“CPUs typically have between eight and 64 cores optimized for performing sequential computing tasks. GPUs can host thousands of smaller cores that operate in parallel to handle massive amounts of data in a shorter time period…. They can run through vector calculations, image and video processing tasks, and data transformations all at once, which is something a CPU couldn’t match” ServerLift, 25 Aug. 2025).
“xAI is actively advancing its second supercomputer cluster, Colossus 2, located at a new facility on Tulane Road in South Memphis. The company has begun installing computing infrastructure, with plans to deploy an initial batch of 550,000 Nvidia GB200 and GB300 GPUs, expected to start coming online in the coming weeks. [Graphics Processing Units can ‘perform thousands of smaller, independent calculations simultaneously, significantly accelerating data-intensive AI workloads'” -Google AI.] This phase includes an additional 110,000 Nvidia GB200 GPUs, powered by a mix of energy sources, including Tesla Megapacks for stable power supply during peak demand. The goal is to scale Colossus 2 to a minimum of 1 million GPUs by the end of 2025 or early 2026, potentially making it the first gigawatt AI training cluster” (Grok, 5 Sep. 2025).
“AI training requires specialized hardware and this is driving dramatic changes. Where cutting-edge systems featured eight GPU accelerators per server just two years ago, by 2027 leading configurations will pack 576 GPUs into filing-cabinet-sized racks consuming 600 kilowatts – enough to power 500 US homes, Goldman Sachs notes” (The Register, 2 Sep. 2025).
“‘If artificial general intelligence is created then there’s going to be artificial super intelligence. And people who have that can control conventional, non-conventional warfare,’ said Rich DiClaudio, president and CEO of the Energy Innovation Center Institute, a nonprofit dedicated to training people in Western Pennsylvania and beyond. ‘I don’t know if that will happen but two countries are saying if that’s going to happen, we need to be in charge of it. And those countries are the U.S. and China. So now they’re off to the races'” (Pittsburgh’s Public Source, 2 Sep. 2025).
“There were 5,426 data centers nationally as of March 2025, and the number is skyrocketing” (Earth911, 2 Sep. 2025).
“AI data centers are specifically designed to support resource-intensive AI workloads. They provide the infrastructure necessary for processing, training, deploying, and continually running complex machine learning (ML) algorithms, large language models (LLMs), and, eventually, autonomous AI agents…. [They] feature more advanced graphics processing units (GPU) (popularized by chip manufacturer Nvidia), tensor processing units (TPUs) (developed by Google), and other specialized accelerators and equipment…. [They] handle huge amounts of unstructured data including text, images, video, audio, and other files. They also incorporate high-performance tools including parallel file systems, multiple network servers, and NVMe solid state drives (SSDs)” (Network World, 4 Sep. 2025).
“Luzerne County [Pennsylvania] Community College President John Yudichak noted that community colleges across the state are ‘collaborating and combining resources like never before to build a new and historic statewide technology and trades workforce consortium’ to address a workforce gap identified in a 2024 report by the state Department of Education. ‘We have an acute workforce shortage by 2032 of 218,000 Pennsylvania workers who will lack post-secondary credentials and skills to meet the demands of the job market,’ Yudichak said. ‘Currently there’s a skills gap of 12,200 workers in the trade and maintenance workforce cluster. [That] gap will grow exponentially if unaddressed as data center development grows in Pennsylvania'” (Pittsburgh’s Public Source, 2 Sep. 2025.
“The University of Virginia, Brightpoint Community College, and Northern Virginia Community College are among the nation’s universities that are part of the first cohort of the Google AI for Education Accelerator, giving students, faculty, and staff no-cost access to Google Career Certificates and AI training courses” (Data Center Knowledge, 2 Sep. 2025).
“‘You should expect OpenAI to spend trillions of dollars’ on data center construction in the ‘not very distant future,’ OpenAI CEO Sam Altman told a group of reporters on Thursday” (Bloomberg News, 15 Aug. 2025).
“Northern Virginia’s energy-hungry data center juggernaut shoulders roughly 70% of the world’s internet traffic. Behind that buildout is an army of blue-collar electrical workers. ‘Forty-five to 70% of construction of a data center is electrical>,’ said Joe Dabbs, the former business manager for the International Brotherhood of Electrical Workers Local 26, the union representing 12,500 electricians in Washington, D.C., Maryland, and Virginia. ‘The teledata portion, the fiber optic portion, the power and power distribution … all the gear work — that’s us. We build all of that stuff.’ Electricians are also needed for maintenance and changing out equipment, Dabbs said, meaning the local union has ‘people in data centers around the clock, 24/7.’ And the demand for workers keeps rolling in — making it a good time to be an electrician in America, where experienced workers can earn over six figures” (Yahoo!Finance, 7 Sep. 2025).
“TPUs, NPUs and DPUs: AI-ready data centers increasingly incorporate more specialized accelerators specifically built for AI workloads. These include tensor processing Units (TPUs), neural processing units (NPUs), and data processing units (DPUs). TPUs speed up tensor computations, or multi-dimensional data structures, so that AI models can process complex data and perform calculations…. NPUs mimic the neural pathways of the brain, allowing for processing of AI workloads in real time…. DPUs offload and speed up networking, storage, and security functions, freeing up CPUs and GPUs to focus on AI tasks” (Network World, 4 Sep. 2025).
“CXL is not just a technological advancement. It’s a paradigm shift. By decoupling memory from CPU sockets, CXL allows data centers to integrate memory pooling, sharing, and dynamic allocation capabilities. This innovation resolves critical bottlenecks and sets the stage for scalable infrastructure ready to handle AI’s complex demands…. CXL enables centralized memory resources to be pooled and shared across multiple devices…. CXL…allow[s] memory to be scaled independently and cost-effectively without adding more physical servers…. CXL’s low-latency architecture ensures smooth communication between devices…. By enabling smarter memory allocation, data centers powered by CXL consume less electricity overall” (DataCenter Knowledge, 3 Sep. 2025).
“Cloud and data center transformation critically depends on the evolution of the network. The network is essential for connecting distributed data centers that guarantee data sovereignty and real-world AI performance. To support enterprise and consumer use cases, AI needs a network that can support the proliferation of data centers, ensure mission-critical security and reliability, and deliver hyperscale performance” (Nokia, accessed 7 Sep. 2025).
Filed under: ETC Wrap |
Educational Technology and Change Journal 
AGI has multiple definitions, and most are somewhat ambiguous.
I hope that improved AI software will enhance my life in various ways. Where I am now, I don’t expect it to make my life worse. I am not concerned about AI systems taking over the world or eliminating humans.
I am concerned about the energy requirements of Colossus. Typical Musk. Touting energy efficiency while building an energy hog of Brobdingnagian proportions.
I look forward to a better world, if I live long enough. There will be pain during the transition. Enlightened governments will help their citizens through this period. I hope the US government will soon find enlightenment.
Hi, Harry. You’re right about AGI. It seems to mean different things to different people. But I believe the simplest definition might help to keep most of us on course. Artificial General Intelligence is an all-encompassing term that defines a goal rather than piecemeal attainments. It establishes a point in time, a finish line, when we have difficulty distinguising a bot from a human being in conversations.
Re the environmental toll of massive data centers — I think it’s just a matter of time, a price for progress, that we might need to pay. It’s a temporary issue that we’ll be working at mitigating even as we move forward. In fact, some of the most innovative work that’s currently going on behind the scenes is exactly that — strategies to systematically reduce the environmental footprints of data centers. -js
P.S. See Innovations to Reduce Data Center Environmental Footprints (Sep. 2025)