The tech giants want more electricity, and their tilt towards nuclear power continues.
Amazon Web Services, the cloud computing arm of global online retailer Amazon, is the latest tech giant to go for nuclear power. AWS needs more energy to feed its data centres with servers for computing, storage and of course, the very power hungry artificial intelligence (AI) technology.
AWS has signed an agreement with Dominion Energy in Virginia, USA, to work on getting a small modular reactor (SMR) up and running. While increasing amounts of electric generation is required for data centres, it can’t be the dirty emissions producing power of the past; hence, nuclear is the option, some 300 megawatts to start with.
AWS also has an agreement with the Energy Northwest consortium of utilities in Washington State to develop four SMRs. These will produce 321 megawatts initially, which can be increased to 960 MW.
Earlier in the week, Google said it has entered into an agreement with Kairos Power in California to develop SMRs using fluoride salt-cooled high temperature technology.
The first SMR from the Google-Kairos agreement is meant to go online by 2030, with more reactors being deployed over the next five years. The project is set to create 500 MW of power generation.
There is much debate around the feasibility of building more nuclear power generation, particularly now solar energy has dropped in price massively. Nevertheless, the tech giants are serious about nuclear power, including Microsoft, which was one of the first to cast side-eyes at the allegedly carbon-free method of electricity generation, using SMRs.
So much so that Microsoft signed an agreement with Constellation, the company that runs the Three Mile Island plant in Baltimore, Maryland, for 835 MW. That energy will power Microsoft data centres in three states, Pennsylvania, New Jersey and Maryland, expanding to 14 other states. If Three Mile Island isn’t familiar to you, go back in time to 1979 and check out what happened at the time.
It's AI-lection time
To the surprise of nobody in particular, AI which should be renamed Imitation Intelligence, is used to automate dark influence operations on a mind boggling scale.
Researchers Darren Linvill and Patrick Warren at the Clemson University in South Carolina mapped out a big AI powered bot network that includes at least 686 accounts, likely more, that uses large language models (LLMs) to create posts on social media that look like real replies to what people say on the Twitter-X social media network.
The pair discovered the network has made over 130,000 posts since March this year, and it aims to influence the United States presidential, Senate and House of Representatives elections, and also the World Health Organisation's Pandemic Preparedness Treaty.
No prizes for guessing who might be behind the network: "One leading target of this campaign is the 2024 Presidential election where this network supports former President Donald Trump and attacks Biden-Harris-Walz. Network post output remained fairly low until just before the Biden-Trump debate in late June. Since then, there have been several days with over 1000 replies from the network supporting Trump or attacking Biden-Harris-Walz," the report found.
Disinformation as reported on by NBC News in this story on Russia's Storm-1516 group is cost-efficient and can be effective. Particularly so if it's augmented by AI and technology in general, to increase its reach and volume to drown out the truth with lies, no matter how improbable they are. You can expect this bad situation to become way worse rapidly as threat actors get better at fakin' it.
The chip biz is a wild ride in AI times
If you're an investor in Taiwan Semiconductor Manufacturing Company (TSMC), rejoice: the contact chip maker is doing stunningly well, with a 54% jump in quarterly profit reported this week. That profit hike is due to AI chip demand, with Nvidia and Apple putting in big orders for semiconductors made with a very advanced process.
However, scroll down in that Reuters piece linked to above and read about another company, ASML, that is making investors concerned the chip boom time could be heading towards a bust.
The Netherlands based, ASML, the name of which originally stood for Advanced Semiconductor Materials Lithography, is the company that makes the huge and very, very expensive machines used to produce the semiconductors for computers and AI.
The machines are so expensive even profit-laden TSMC is going "nah, that's too much". So, Europe's most valuable company is cutting its forecasts for next year due to weak sales - in other areas than AI of course. We'll see where that ends.
Original chip gangsta Intel which has been clobbered badly by investors for mostly missing the AI train, and which is now rumoured to be on the block after mass redundancies, losses and its share price falling off a cliff, did something unusual this week.
Intel and AMD which used to be bitter rivals have joined forces to help further the chip architecture that started the microprocessor revolution, the x86. It's an advisory group, to promote the chip architecture behind both Intel and AMD's fortunes. Does that make sense, or is it a sign of desperation indicating that the two companies are losing the battle against chip designs from the UK's Arm? We're about to find out, probably as soon as next year.
5 Comments
Current Small Modular Reactors (SMRs) are seen as one of the most effective energy generating technologies we have in the world today to supply us with large scale, efficient and safe power. Linking these to the technologies used at 3 Mile Island just because they have the word 'Reactor' in them seems disingenuous.
Why? It is the reactor where all the danger is.
SMRs have the same issues as the big ones. The primary value proposition of SMRs is that their cost can be brought down because they're 'modular' - not because they don't have many of the same safety issues.
I guess because they're much smaller, instead of a humongous BOOM, you get a smaller boom. That's a good thing, right?
Read the article again (or do a quick web search). Microsoft is literally re-opening a reactor at Three Mile Island.
For further research, read about the inevitability, and inherent, emergent, unpredictability of accidents in complex systems: https://www.jstor.org/stable/j.ctt7srgf. I challenge you to read that book and maintain your current views.
Anyone claiming SMRs are efficient and safe is making the same claims that have been made about all nuclear generating technology. When did anyone ever say, we're building something inefficient and unsafe? But in fact, they haven't been safe in operation, even before considering how humanity is going to contain the nuclear waste more or less in perpetuity. It's bold to assume that we'll maintain civilisations capable of doing that for thousands of years. And it's already not going at all well at Sellafield (UK). The latest estimates to deal with it are over half a trillion NZ dollars. There's a huge physical and cyber security risk (e.g. Russia and China have both already hacked Sellafield's systems).
Also, the cost of nuclear power is higher than any other power source, even before factoring in the cost of dealing with waste containment. And SMRs are just non-production concepts at this stage. And they are not 'large-scale', by definition.
SMRs in the real world are few. China & Russia from memory are the only live instances, and they're very few. There are lots of designs, and not a few prototypes. The Google link says:
"Kairos Power’s technology uses a molten-salt cooling system, combined with a ceramic, pebble-type fuel, to efficiently transport heat to a steam turbine to generate power. "
China's (only?) live SMR uses a pebble-bed. (just checked. went live 2021.) But theirs from what I know sounds quite a bit different to Kairos's.
So bleeding-edge A.I. is pushing bleeding-edge nukes? What could possibly go wrong ...
We welcome your comments below. If you are not already registered, please register to comment.
Remember we welcome robust, respectful and insightful debate. We don't welcome abusive or defamatory comments and will de-register those repeatedly making such comments. Our current comment policy is here.