AI industries and industries: Addressing industrial challenges to commercialize AI

This article is part of a VentureBeat special, “AI at Scale: From Vision to Viability.” Read more about this story here.

If you were to travel 60 years back to Stevenson, Alabama, you would find the Widows Creek Fossil Plant, a 1.6 gigawatt plant with the longest chimney in the world. Today, there is a Google data center where the Widows Creek plant once stood. Instead of using coal, old transmission lines bring renewable energy companies online.

This transformation, from the fireplace to the digital factory, is a sign of the global shift to technology. And we’re about to see intelligent manufacturing skyrocket thanks to industrial AI.

These data centers are decision-making engines that use computers, networks and storage to turn data into information. A place full of data is stability in the computation time satisfying the endless demand for artificial intelligence.

AI-enabled architectures take on many of the challenges that define the industry, from power to scalability and reliability, which require innovative solutions to old problems.

New employees: Power calculations

In the age of steam and steel, workers worked day and night. In today’s AI factories, output is determined by computing power. Training large-scale AI models requires large-scale software development. According to Aparna Ramani, VP of engineering at Trimthe size of the studies of these samples is about a four things a year across the industry.

This increase is on track to create barriers that existed in the industrialized world. There are supply chain constraints, for starters. GPUs – the engines of AI transformation – come from a few manufacturers. It’s very difficult. They are in great demand. And so it should not be surprising that they are subject to price fluctuations.

To avoid some of the limitations, big names like AWS, Google, IBM, Intel and Meta are developing their own silicon. These chips are optimized for power, performance and cost, making them professionals with unique features in their work.

This update isn’t just about hardware, though. There is also concern about how AI technologies will affect the job market. Research published by Columbia Business School studied investment management companies and found that the implementation of AI leads to a 5% decrease in the share of employees, reflecting the changes observed during the Industrial Revolution.

Professor Laura Veldkamp, ​​one of the authors of the paper, said: “AI can be a game-changer for many, if not all, of the economy. “I am optimistic that we will find jobs that help many people. But there will be a transfer fee. “

Where will we find the power to expand?

Cost and availability aside, GPUs that act as AI factories are notoriously power-hungry. When the xAI team brought their Colossus supercomputer online in September 2024, they had access to between seven and eight megawatts from the Tennessee Valley Authority. But 100,000 H100 GPU clusters require more than that. Therefore, xAI introduced VoltaGrid mobile generators to make up the temporary difference. In early November, Memphis Light, Gas & Water partnered with TVA to provide xAI with an additional 150 megawatts of power. But critics argue that the use of the site is disrupting the community and polluting its air quality. It’s Elon Musk they already have plans for 100,000 H100/H200 GPUs under the same roof.

According to McKinseyThe energy needs of data centers are expected to increase to nearly three times the current capacity by the end of the decade. At the same time, the number of processors is increasing and their performance is decreasing. This means that the performance per watt is still going up, but at a lower speed, and not fast enough to keep up with the horsepower requirement.

So, what will it take to keep up with the heat of AI technology? A report from Goldman Sachs suggests that US corporations need to invest $50 billion in next-generation data center infrastructure. Researchers also expect that data center energy use will drive an estimated 3.3 billion cubic feet per day of new natural gas demand by 2030.

Growth becomes difficult as the AI ​​industry expands

Training models that make industrial AI more accurate and efficient can take thousands of GPUs, all working together, for months at a time. If the GPU fails during training, the run should be stopped, returned to the search area as soon as possible and restarted. However, as the complexity of AI industries increases, so does the chance of failure. Ramani addressed this concern during the Demonstration of AI Infra @ Scale.

“Stopping and restarting is very painful. But it gets worse because, as the number of GPUs increases, so does the chance of failure. And sometimes, the number of failures can increase so much that we waste more time in reducing these failures and you don’t complete the course.”

According to Ramani, Meta is working on long-term projects to detect failures sooner and get back up and running faster. Furthermore, research involving asynchronous training can increase fault tolerance while simultaneously optimizing GPU utilization and distributing training runs across multiple data centers.

Always on AI will change the way we do business

Just as old industries relied on new technologies and organizational forms to create products, AI industries consume computing power, network resources and storage to create tokens – the smallest piece of information that the AI ​​model uses.

“This AI factory is creating, creating, creating something valuable, something new,” Nvidia CEO Jensen Huang said at the time. Computex 2024 highlights. “It’s easy in almost all industries. That’s why it’s the new Industrial Revolution.”

McKinsey says artificial AI they have more potential That equates to $2.6 to $4.4 trillion in annual revenue across 63 different industries. For each application, whether the AI ​​factory is in the cloud, installed at the edge or autonomous, the same challenges must be overcome, just like the industrial factory. According to the same McKinsey report, achieving even a quarter of this growth by the end of the decade will require another 50 to 60 gigawatts of data center capacity, to begin with.

But the consequences of this growth are about to change the IT industry continuously. Huang explained that the AI ​​industry will enable the IT industry to create a $100 trillion business intelligence. “This will be the manufacturing industry. Not the computer manufacturing industry, but the use of computers in manufacturing. This has never been done before. The most amazing thing.”