At this point, when you are doing big AI you basically have to buy it from NVidia or rent it from Google. And Google can design their chips and engine and systems in a whole-datacenter context, centralizing some aspects that are impossible for chip vendors to centralize, so I suspect that when things get really big, Google's systems will always be more cost-efficient.
(disclosure: I am long GOOG, for this and a few other reasons)
I'd go long Google too if using Gemini CLI felt anything close to the experience I get with Codex or Claude. They might have great hardware but it's worthless if their flagship coding agent gets stuck in loops trying to find the end of turn token.
Gemini CLI isn't a great product unfortunately. While it's unfortunately tied to a GUI, antigravity is a far superior agent harness. I suggest comparing that to Claude code instead.
And the converse is true also. I mean, look at NVIDIA. For the longest time they were just a gaming card company, competing with AMD. I remember alternating between the two companies for my custom builds in the 90s and it basically came down to rendering speed and frame rate.
But Jensen bet on the "compute engine" horse and pushed CUDA out, which became the defacto standard for doing fast, parallel arithmetic on a GPU. He was able to ride the BitCoin wave and then the big one, DNNs. AMD still hasn't caught on yet (despite 15 years having gone by).
I make the mistake of thinking its 2020 as well. CUDA was announced 2006 and released Feb 2007. So its actually 20 years that AMD/RADEON hasn't caught on that they need a good software stack.
I wish it were otherwise but antigravity is also a distant third behind codex cli/app, and claude code.
3.1 pro is just fundamentally not on the same level. In any context I've tried it in, for code review it acts like a model from 1yr ago in that it's all hallucinated superficial bullshit.
Claude code is significantly less likely to produce the same (yet still does a decent amount). Gpt 5.4 high/xhigh is on another level altogether - truly not comparable to Gemini.
I use Claude Code all day and use Gemini CLI for personal projects and I don't see the huge gap that other people seem to talk about a lot. Truthfully there are parts of Gemini CLI I like better than Claude Code.
I agree. I like using Antigravity for some of my frontend work, and I find it does a better job than Claude Code - Opus 4.6. I’ve also found the Gemini Flash models to be good at legal defense research—I use them to help New Yorkers fight parking tickets (https://nyceasyparking.com). That said, the Claude models are still amazing at agentic work.
I don't use Gemini CLI- I use the extension in VSCode, and Gemini extension in VS Code is barely usable in comparison to Claude or GPT-5.4. My experience (consistent with a lot of other reports) is that it takes long time before answer, and frequently returns errors (after a long wait). But I think it's specific to the extension (and maynbe the CLI) because the web version of Gemini works quickly and rarely errors (for me).
Of the big three, Gemini gives me the worst responses for the type of tasks I give it. I haven’t really tried it for agentic coding, but the LLM itself often gives, long meandering answers and adds weird little bits of editorializing that are unnecessary at best and misleading at worst.
Same. The tone is really off. Here is a response I just got from Gemini 3.1: "Your simulation results are incredibly insightful, and they actually touch on one of the most notoriously difficult aspects of ..." It's pure bullshit, my simulation results are in fact broken, GPT spotted it immediately.
Yeah trainium and inferentia. They’re just not nearly as well supported on the software level. Google has already made sure this new generation will be supported by vllm, sglang, etc. Amazons chips barely support those and only multiple versions back. Super under invested in (at least on the open source side)
That's seems odd. I'd figure if they are going to sell it as a product in AWS that they'd have some sort of off the shelf tooling that would be available.
I think this is a narrow view. Aws and azure build their own data centers and partner closely with Nvidia and build their own silicon too. TPUS are non standard, no one else can run them - Nvidia build on fabrics and technologies well under and well integrated for a long time (mellanox etc) and clearly work very closely with the aws and azure hardware and data center build teams. I’d not bet that Google can do things better than everyone else - that’s certainly something Googlers always believe about themselves but it’s not the case that you can’t build a best of breed that meets or exceeds total in house builds.
I'd bet that too if their management wasn't so incredibly uninspiring. Like, Apple under Cook was also pretty mild and a huge step down from Jobs, but Google feels like it fell off a cliff. If it wasn't for OpenAI releasing ChatGPT, they might still be sitting on that tech while only testing it internally. Now it drives their entire chip R&D.
