Jul '24 in General discussion
GS Top of Mind – GenAI: Too much spend, too little benefit?
In its June 2024 research report on GenAI Goldman Sach pulls together a variety of macro views of the state of GenAI affairs. This is done against the background of a projected AI capex of over USD1 trillion in the coming years, to be flowing into research of the technology itself, data centres, chips, infrastructure and power grids. Considering these hefty expenditures and the to date seemingly sparce, tangible benefits a debate is brewing among industry experts whether this colossal investment will yield commensurate returns.
The Skeptics
MIT’s Daron Acemoglu and GS’ Jim Covello (Head of Global Equity Research) are among the skeptics. Acemoglu forecasts only a limited economic upside from AI over the next decade, predicting that GenAI will increase US productivity by a mere 0.5% and GDP growth by 0.9%. He argues that AI’s current architecture is not designed to solve complex problems that would justify its high costs. Covello shares this sentiment, emphasizing that AI technology, unlike the internet in its early days, is not only exceptionally expensive, but also lacks the product roadmap that existed at the inception of other technologies and hence AI is not poised to deliver cost-effective solutions to complex problems. Covello also doubts that AI’s costs will decline significantly, making widespread AI-powered automation economically unfeasible.
The Optimists
On the other side of the spectrum, GS’ Joseph Briggs (Global Economics Research), Kash Rangan (US Software Equity Research Analyst), and Eric Sheridan (US Internet Equity Research Analyst) express optimism about AI’s economic potential: They argue that AI could ultimately automate 25% of work tasks, potentially increasing US productivity by 9% and GDP by 6.1% over the next decade. Briggs highlights the historical trend of technological innovations leading to new task creation and labour reallocation, which he believes will apply to AI as well. Rangan and Sheridan note that current AI capex spending (in some cases over 30% of revenue), while substantial, is no outlier compared with previous tech investment cycles.
The Constraints
Despite the optimism, significant constraints loom on the horizon. A lot of attention has been paid to chip shortages. Less focus is placed on a looming power shortage, which could constrain the expansion of data centres and, consequently, AI development. For the US its aging power grid, combined with regulatory and supply chain challenges, poses a formidable barrier to scaling AI. Brian Janous of Cloverleaf Infrastructure sees need to add well in excess of 100 GW of peak capacity to a system that currently handles around 800 GW at peak, while wait times of power projects to connect to the grid are currently 40-70 months.
1
2 comments
GS Top of Mind – GenAI: Too much spend, too little benefit?
skool.com/generativeai
Learn and Master Generative AI, tools and programming with practical applications at work or business. Embrace the future – join us now!
powered by