Facebook is set to buy billions of dollars worth of AI chips. But Elizabeth Warren is worried about the climate impact. Will ChatGPT need an EPA permit?
Whole New Industry
Jensen Huang, Nvidia’s co-founder says AI Is a ‘Whole New Industry’.
“This last year, we’ve seen generative AI really becoming a whole new application space, a whole new way of doing computing,” Jensen Huang, Nvidia’s co-founder and chief executive, said Wednesday. “A whole new industry is being formed, and that’s driving our growth.”
“This is transforming everything about computing,” Intel Chief Executive Pat Gelsinger said at a company event Wednesday.
Fueling that growth, tech companies have plowed money into AI in recent months. Meta Platforms Chief Executive Mark Zuckerberg boasted in January that he was planning to buy billions of dollars worth of AI chips from Nvidia this year. Google recently rolled out a new, more powerful version of its Gemini AI system. Microsoft has its own AI assistant tools and has bought tens of thousands of Nvidia’s most advanced chips.
Huang said generative AI is kicking off a new wave of investment worth trillions of dollars, which he believed would double the amount of data centers in the world in the next five years and deliver big new market opportunities for Nvidia.
Nvidia’s boom is largely dependent on a handful of those companies. Nvidia disclosed that 19% of its sales in its last fiscal year came from one end customer. It didn’t disclose the identity of the customer, but cloud-computing companies like Google, Amazon and Microsoft accounted for more than half of the revenue for its data-center division in its latest quarter, or over $9.2 billion.
OpenAI, which counts Microsoft as its biggest financial backer, is itself looking for ways to increase access to chips that power AI. Chief Executive Sam Altman has discussed with investors and government officials a build-out of AI-related infrastructure, including chip factories, that could cost as much as $7 trillion, The Wall Street Journal reported this month.
Nvidia’s early-mover advantage—the company began investing in AI more than a decade ago—has made it difficult to displace, though, in large part because of its head start in software. The company has developed a vast amount of software catered to healthcare and robotics, among other key applications of AI, and the tools it created to exploit the capabilities of its chips have become an industry standard.
Is AI a new industry or a way to enhance existing ones? Both?
I’s be lying if I said I looking forward to a new and improved Facebook given that I don’t use Facebook at all.
Perhaps AI will soon be embedded into everything. This does not have me concerned, but Senators Elizabeth Warren and Lindsey Graham are leading the charge against it.
Strange Bedfellows
Please note Senators Elizabeth Warren and Lindsey Graham propose a Licensing Agency For AI and Other Digital Things
I am quite certain that dynamic duo could ruin anything.
Climate Concerns
AI climate concerns are now in the spotlight according to the Wall Street Journal.
Mr. Biden is lucky that the AI revolution accelerated under his watch, just as Barack Obama was fortunate with the shale fracking boom. But AI advances are happening despite government, not because of it. Mr. Biden’s executive order last autumn has created new uncertainty about whether and how regulators will permit AI.
Some Democrats are already threatening to suffocate AI with—what else?—climate regulation. Democratic Senators this month introduced a bill that would direct the National Institute of Standards and Technology to recommend administrative actions to mitigate AI’s environmental impact. Will ChatGPT soon need an Environmental Protection Agency permit?
Progressives fret that AI systems will generate as much CO2 emissions as entire countries. Possibly. An AI-driven web search consumes four to five times as much energy as a conventional one. Can the U.S. electric grid, already creaking under the force-fed green energy transition, handle the increasing demand? Is anyone in government thinking about this?
Note to Congress and President Biden: Industrial policy doesn’t make countries or businesses great.
How Good or Bad is Google’s Gemini Artificial Intelligence (AI) Tool?
Yesterday, Google pulled the plug on Gemini image generation for inserting diversity into historical images. What about simple math and logic questions?
For discussion of some glaring flaws, please consider How Good or Bad is Google’s Gemini Artificial Intelligence (AI) Tool?
I’m not ready to risk my life on it yet, but there is little doubt AI is coming big time.
NVDA sells at 40 times sales.
LOL…..