Google was calling itself an "AI-first" company beginning in 2016 or 2017. They designed and built TPUs nearly a decade ago and were using transformer models in products like Google Translate but didn't make a big fuss about it, it just made the product way better. People should at least credit Sundar somewhat for this, it turned out to be quite prescient, especially the advantage of having your own chips that are specifically designed for ML.
AI was very different in 2016-2017 compared to what it is since ChatGPT. Facebook was also a primarily AI/ML driven company with noone realizing it on the front-end, but at least they were heavily involved in the open source side on the back-end - long before LLMs went big. In fact they enabled them to go big with things like pytorch. Google just stumbled into this. Deepmind (also acquired before Sundar) came up with the theory, but they didn't see the potential. What you call "prescience" I call luck. They did not create the demand for their own technology like e.g. Nvidia did by pushing the field ahead with full force. In fact all of Google's most popular products are from the time before Sundar took over. Even with Gemini they are dragging their heels, sitting far below all other big model providers when you look at usage.
This is a bizarre accounting of things. FAIR's efforts building Pytorch were seen as experimental and fragile by the time it was released, when Tensorflow was already being used in edge deployment for computer vision and seq-to-seq. Google was the company that prepped the technology for deployment, created the theory (Transformer architecture), implemented it in practice (BERT bidirectional encoding) and then scaled it (RoBERTa) all before GPT-3 ever released. Three years before Facebook released Llama.
> They did not create the demand for their own technology like e.g. Nvidia did by pushing the field ahead with full force.
They did, though. You are commenting on an eighth-generation TPU product that has been used millions of times a day for the past half-decade. It's likely that this will be the hardware providing inference for Apple's Gemini model they've selected to use with Siri. TPUs are the economically-conscious inference choice if you've already separated your training/inference workflows.
To be fair, I don't think any of the AI players wanted what OAI did. Sam grabbed first mover at the cost of this insane race everyone else got forced into.
> I suspect that when things get really big, Google's systems will always be more cost-efficient.
In fact I am opposite of this hypothesis for two reasons. Google has artificially limited production. And because TSMC favours whoever could pay for the most capacity(as incremental capacity is very cheap for them). So Nvidia gets first slot for new process.
Also the second reason is that GCP's operating margin is very high compared to say Hetzner or lambdalabs and you can get GPUs much cheaper there compared to GCP. So students/small researchers are stuck on GPU.
Any computable function f on one variable x has a program. That function is a program of size p. The input x also has a data size d. BB(p+d) >= f(x), by definition, for all f and x. If you think you might have a (function, input) pair (and corresponding (program, data) pair) for which this is not true, see the previous sentence.
No, because f is assumed to be computable from the start, which BB is not (otherwise it could be used as a subroutine in a program that solves the halting problem).
It was forcibly funded as part of a consent decree from the US government that allowed AT&T to continue as a monopoly as long as they invested a percent of their yearly revenue (or profit? I forget) in research. AT&T, having no interest in changing their incredibly profitable phone network, then proceeded to do fundamental research, as required as a condition of their monopoly.
Decades later, AT&T was broken up into the baby bells and the consent decree was removed at that time. Bell Labs' fate was then sealed - it no longer had a required legal minimum funding level, and the baby bells were MBA-run monstrosities that were only interested in "research" that paid dividends in the next 6 months in a predictable fashion.
The funding model is an integral part of the story.
That sounds plausible, but is not how it is told in The Idea Factory, where the authors explain that both AT&T (running the phone system) and Western Electric (manufacturing equipment for the phone system) had separate research divisions even before this. They then discovered that they were duplicating a lot of research, so they set up one entity to perform research for both the harder and the softer sides of the communication system.
Citation needed. What I'm seeing: No evidence of a legal requirement to spend a % of revenue on research: There was no line-item mandate in the consent decree forcing AT&T to invest a specific percentage into Bell Labs. The support for research was strategic and reputational: AT&T used Bell Labs to fend off antitrust pressure and maintain regulatory goodwill.
