If you want a government job, you need to be member of the party. That is no different then the old USSR system.

You end up with a overcount of "members" as people who ideology subscribe to the system, vs people who just become members for the better jobs. And yes, those people also get fired or leave government jobs (especially given the financial issues in the local governments the last few years because of the building crisis / lack of income).

So this type of throw towel over everybody to make a point is just a useless statistic.

Its not just the US ... Where do you think all those 5090's end up? Its just crypto mining all over again. Most people do not realize that China was the biggest crypto mining operation because of its cheap electricity and people willingness to take big risks.

The same happens now also, with people buying up 5090's (and lower class cards) in bulk, to strap into AI rigs. With Nvidia discontinuing the 32GB 5090s, we already see pressure on the lower end cards that had a better price / performance for a while now.

China will do China things *lol* ... People there are just more willing to take bigger gambles (and the cheap elec helps a ton). But that is also a factor why a lot of consumer level hardware gets under pressure. Its not just the supply chain to the enterprise market that is eating a lot for AI.

Consumer get the cheaper energy, while industry normally get the twice as expensive energy.

But for AI/Datacenters, there is a exception and they get the cheaper consumer level energy.

You need to take in account that the way China taxes is not the same as here in the West. And a lot more is put on the local governments regarding generating their own revenue (why there was so much building in China because those projects funded local government). A bit of a side topic lol. Its not so simple as they are making a loss on something because how its structured.

AI gets the cheaper tariffs then the rest of the industry, that is a fact. They simply have less energy profits on the AI energy. Now if you consider that subsidized or not, ... in both cases China is still making a profit.

What is overlooked is the massive amount of energy generation that China has and still keeps expanding. And that is a direct reflection upon why the prices for industry and consumers are so cheap. This buildout ironically had nothing to do with AI but was more about lower energy = cheaper industry production ... and a lot of buildout was also to build more independence from energy imports. They are not just throwing solar and wind turbines around like candy, the amount of reactors constructed and still under construction, ...

> China currently has 36 to 38 nuclear reactors under construction, maintaining the largest nuclear buildout in the world and accounting for roughly half of all global nuclear plant construction

So yea ... The AI in the west have a issue because it really does not matter if you cut China from Nvidia GPUs or advanced nodes, they compensate on the power level.

Edit: And a little extra, the irony is here in the West building nuclear is expensive as hell because we barely do one or two at the time. With a lot of lost expertise. China doing tons of them non-stop cuts down on the costs a lot.

* Models like DeepSeek V4.1 Flash are much cheaper on DeepSeek their API directly because of the cache handeling is better. Neuralwatt can hit up to 98% but DeepSeek can do 99.x... That may not sound like a big difference but it quickly widens the gap on long tasks to grow 2x a 3x in price. DeepSeek their cache handeling is S-tier (with a ton of features, for instance 24h caching).

* The same issue is also present if you compare GLM 5.3 Flash with z.ai vs Neuralwatt. Its just way more cheaper from the source, then from Neuralwatt.

* The energy numbers from Neuralwatt are ... to be taken with a ton of salt. Past energy numbers had the same models (for instance) GLM 5.2 up to 6x cheaper in energy usage, then after they "fixed" issues with the energy numbers. In reality, those energy numbers are just a different form of billing, but not a actual representation of the energy usage of AI models. Things like profits are inside those energy numbers. So seeing 4kWH used for a model, does not mean that it uses 4Kwh.

Edit: That are some interesting downvotes ...

To answer the questions. It was stated by the CEO himself in one of the video blogs that the energy prices inc their profit margins. Regarding their published numbers ... I like to point out that this is the same company that had up to 6x cheaper energy numbers at the start of the year until they got updated. Again, its in one of those video blogs the CEO did. Its around the same time when they increased the price from $5/1kwh to $10/1kwh.

Yes, DeepSeek API is cheaper then Neuralwatt. I have done way too many comparisons between NW and other providers, regarding their prices. Over long sessions, that gap grows because of the differences in caching. You need to use the NW Flex option to reduce the impact but then your constantly waiting on responses (good for overnight work, not great in prime time).

Edit 2: I am getting a little bit fed up with the people who downvote and do not give their reasons for the downvotes.

This screams to be that Sol vs Terra model problem that OpenAI had. On paper half the price, in actual usage the price gap was so close for less good results, that everybody just spammed Sol.

One of the issues mentioned is PFAS, ... ignoring the fact that we have tons of PFAS pollutions already in our soils and not a darn thing is done.

We live very close to one of those sites, but its always excuses after excuse to solve. But PFAS on labels, panic, panic, ... The whole label issue can easily be solved by random sampling at import and a fine system. That automatically puts backpressure to fix it. No need for certificates left and right.