And with it selling at 40 times people think you are jealous that you don’t own a small piece of it.
quote:
“I think we’re all trying to find the answers as we go, and it’s changing every day — and it’s not just our industry, but it’s every industry that AI will be affecting, from accountants to architects.”
link to hollywoodreporter.com
“Huang said generative AI is kicking off a new wave of investment worth trillions of dollars,”
Yup.
And that’s also all it’s kicking off: Yet another way for gullible backmarkers on Fed welfare to waste resources and flaunt their stupidity by what they believe is “inveeesting” some more of the fresh print The Fed stole for them
While in the real world: “Generative AI” is; over any domain with more dimensions than rather simple optimization problems; probably the least efficient of all possible means of programming a computer.
The few “wow” experiences elicited; are solely due to the utterly enormous “investment”/resource misallocation thrown at it by the ever more clueless hence gullible Fed welfare idiot class. If practitioners of _ANY_ other means of instructing/programming/training a computer, could even halfway be bothered to waste their time on such trivialities as creating a cartoon drawing of a seal with shades on (and then parade the triviality around like something to be proud of..); they could have done it with far less than one millionth of “Generative AI’s” required processing cycles.
But alas, they’re simply not that stupid. And hence have better things to do. What is sad, though; is that more and more of even such better things; will be directed by this latest “AI” fad among the Idiots-on-Fed-welfare classes: As NVidia is a prime example of: This nonsense is what they’ll now focus their development on. Hence what they’ll hire graduates who honestly deserve better, to work on. Rather than something useful and not purely resource wasting.
This is the best thing I’ve read in a long time: “While in the real world: “Generative AI” is; over any domain with more dimensions than rather simple optimization problems; probably the least efficient of all possible means of programming a computer.” very good sir.
And just remember that genitive AI isn’t even real AI.
Now we know the next tulip craze.
It is conjectured that Nvidia’s H100 chip usage will consume as much power as Phoenix by the end of 2024.
Given the current state of energy production, transmission and storage in the US and ROW, some sectors of the economy will have to sacrifice their energy consumption should AI’s growth continue.
link to tomshardware.com
No need to pay for heatpumps and radiators, when some “AI” “investor” will pay the better part of your heating bill if you heat your house running models on H100s in the winter….
Heck, enough demand for compute, may even make steam engines viable again….
Less flippantly: Given sufficient demand for compute power: Any requirement for just heat itself; home heating; stoves; swimming pools and tub, general hot water, industrial and chemical processes….. ; may well become more economically delivered, by colocating with compute power for hire and using its waste heat.
Ha! Ha!
The US grid is 3rd world. How about the US ramp up its antiquated grid first then we talk pie in sky.
Assuming a world where available energy is what constrains computing cycles; and where simultaneously computing cycles are used for something valuable: No matter how poor the electric grid: It still makes more sense to heat your house running current though an H100, than to run similar current through a simple resistor for heat…..
Say you live in Fairbanks in a standard issue poorly insulated house: If chips were cheap enough per cycle: Why not heat the place by passing the kilowatts required to keep it cozy through chips mining crypto? Competitive pressure would most certainly prevent you from recovering your entire heating bill, but you would get some back….
Of course, in practice, in colder parts of the US people almost exclusively heat with gas directly. But in some cold parts of Northern and High Alpine Europe; they do use electricity.
Of course, also in practice: H100s are still quite a lot more expensive than resistor wire…. Enough so that assembling radiant floor tile out of H100s, is still pretty much a loser anywhere.
A somewhat lower hanging fruit, is colocating chemical and industrial processes (say waste incineration) requiring heat as a specific input, alongside datacenters. That is already either being done, or at least in the planning stages several places in colder climates. Ditto remote heating: Using datacenter waste heat to heat water which is then pumped through radiators in nearby residences and businesses.
Fundamentally: As long as there is specific demand for electrically generated heat in and of itself; while at the same time compute capacity is constrained by access to electricity; it ultimately makes no sense for the former to be bidding electricity away from the latter, instead of rearranging for better synergy.
AI is a branch of signal processing. It’s been around for a long time. It’s appeal is you substitute expertise for training a NN using gobs of data.
Our government can’t even secure the border. I’m pretty sure they would be clueless in regulating AI.
