The report was the latest take from Bain on AI’s moment, an AI boom that is at its highest investment levels ever. This influx of capital is a reminder that AI companies have to move quickly in what is a fast-changing and fluid environment. Switzerland has been buzzing this week as the country prepares to roll out its own, an LLM similar to ChatGPT. This cutting-edge model, which can effectively speak in 1,000 languages and dialects, was built on its own infrastructure. Development of the model cost around $US50 million, showing the financial commitment at stake in this tech race.
Over the last several years, AI models have rapidly increased in size and complexity. OpenAI’s GPT-3, released in June 2020, had 1.75 billion parameters. Its successor, GPT-4, released in March 2023, left it far behind with an estimated 1.5 trillion parameters. The extremely fast pace in which AI capabilities are advancing dramatically increases the stakes of what happens next. Industry experts are already estimating that OpenAI’s next model, GPT-5, which is expected to be released in 2025, will require 10x the capital to train as its predecessor.
The Financial Burden of AI Expansion
The pace of AI technology’s growth requires a massive scaleup in global energy provision and the infrastructure to deliver it. A recent study indicates that the global electricity supply must grow by 20 percent by 2030 to meet the demands of expanding data centers. In the United States—the largest single market for data centers—new data centers will have needed about $US500 billion per year in capital investment.
The top 10 AI companies will need to make an additional $US2 trillion in yearly revenue just to support their growth. This revenue is absolutely critical to offset capital expenditures on their data centers. And fellow resident AI company OpenAI is constructing its own Stargate AI data center in Texas. The massive project will cost an estimated $400 billion and will be able to supply 7 gigawatts of energy. This ambitious project sheds light on the financial peril companies of all sizes are up against as they attempt to grow their businesses.
Nvidia has already committed to investing $100 billion into OpenAI. This investment will make millions of AI chips, underscoring the intricate corporate collaborations that drive innovation in AI hardware. These investments are necessary for maintaining the status quo. Beyond that, they create the opportunity to craft transformative, new models that can completely upend whole industries.
Australia’s Strategic Position in the AI Landscape
Australia is definitely making moves to set themselves up to benefit from this AI boom. Sovereign Australia AI has ambitions to build its own LLM. It will include between 700 billion and 1 trillion connections, and they expect the price tag to get up to $100 million. On top of that, Maincode plans to release Australia’s first sovereign LLM, called Matilda, later this year.
Open-source LLMs have proven to be a game changer. They provide opportunities for countries such as Australia to develop their own AI capabilities, lessening reliance on foreign technologies. As Davis observes, “You can accomplish a tremendous amount of work just by taking one of the pre-trained leading open-source models and doing fine-tuning. This strategy allows nations to leverage existing resources while tailoring solutions that meet specific local needs.
Simon Kriss emphasizes the importance of practical applications: “What we need is just a simple large language model that will handle all the text-based tasks that we have. All the chatbots and transcription services.” The focus on functional applications rather than merely increasing model complexity could redefine how AI is integrated into various sectors.
The Future of AI Models and Their Applications
As the conversation on AI improves, some experts in the industry are challenging the notion that bigger models are better. Nicholas Davis illuminates a frequently held misbelief. He makes the case that no one should take for granted that the future will be based on bigger and more complex models. This sense of restraint belies a new spirit of realism, a realization that efficiency and common sense need to drive the next iterations of AI innovation.
Dave Lemphers highlights a prevalent trend among major AI companies: “The prevailing wisdom for these [big AI] companies has been to make the model bigger. You just have to throw more [computing power] at it.” The focus on raw size may overlook other innovative approaches that could yield equally effective results without exorbitant costs.
As Hoecker notes, “AI is moving at a pace quicker than we’ve experienced any other technology transition to be honest. The fast-paced developments suggest a critical juncture where companies must balance ambition with practicality.
“What are the practical use cases?” – Hoecker