My point is that we keep making bureaucratic rules, that make every step more difficult for entrepreneurs here in Europe, instead of having a proper compliance setup. There is a entire industry around yearly audits (that cost 10.000's), like people who do it non-stop suddenly forget how to do it.

That is the issue ... Its not about rules but about barriers that make often setup in such a way to be barriers of entry/working, then compliance.

Tell me, how many times do you go to have your car checked. Yearly or every two years? That is one of those example where in some countries its yearly, other twice. Do the countries with two years have issues with their car park. No ... The result is the same but some people get to be inconvenienced twice (and pay twice). Do you understand it a bit more now?

There are rules and there are rules that add little beyond making everything difficult, when it can be solved in a better way.

A new regulation when importing into Europe, is that you need to show that import materials do not have toxic substances, ... on ... shipping labels (meaning the physical labels are free from harmful chemicals like BPA, PFAS, or PVC).

You need from every company that ships from China, they need to provide certificates that their shipping labels are toxic free.

Pulled from AI because i hate summarizing this...

* EU Ecolabel: The gold standard in Europe. It officially guarantees that a product has low aquatic toxicity, a reduced environmental footprint, and strictly limited hazardous substances throughout its life cycle.

* EN 13432 Certification: If using compostable labels (like those from Better Packaging Co.), this mandatory European standard ensures the label, liner, and adhesive break down safely in industrial composting without leaving toxic chemical residues.

* FSC (Forest Stewardship Council): Certifies that the paper backing and label face stock are sourced from sustainably managed, non-destructive forests.

* Phenol-Free / BPA-Free Declarations: Thermal printing labels must formally declare compliance with EU Regulation 2016/2235, which legally restricts bisphenols in thermal paper across the European Economic Area.

Its a mess of paperwork ... We have not talked about a ton of others, and those are on the import side.

We had some fun discussions about vitamine pill manufacturing, where importing materials made it "made in China".

But if we imported the raw material, and the capsules are made in Europe, and its mixed / bottled here. Well, made in Europe. But the prime materials that goes into your body is from China ... Does not matter.

But wait, if the capsules are made in China and mixed / bottled here in Europe, then its still made in China. Because the capsule somehow made the difference?! I do not remember the details but the amount of easy customer cheating that can be done with "Made in EU / Germany / France"..

I can go on with some of the insanities i get told about ... There are good rules but a lot are like this mix that make no sense, and you can almost hear that somebody pushed for exceptions when the rule was designed. So you get this odd crap that half the time makes no sense.

Hard to do business when the rules are often a issue.

Here is a idea, why not combine both? Have the advantage of Rust its build in checks and LLMs independent checks. Now you get both for a even more safe program.

I simply mention what came to mind ;)

A quad 6000 with 96GB, can run this model at NVFP4. That is 60.000 Euro for the GPUs and lets be generous with another 20.000 for the rest of the system. The price of a single developer for a year.

Flash is about speed ... Flash models are supposed to be fast, way faster then their big brothers that are "better" but way slower.

Its just that up to now, getting more speed involved cutting back on the parameter count, what ended up making the Flash models more "dumber" in exchange for speed.

What we see with DS v4.1 Flash, is that DeepSeek has found a way to make a Flash model, that is 2x a 2.5x faster then the older Flash version, while increasing the intelligence (more parameters). To the point that it goes past Kimi K3 and GLM 5.3 in most tests, with a blazing 250 to 400t/s.

AND its also priced as a Flash model (they even reduced the price back to almost old v4 Flash price), despite it now rivaling those 10x to 30x more expensive competitors.

The issue that people can not fit it into local setups, is not how companies design their models. They design it for their own needs. A old flash needed less parameters to be fast, and local users had the benefit of it fitting in 256GB memory.

Companies who run locally, are perfectly able to buy a few H200/B200 and get a setup that run a model that almost rivals Opus 5.0 in their office. How to say this without getting downvoted. People get way too fired up if a model does not fit, despite that they can still run the old v4.0, qwen 27b, 35b, 3.8 Next and other models. The fact that these models are being released for free, is already amazing by itself. I am still waiting to see what Anthropic and OpenAI and Google are releasing for free... O wait ... ;0

The problem is that your too focused on the CPU only. There was GPU improvements (2d) then 3D, the constant improvements in sound cards. Printers ...

There was very strong depreciation in that time. I think that the poster before with his $10.000 > 1000 is somewhat exaggerating but like i said, my $5000 range system was not worth reselling. So did several other systems we acquired at the time, it simply became hand-me downs (for family who needed a PC for Lotus or WP but not a "advanced" PC).