AI is just repackaged censorship at times. Actually better for the companies because now no one person is responsible.
You don’t buy tulip bulbs to plant and enjoy the flower.
You buy tulip bulbs to sell to someone else at a profit.
Just a reminder.
Back in the 16th century, anyway.
From my professional and personal need, AI is a solution in search of a problem. However, If I was a government, in advertisement or credit rating company, looking to get deep information on every individual, AI is the answer.
I forgot one major use for AI: internet censorship.
Wonder when the boat will tip over?!
“Yesterday, Google pulled the plug on Gemini image generation for inserting diversity into historical images. What about simple math and logic questions?”
Well, it was designed by leftists, just as Chat GPT was. According to leftists, math is racist.
Homework, reading, writing, and maff are for white people. Basse’ball after school is for POC’s.
“An AI-driven web search consumes four to five times as much energy as a conventional one.” Who is going to absorb that cost, the tapped out consumer? IE, companies that advertise on search engines will have to raise the price of their products since I doubt Google will be willing to absorb the cost. Seems it wilI take a lot more debt to cover it. If debt hits a wall and the consumer can’t cover it, seems AI would just become a large overhead for the companies whose stock is blasting off.
Nvidia will afford the costs as they are making so much money selling chips.
This is where the New York judge’s ruling on Trump comes in. It will be financed by incredibly high fines on individuals who don’t toe the line. Since there are no victims, the fine will go to the government who will then give it to favored companies and individuals with exceptional political talents. Problem solved.
This is where fusion power generation comes in. Many major advances occurring. Microsoft has a contact with a fusion power development company to deliver a working plant to power one or more data centers by 2028.
“An AI-driven web search consumes four to five times as much energy as a conventional one.”
Touche!!!
And for richer problems, that four to five grows far beyond exponentially.
It’s why “AI” always ultimately “fails.” From the 50s and onward. Early gains always looks impressive (at least to the gullible), but then the hockey stick turns upwards; into effectively vertical; so abruptly, that any real world usage ends up very bounded.
Ultimately, efficiency is what matters. And “generative AI” is about the least resource efficient way to program a machine imaginable, for all but tightly constrained “optimization” problems.
AI is real, in 5 – 10 yrs AI will eliminate most of the white collar jobs that require a college degree thus effecting mostly latte carrying brainwashed liberals. Middle America will not be effected, can’t wait to see how these clueless libertards survive on their own without a handout from government.
They’ll just start acting like degenerate red state Republicans. Claiming moral and economic superiority while hoovering up tax dollars from neighboring states is a time honored tradition in the south.
You are so upset that Republicans do not fall for deranged liberals’ vote buying schemes using tax payers money in the form of handouts. BTW, there are not enough tax dollars to buy votes anymore, it’s all borrowed/printed dollars 34 Trillion and counting.
Speaking of “clueless” the word you wanted to use was AFFECTED, not EFFECTED.
SPY 1M needs a correction, for a few months, for a sling shot up.
Yes an entire new ecosystem will be needed for this new industry
1. New power grid – AI uses 10x as much power as a normal computer (for now)
2. New networking infrastructure – The current network is too slow and constrained for AI
3. New laws for deep fakes, copyright, foreign use, competitive use (AI monopoly?), criminal penalties for misuse, etc. If AI commits a crime who is legally responsible?
And then the unknowns…. will AI be allowed to trade stocks, bonds, commodities, etc? If so, will insider information count if it can read everything on the internet, even undisclosed financials then trade for profit?
Will AI be used to profile all Mishtalk readers by comments and build forensic profiles and cross match those with other sites like ZH and upload to a massive database at a three letter agency?
OK that last one is a joke cuz it’s probably already being done, AI will just make it faster.
But the most important question of all is how do I profit from all this? Will AI know the answer? I am gonna ask.
AI already trades stocks and bonds. Algo’s have been doing that for years and yes, they scan the internet now reading news, financials etc.
Not sure what new networking infrastructure you imagine we can add. Fiber is already pretty much everywhere in major cities now and Cellular is not going to get much faster than what we have now. Plus if you ask a question and get an answer, why would that small amount of info transfer require a lot more bandwidth?
