cloud in 2023 but decided to not compete against their customers by 2025. Instead, they're backing Neoclouds. Makes more sense.
They have the cash. Instead of returning the cash to shareholders in buybacks and dividends, it's smart that they're putting it to use strategically.
If they just returned all the cash to shareholders, 5 years from now, people might complain why didn't Nvidia strengthen their grip on the market with their cash?
reply