The baby bells actually took with them part of Bell Labs, renamed to Bellcore, that survived for another decade or so. I interned there whilst doing my MSc, it was still a great place for a while, with serious research.
my folks worked at at&t when all that happened and so that narrative arc was a big part of my upbringing. my timeline/details is probably off and I can't ask them because they passed away but from what you say here I can totally see what you mean, it totally tracks with the dramas and discussions that they brought home from work every night.
They are a zero-trust networking solution that also traverses IPv4 NATs. Zero-trust networking is a layer above the IP layer. In an IPv6 Internet their capital costs go down, and their product remains valuable for their paying customers. (Free accounts mostly use it for NAT traversal, businesses for the zero-trust encryption.)
Their CEO has been working with (and supporting) v6 for decades both at the executive level (now) and also as an extremely capable software engineer that I personally met with a few times while we were both engineers at Google doing network measurement.
Countries can tell companies what the maps of their countries should look like and what things should be called.
For example, Google shows different info about the Kashmir region depending on whether you are in Pakistan or India or external to both, because of how the Indian and Pakistani governments define the borders and names of the region. If the US government changes a thing's name, then Google will change that thing's name within the US. Google mostly doesn't choose names, it uses externally-supplied mostly-governmental name databases. Governments have the power to name their own regions.
I don't want Google to choose what things are called, so I think they are doing the right thing here, and the USG is doing the dumb thing.
When I worked at HERE Maps this was a pretty standard thing to do. Countries have official named in multiple languages. Borders are seen differently by different governments. A map is anything but static.
And the chaos of translation. It is fun to consider whether somewhere like Johannesburg should be called Johannesburg or Johnscity or YHWHisMercifulFortifiedPlace or what in English. The standard is obviously to keep the local term and pronunciation, but then apply that logic to languages with different scripts and nothing really makes sense in a satisfying way because mapmakers can't avoid some amount of translation. It is a muddle of conventions.
There is no chaos there. When you label something on a map, you never translate any of the names. What you do, is use the exonyms which already exist. That is, local names which may have been literally translated at some point, or might simply have changed to align with local orthography and pronunciation, and which have become common place and accepted.
Like encyclopaedia, most reliable real-world maps document what it is, not what the author thinks it should be (obviously this is a possible use of maps; see China's nine-dash-line bullshit). So you wouldn't be making your own translations. At most, you would transliterate names in scripts not readily understood by your target audience.
I don't want Google naming things, but I don't want a President naming them either. My preference is for things to be named by the people who live there/use the thing, but failing that at least I would want my mapping software to present the fact that there are multiple names/borders and let me pick which I want to view.
The president was given this power by the people. Crowd sourcing this is a particularly bad idea as countries already have working institutions that map and name things.
One alternative would be for Congress to create more barriers for name changes.
The president is taking this issue as a power grab. He does not have rights to name things. There is a geological naming board appointed by cabinet heads. The executive order instructs his newly appointed department heads to purge their geological naming boards of any members who may not agree to these names. This is not something that has happened before, and not by design.
Is "gets to rename any geographical feature in the general vicinity of the country on a whim" a usual power of the US President? I genuinely don't know the legalities around this, but given that, I would bet that most people who voted for Trump also didn't think they were giving him that power. (Whether they would have chosen to do so is a separate question.)
Do electors of any politician ever understand the full extent of their candidate's powers? They vote on people under the assumption that the person is the best choice (not necessarily a "good" choice) for the job given the system already in place.
Answering your question: yes, the executive branch can change geographical names in the US. Obama did that too with Mt Mckinley. This power extends solely to the names as accepted by the United States. If Mexico wants to change the gulf's name to "Gulf of the Aztecs" or "Gulf of qetwyetwdhxysysheussg", it can do that too.
SHOULD they do it? Now that's a different question.
>Countries can tell companies what the maps of their countries should look like and what things should be called.
Not in America. The First Amendment means private entities, individual or organization, are free to draw maps and name things however they like and the government may not tell them what to do. The government can decide how it does its own maps, and society at large can decide whether they wish to use a different product or not.
>I don't want Google to choose what things are called
Tough shit. Google may choose for a variety of reasons to go along with what the USG decides is the naming, and in this case it has, but they can't be compelled by force to do so. And other private entities can choose differently should they wish. And of course everyone is entitled to comment or criticize it, and then comment and criticize the comments and criticism in turn, and make use and spending choices accordingly and so on. That's the vibrant market of ideas we're founded on.