Suggest you watch this video. The network issue is topology and protocol not medium (fiber).
link to youtube.com
“1. New power grid – AI uses 10x as much power as a normal computer (for now)”
Not far off. The Mac Mini uses about 40 watts. The RTX 4090 video card uses 330, and there will be racks of similar cards in the data center for AI.
The other day I ran across a discussion that mentioned 10 Gb Ethernet uses noticeably more power than regular Ethernet, so there is that too.
All this “long range” thinking about second order implications won’t matter much if AIs keep getting embarrassed by crap-tastic results.
The hype-to-reality ratio in AI is *huge*.
In fact, it is impossible to think about Google’s AI fiasco without also immediately wondering about just how crappy Google’s *search engine* might really be behind the scenes (although obscured by its lack of sizeable competition).
In a clear case of synchronicity this is from an invest with us email that arrive this morning.
“AI-driven technologies need substantial energy and energy infrastructure to support them.
Schneider Electric projects that AI-driven power use will grow 25%–33% annually through 2028.
Much of that power use will happen at new data centers sprouting up around the country. The US data center construction market is projected to grow from $20.21 billion in 2022 to $28.56 billion by 2028.
Boston Consulting Group expects data centers’ share of electricity use to triple to 7.5% by 2030, driven in large part by the expansion of AI.”
During the gold rush there was more money to be made supplying the miner’s equipment than mining. Nvidia has a temporary lock on the processors and their stock price reflects that, so maybe power and data center support are where it’s at.
Datacenter’s profit margins are razor thin, it’s a commodity these days. Utilities have been beaten down and I am tempted there but we have huge labor shortages (read cost overruns) for building new utility plants or anything else for that matter.
Increase demand for electricity and insufficient supply will drive electric costs up and generate way more inflation so my bonds play is good. I just loaded up on yesterdays 30 year TIPS. Those are only sold once a year. T-bills are still a buy for me despite all calls for rate cuts this year.
The best play may be infrastructure companies like Quanta ticker:PWR. I am kicking myself for not loading up when it was $80 a couple of years ago.
But at some point, they too will be hit by the labor shortage. Another 4 million people are slated to retire this year between boomers, the sick and the rich.
Energy will not be a problem because of all the infrastructure that has been built to provide charging for all the electric cars that will replace our current (pun intended) manifestly evil cars.
If companies have the right to finance political campaigns through the Citizen’s United decision of the Supreme Court then I see no legal reasons why AI would not have the same right if it incorporates itself under US law and choose those politicians who favor them over us as the corporations do.
Government activity always follows innovation and eventually stifles it. The opposite is rare. The only reason regulation has not stifled the computer industry is because those in government are too stupid to figure out how to gain from it. Not to fear, they will get there.
Yes.
They will use AI to draft more regulations and legislation.
No, AI won’t require EPA permits. But growing data center loads are helping to set utilities up for an epic battle.
Read any utility industry news and you’ll see that many are postponing or cancelling coal and gas plant retirements in order to meet the rapidly growing data center loads.
States that prevent these extended lives will just push data centers into states that allow them.
Here’s one. There are many. It is a big trend. link to bloomberg.com
There is always a new need for more energy. Our economies and living standards cannot grow without more energy.
And lately, the focus is on more electricity, which has grown worldwide from 17% of all consumed energy in 2010 to 20% in 2023. In that timeframe we have added demand from more computers, smartphones, internet use, cryptocurrencies, EVs, and from some of the world’s poorest people who are starting to access electricity for the first time.
Over the next 10 years the demand for electricity is going to keep growing as all of the above reasons (and more) keep growing. After all, there are still 1.2 billion people in the world without electricity. And AI, which is another new area of demand, explodes. And then there is EV charging
A lot of investment will be needed to expand, modernize, and toughen the electric grids. Whether the investment comes from government or the private sector doesn’t matter, as long as it happens.
Given the strong anti-government sentiment here, I suspect that many would prefer if the private make the investments needed.
I am not a big fan of government either, but I accept that they have a purpose, and I would not oppose more government investment in improving our electric grids.