Maybe my memory is off, but hardware in that time was way more expensive, then it is today. While todays hardware lasts WAY longer. The wife is still on a 8250u/8GB laptop, from like 8 years or so, and i can not pry it out of her hands. Needed a new battery lol ...

There really is rarely pressure to upgrade these days. That is not how i remember the 90's... Where upgrades tended to have much bigger impacts. Especially as gamers. But even on side equipment like CD players, CDRs and later DVDs.

Todays PCs is like ... raytracing, no interest. 4k? 1440p is perfectly fine. CPUs? Over powerful for most folks. The only new thing is AI and that is a totally different issue.

The problem is also that at a lot of code bases are not designed around LLMs and their token usage.

If you have a monolete codebase, you need to clearly define in the prompt what modules are involved. And even then, your wasting tokens with the first 1 or 2 steps where it needs to located the modules.

if you have a github codebase, with a lot of your code into nice little repos, its even worse because then the model needs to pull data, and a ton of more steps.

Most people do not open their coding agents in the module directories because "it may need something out of it".

I mean, we used to program by creating utils directories to deal with repetitive code but models (a) find it and use it (but it cost steps and tokens to read), (b) do not realize it there and make their own version of whatever or (c) combination of both.

And ironically, i feel like we need to give up on this idea of reusable code, and literally keep things into single modules, with as minimum external dependencies. That in return reduces searches and thus steps/tokens burn. But very few agents / harnesses have proper implementation of groups/projects and sub-module structures. Aka they only open a single dir, so your then forced to create dozens of tabs > per dir > cli ...

2. That is also a issue if you open multiple agents. Maybe now your working in A, B but C, D, E are not doing anything. And their cache expires... Now you go back to D because A, B needed to be done, and now your paying Cache Write + Input cost.

Its hard to have a good flow to keep things cached, when to really /new and when to not have it expire (and that assumes there are no issue with the provider moving your session around and forcing new cache hits. MiMo did that a lot in the past).

Something that i also advice more and more to people. Get a microphone, download openwisper and talk (text to prompt). You tend to give more information vocally, then writing as its in our habit of programming to not be verbose. Its like people are afraid of long prompts. While just talking to the LLM tend to give it much more information to work with, often resulting it being able to skip steps.

386 to 486 to the first Pentium (with the bug!)... You did not upgrade in place, it was often a new system. Sure, you maybe kept your screen, keyboard etc but ... The only upgrade we had on the same MB, was a coprocessor upgrade. Remember those? Each new generation of CPU was a new motherboard. Upgrading CPUs in the same MB really became a thing only later on.

GPUs had a shelf life of barely a year. Its been 35 year but i remember TNT to TNT2 having like 9 month in between. Moving from 2D to 3D involved a constant cost as GPUs evolved fast and the latest games required latest hardware.

We have not talked about the ISA, AGP, and PCI fun ... The “bus wars”.

DOS to Windows 3.1 (and OS/2 somewhere in between) to 95 ... with software being pushing hardware, just like games did.

This is why people are spoiled with cheap PC hardware where its cheap, and easily lasts 4+ years. Even with the bad memory price and more expensive GPUs, your can stil buy a $1500 system that will last you years (with maybe some lower game settings later on ... or the catalog of 10.000s games that will easily run on a mid tier GPU).

PC hardware has become boring but extreme stable. You can run GPUs for year, switch MBs without issues while keeping large amounts of old hardware. That was NOT the 80s and 90s that i remember.

Carrot: Free vacations? Merch? Benefits ...

Stick: Scared because taking actions against the Buy (aka you own nothing) is paramount to going in a fight with a entire industry.

We saw the same issue with GLM 5.2 when they still published publicly who the providers are on their website. Most ran FP8 but one was doing FP4, so you had this issue where one moment you had the better FP8 and another session you had the FP4 provider.

You can check the internet archive, it was in the FAQ part before they hide/removed it. So if you looked up the providers, and the published quants, yea, ...

Given that a lot of complaints are coming from people that felt OpenCode Go Flash feel like a step down compared to old OpenCode Go/DeepSeek API directly, it smells of a quantized down provider is mixed in.

Its a system that mostly benefits large companies. Just about everybody else hates the system.

It places all the work on the affected parties to prove their innocence.

And suing the fake claim, is years of work and cost. This is why companies like Tracer.AI, ... get away with it. And how the system even gets misused to remove competitors or flat out blackmail.

Even worse is when we have hosting companies in Europe, who simply take American DMCA claims for granted because investigating them costs time and money.

The entire system is broken (by design?).

/Edit: what is with the downvotes? Did i write something that is incorrect?