noodlesUK

I want so badly for starlink to offer a sensible well priced self-service B2B offering with fleet management. Any kind of fixed, high bandwidth IoT device that currently has a SIM card could easily use a starlink. As it stands SMBs who want this have to screw around with residential starlink and live in fear that they’ll get banned.

One thing that I'm not seeing mentioned very frequently is that it's really hard to actually get the GPUs that hyperscalers advertise depending on your region, use of cloud credits etc.

On Azure if you have credits (even hundreds of thousands of USD of credits through programmes like MSFT for startups) you just can't get any GPU quota in any region, even if you beg and plead with support reps. Even if you have cash it's often difficult to get any of the cheaper cards. You can forget about spot instances etc because the regions are essentially all at 100% capacity.

I've heard on other hyperscalers that it's also challenging.

This is super chilling for anything other than GenAI inference/training because if you have some more traditional workload that requires GPUs (traditional ML, other kinds of modelling) you're SOL.

I think one of the things that people analysing the finances of the AI sector often miss is that the capital expenses associated with data centre buildout involve spending a lot of money on compute hardware which depreciates and becomes obsolete very fast.

Large companies have often changed the depreciation schedule for server hardware to 6 years rather than 3. I think it remains to be seen if this is justified as it’s not as though the pace of change in hardware has slowed down by 50% arbitrarily.

I think it’s pretty obvious to me that a data centre full of ampere GPUs is worth less than a data centre full of Blackwell GPUs. I suspect this is also why AI companies prefer to talk in terms of gigawatts of compute, as if efficiency per watt does not exist.

Data centre costs are not one-off. Some are, for buying land and building buildings and infrastructure. The actual compute however is a constant investment. I’m not familiar enough with the market to know what proportion of expenses are ongoing but I bet it’s higher than people think.

The company behind Talos Linux has been acquired by Yardi (of WeWork fame).

I use Talos for my homelab and have been considering using it for work. I'm a little worried about the future direction. I'm not sure why they want to add VMs as a native resource rather than going in the direction of KubeVirt as a first class citizen (which already worked well).

I have no business other than as a customer but we’ve had decent success with getting cheap archive imagery at fairly high res out of SkyWatch. It’s worth a look. If you need a very large area the price will grow quite a lot though.

I really hope that in the coming years we will be able to get smaller phones back into the market. Apple compared this to the size of a passport. A passport seems even bigger than a normal iPhone, even when folded. A number of people (especially women with smaller hands) I know are holding onto their iPhone Minis for as long as possible. I’d love a phone that was roughly normal iphone sized when unfolded but was a flip phone à la Motorola razrs in years gone by. I wouldn’t even need a screen on the outside. I know it’s hard for a company like Apple to be able to market “less phone” as a good thing, and I know the iPhone mini didn’t do great commercially but there’s a small cohort of users clamoring to have something for them.

I look forward to DuckDB being the engine that underpins the next gen of analytical data tooling. DuckLake already looks amazing and with the Quack protocol seems like it will be a great natural fit for lots more types of tooling, such as sensor data etc.

I think what has surprised me is just how quickly the pendulum has swung. From becoming useful in early 2025 to cracking down on costs has been maybe 18m in total.

This is literally the purpose of most of the European EUDI wallets, but automated. You create an account with whatever part of government handles ID, you get a "wallet", and then services can ask you to share minimised data from the wallet. This can be implemented in such a way that the government doesn't know who requested and the requester doesn't learn anything about you other than the minimum covering set of attributes required.

I think that if we are considering technical solutions to the social problems here, the answer is well understood by professionals working in age/ID verification. You allow someone to perform a more thorough check verifying your age and identity (such as the government or your smartphone vendor), and then you use a privacy preserving protocol to provide a proof of whatever attribute (such as age) you are trying to verify in the absence of the remaining personal data. This can be as simple as exposing a "is age over {16, 18, 19, 21, 25}" api in the OS or browser, or as complex as a ZKP as needed. What doesn't work is requiring each service to independently perform verification which leads to a massive expansion in the proliferation of personal data and is incompatible with the modern approaches to data protection.

This is a solution that hopefully will cease to be relevant as soon as the remaining infrastructure is finished.

I have but it seems to me that the stuff that's actually particularly well priced is really old (2016-2018ish). Even a single DIMM of used 32G DDR4 from ETB is now > 250GBP. The used market is also blown to pieces because of the price pressure.

> If you serve large files, a CDN may have a very narrow use case where the budgets make sense. If you are already pushing 23 TiB/month, than cloud providers are usually not work the effort. Some rent colocation rack space. =3

Unfortunately I'm needing to run a lot of batch compute jobs (for which the hyperscalers are just insanely expensive - even to have a machine that outclasses a nice laptop becomes silly very rapidly)

I'm considering buying some machines and racking them in a colo but it feels like buying right now is also insane because of current pricing.

Yes but paying massively inflated prices for compute are not anymore. Even for consumer gear I am being quoted prices that are more than 2x what they were only a year ago (and availability is hard).

This is such disappointing news. I was planning on migrating some of our workloads to hetzner specifically to take advantage of the AX162 pricing which was incredibly competitive.

Does anyone else have any suggestions for competitive pricing for this kind of thing (e.g. batch jobs)? Was this applied retrospectively to existing customers?

I would love to see regulation that required making bootloaders unlockable to enable this sort of thing. People have been making clusters of consumer hardware for decades: I’m sure people remember the PS3 supercomputers of the mid 2000s.

I personally have lots of batch jobs like CFD simulations that could easily run on a fleet of phones with no real reliability issues, and I’d love to reuse old hardware and give it a second life. I’m already considering running old servers from e.g ETB but the cycles per watt on a phone are probably much better.