WTF…why the down votes? This is a great comment IMHO, and it makes perfect sense from an investment opportunity standpoint.
Papa and I certainly have a strong disagreement with regards to climate change, and but his investment strategies are generally sound in my view.
Many have become so ideologically driven that unless they are 100% ideologically aligned with another they dismiss everything that a person says with whom they disagree with on one topic. Very foolish. I read alot of the comments posted on Mish’s articles. Some are good while others aren’t. Bottom line is I read them to get a different perspective on things. If we keep this ideological ego driven crap up, we are doomed.
My ideology doesn’t really fall into any one category. My politics are of the Woodsie Guy variety. I’m a party one, and I’m perfectly fine with that.
I suspect 90% of those downvotes come from the zerohedge trolls that come here from Mish’s syndication there. I understand mish run’s a business and wants the traffic but the price to pay (toxic trolls) is a bit too much. It won’t be long before Mish starts having to swing that ban hammer in full force. I have already seen attacks on Mish and others from these trolls.
I really wish we had the ignore button back, that seems to kill most of the trolls because they got no attention and go away.
It’s an election year so the toxicity is only going to ramp up as we get closer to the election.
@Mish – Consider suspending your posts on ZeroHedge until Nov 5 or bring back the ignore button.
Reminds me of my old procmail filter that sent emails from my list of certain folks to /dev/null automatically upon receipt.
The bizarre thing about technology is we had far more control and options decades ago than we have now.
link to en.wikipedia.org
I wish we had kill files like back in the Usenet days. Bring back the VAX too!
And to think: We even had planes that didn’t fall out of the sky at one time. And roads without craters. Houses to live in, instead of cardboard boxes. Schools where kids were taught, not shot at…
Them’s were the good ol’e days……
We are doomed.
Doomed I say.
“Here are your winnings, sir.”
“Thank you”.
Had the US actually built the 400 planned nuclear reactors instead of the paltry 105 that are now down to about 96, we wouldn’t have even needed these inflationary green initiatives in most places. We certainly wouldn’t be talking about oil and gas the way we do now, nor Middle East wars. The NRC and Congress are to blame for overregulating us to death, because radiation scares people.
Water under the bridge.
Government has its uses and government subventions also have their uses. It often is the only source of seed money for expensive projects whether it be financing Columbus’ voyages to the national highway system and beyond. The justification governments have to give in order to convince people to buy into it are often very creative.
In another few years, ChatGPT will require for Elizabeth Warren to obtain a permit before opening her mouth.
Hey look it’s the Y2K pump all over again. I wonder what comes after……
Thought about that. The difference may be the end point. With Y2K, once 1/1/2000 arrived, the need for IT spending at the pre 2000 levels ended. Kind of a hard cliff. No such end point for this tech cycle
No end point unless it doesn’t live up to the hype.
I think off the dot.com bust a point of comparison. Nvidia’s announcement is obviously self-serving.
There was coincidently an internet bandwidth rush at the same time that had fibre transceivers selling for hundreds of thousands of dollars that within a couple years were below $500. A GPU advantage is even shorter. This will end the same way….. the fibre transceiver companies went bust or were bought up for pennies
19% to one customer … Uncle Sam? He’s busy reducing inflation by building AI clusters for more diversity, especially weapons systems diversity.
I call myself my wife’s “Artificially unintelligent Husband” using a ROBOT Voice in front of Friends at parties and get huge laughs.
I sound like the CONE HEADS on SNL. I am good at it.
It is REALLY fun, because I then ask my pals and gal friends: “Ask any question and I will answer it” – – again in SNL CONE HEAD vocals and I make up answers quickly or say: “that question is being reviewed by my Masters” and “NO, you are wrong,
>>>>>Robots rule the world of answers and you must comply by stating your intentions..”
IT IS A HOOT!
For everyone’s situational awareness, the Biden regime issued an Executive Order last October on ai, and the National Institute of Standards and Technology, (NIST) is the federal government lead on developing ai standards like “system cards” that will probably become policy and/or law. They publish the peer reviewed papers and collaborate with OpenAI and others. For the DoD, the lead office is the CDAO at ai.mil.