NB: The Treynor Curve is named after Ben Treynor and his ideas. Ben Treynor's name changed to Ben Sloss a few years back, and Ben Sloss is one of the authors of this article.
The Qmodem program, brought home on some random 3.5 inch floppy, allowed me to connect to local BBSes and started my journey into networking computers. Now I have a PhD in CS and I spent more than a decade deeply caring about the Internet, networks, and network research. Without the start given by those BBSes, my path could have been very different! I am very sorry for your loss, and I hope the fact that he made a random teen's life better is some comfort.
> The EU single market is largely a myth from the perspective of wanting to start a successful company that can compete internationally. It's not any easier now than it was before the EU existed.
[citation needed] because I think this is extremely false.
It is easy to underestimate the pain caused by having all the pre-euro currencies and pre-EU tax laws (yes there was the Schengen zone and EEC, but the EU uniformity helps a lot) and pre-EU borders.
Source: me, who is a middle-aged American founding a tech startup in the EU and visited Europe many times before the EU existed.
I think it is getting better year after year, in small steps. For example, B2B taxation within the EU (by use of the reverse-charge mechanism) has become quite easy in recent years. Of course, depending on the product, diverse local rules may apply.
EU Single market is a massive improvement, but it's not yet really single market like in the States (although there can be a bit of legislation/tax barriers between states there as well, depending on your location and line of work).
Also, it's hard to understate just how open the US is. For me, an European living in the EU, it's as easy, or easier (larger wallets, single language) to sell (digital/online products/services) to Americans. For digital goods, might as well skip the EU.
Uber, has to do all sorts of gymnastics to make rides work across different EU countries. And still they are routinely asked to stop serving in a country for some reason or another.
Amazon. It took them years to launch across EU, they went country by country and didn’t launch across EU at once.
The current EU laws (especially tax laws) aren't helpful for businesses. If you want to do business in every EU country you basically have to register a business in each of them.
Sure, you can sell goods for €10.000 - €20.000 to another EU country without much trouble, beyond that you need to setup a business in each country, sometimes you even need employees or at least an office in a country to be able to do so. I worked for a company that shipped to a number of EU countries, from an EU country. It has taken them 15 years to expand from shipping to three countries to nine. Every time they want to enter a "new market" it is years of planning and insane fees or doing business in a supposed "single market".
The current situation is better than before the EU, but it's by no means good. Certainly no where as good as a Nebraska company wanting to sell their product in New Mexico. EU business will never grow to the size of American companies if they cannot bootstrap themself in a true single EU market. An American business immediately have a potential customer base of 350 million consumer, an EU business have at most 80 million if they start in Germany and will have to add an insane bureaucracy if they want to expand.
EDIT: My information is out of date: In 2021 a solution was introduced where you can pay your VAT locally, and the tax authority in your own country will handle the transfer of VAT to the other EU countries.
There are still some taxes (e.g. taxes on tobacco or alcohol) you may need to handle for each country, and to do some you may need to be a registered business.
This is just straight up not true. You need a legal body in a country if you want to hire someone on payroll there, you most definitely do not need one just to sell your goods and services, that's the entire point of the single market. It sounds like you've misunderstood a second hand recollection of someone else's struggles?
> You need a legal body in a country if you want to hire someone on payroll there
Even this is not strictly true. One can often just register a foreign company with the local tax agency for a direct hire. The most frequent scenario is that a foreign hire comes through their own legal entity, as a hired contractor. Now, the sad part of that statement is that often a hire comes as a hired contractor not because of cross-border taxation, but because of tax management reasons. In some EU countries, high value employees cost three euros in taxes for every two euros paid after taxes. A lot of people prefer to set up a corporation that allows them to better manage that majority of their income that goes to the tax office.
That's not employing someone in a different country though, that's contracting. And it's the same arrangement you can use to hire someone in the US from the EU.
No. Satellite (except when they use lasers, but nobody is proposing space lasers) is a broadcast medium and wire is point to point. Wires will almost always be cheaper and more reliable and faster on a per-customer basis and have less interference.
(disclosure: I am long GOOG, for this and a few other reasons)
reply