To what extent is this possible for actual metal hardware? I'm sure lots of us are running PXE/TFTP systems and HTTP would be a heck of a lot simpler.

I think that SUSE and RH can definitely work well in a fairly secure setting as needed. I certainly don't think it's any less secure than your typical corporate windows setup.

It definitely isn't a good look but I'm not entirely sure where this lands on "the line drawing on my IKEA instruction manual doesn't look like the furniture" to "VW diesel emissions report" spectrum. I'd appreciate if any bioscientists in the audience could clear that up a bit.

Exactly what is the "data" that's being shown here? Is it essentially some kind of marketing material showing "this sort of thing is what you should expect to see" or is it actually data or for compliance? If it's essentially marketing material or an instructional example that isn't meant to be representative it being magically clearer than real life doesn't seem like a great sin (unless it's being claimed it is representative). If it's something to be relied upon for compliance or as data to be used, that's pretty damming.

I don’t think it needs to be the FTSE and I don’t think it even needs to be all that much in UK funds but if you look at the default allocations of many large pensions it’s single digit UK equity exposure which I think isn’t really acceptable.

I think European countries need to get serious about investing money locally. The UK is a particularly egregious example but it’s taking begging and pleading with the mansion house accord to even convince pensions to try to invest in the UK economy. Every country should make a portion of local (at least within Europe) investment a prerequisite for whatever favourable tax treatment pensions and similar products get.

So much wealth is tied up in pensions and it’s folly to let it all go to supporting the U.S. and eschew local investment altogether.

From initial application to permanent residency it costs around £6k in visa fees and around £5k in health surcharges over the course of 5 years. Citizenship is another ~£2k all in.

You make three applications with those fees divided slightly differently each time.

That’s without any legal fees if you need someone to help you prepare your application which will be ~£2k per application.

If you have non-British children or stepchildren (which is generally quite rare) it’s approx those fees per kid as well.

For most other European countries including Ireland the fee is <€500.

The asylum system and immigration system are surprisingly disconnected from each other in the UK.

Pretty much all forms of permission to stay in the UK other than asylum can only be granted from within the country if you hold an existing long term status. So if you're visiting as a tourist you can't then decide to apply for a spouse visa or even a working holiday or student visa without leaving the country first. If you're already on a student visa or a work visa or similar you can change categories without having to leave.

The graduate visa is essentially an extension to the student visa with slightly different permissions - it makes sense that you can only apply to extend if you're in country and you view it from that lens.

The historic reason behind all this is that there used to be a substantial difference between being granted "leave to enter" and "leave to remain" (out of country vs in country applications). Leave to enter used to be granted by embassies etc and the foreign office, but leave to remain was granted by the home office. Now the home office handles everything in the UK centrally so the distinction is not significant.

Many other countries including UK enforce a similar rule. It's very inconvenient in those countries, but there's a significant difference: in most other countries that have this kind of policy, visas can typically be processed in a timely fashion (and are actually processed at all). It's insanely expensive and very arduous administratively to get a visa for the UK as the spouse of a British citizen, but the process will typically only take a month or so.

I'm really looking forward to the temporal api being universally available. Moment and Luxon are fairly good but sensible date/time handling is something that really ought to be baked into the platform ootb.

I think all of this article is true, but I suspect that the vast bulk of people entering retirement each year have few if any of these things beyond perhaps the sense of identity.

It's very true that retirement means losing many of the fun perks associated with a high-flying career, but you have to have those perks in the first place.

Even here on HN, many of us are living fairly ordinary lives, perhaps working on cool technology, but I think even in our fairly exclusive club, not all that many of us have PAs and frequent high-class work travel, even at the apex of our careers.

For those who have all those things and lose them it can be very profound. For a lot of people in that position they spend almost all of their time working in some sense or other, and retirement ends all that. I think that's why people in super successful careers often end up doing advisory work or ending up on boards, so they can keep some of the perks with less of the commitment.

I'm shocked that someone would write a blog post like this in which they openly admit to something that is widely understood to be fraud. Even if I'm sympathetic to why this individual chose to do this, and the technical side is interesting, I think the decision to just publicly tell a story in which you criminally defraud the villain is not a choice I'd ever make.

IMO this is not at all sensible. Hiring does not require an AI agent on the candidate's side engaging with an AI agent on the prospective employer's side.

What needs to happen is a sensible evaluation of the candidate's and employer's mutual suitability. We've had systems to do this in various ways for decades/centuries. What AI has enabled is a kind of shotgun approach to hiring that benefits nobody. The solution to this issue is more/better human interaction, not "agent-to-agent" communication.

I think many of these issues are also solved by Podman and systemd depending on what kind of "production" you're building for. If you're building a linux-y appliance and you need to run a few containers I think Podman is a much better and more ergonomic way of doing so. I think perhaps that's less true for running a web service (where the linux environment is just a means to that end).

I think it's hard to overstate how annoying working in a hardware-adjacent part of the industry has been in the past few years. We've only just recovered from the covid era supply chain issues, and it looks like we will continue to have shocks, after shocks, after shocks.

I recently bought a sample for a NUC-style board that contains an N355 processor, ram, and disk in a little kit. In the couple of weeks between when I put the PO in for the sample and the time it was delivered it had gone up in price from ~£600 to ~£1000. That might be a huge difference in BOM costs for a given product based on that design.

Real-terms prices for embedded computers must now be 2-4x what they were in 2020 for "similar" (i3, i5, i7, though I know there have been substantial improvements since then) positions in the market due to inflation and price gouging.

It's terribly difficult to keep up. This impacts far more than just AI. IoT devices, scientific instruments, kiosk systems etc have all skyrocketed in BOM costs because of this. Almost everything contains RAM, and a ~2x increase in an already expensive component might scupper a whole design.