The Cost Analysis and ROI Question
Welcome back! In this second part of the AI Diplomat Insights Series, we dive deep into the ongoing debate: Are we in an AI bubble, or is this the dawn of an AI revolution? We'll analyze expert opinions and break down the financial viability and future of AI development. Let's explore these perspectives and understand the complexities involved.
As highlighted in the Sequoia blog post, the discourse around the cost and return on investment (ROI) for AI is multifaceted. There are two distinct discussions: one about the ROI for companies building massive AI infrastructure, like Microsoft, Meta, and Google, and another about the ROI for businesses using AI. While related, these discussions are not identical, and conflating them oversimplifies the issue.
For infrastructure builders, the ROI question is crucial. These tech giants are investing billions, even trillions, in AI infrastructure, and it's reasonable for investors to question the potential revenue upside of these investments. Some might argue that these companies are overvalued, leading to investment decisions like shorting their stocks. However, the ROI question is more complex due to several factors:
- Timescale Calculus: Wall Street typically focuses on short-term ROI, pricing in the future value of a company today. Conversely, executives investing in AI are thinking decades ahead. This difference in time horizons creates tension and misalignment in evaluating AI's potential.
- AGI Potential: Companies are betting not just on current productivity gains from models like GPT-4, but on future advancements with GPT-6, GPT-7, and beyond. The potential economic opportunities from these advancements justify significant short-term spending, despite appearing excessive now.
- Capacity for Loss: Unlike past bubbles, where leverage and unhealthy financial actors played a significant role, current AI investments are made by financially robust companies. These firms can absorb substantial losses without risking systemic failure, making their investments more sustainable.
Enterprise ROI and Cost-Effectiveness
On the enterprise side, the ROI equation differs. The current prices of AI APIs make this a costly new technology, but businesses are finding ways to optimize and justify these costs. There are several key points to consider:
- Matching Models to Problems: Enterprises are increasingly focused on aligning AI models with specific business needs. This includes using less expensive, non-state-of-the-art models for suitable tasks, balancing cost and performance effectively.
- Batch Processing: Many business tasks don't require immediate AI inference, allowing for cost-saving batch processing. This approach reduces expenses while still leveraging AI capabilities.
- Price Competition: Significant price competition among AI model providers is emerging, driving costs down. In markets like China, price wars are commoditizing AI models, making them more affordable. Globally, we see models like Claude 3.5 Sonnet outperforming at lower costs, exemplifying this trend.
Expert Opinions and Market Analysis
The Goldman Sachs Global Macro Research Report, Issue 129, titled "Gen AI: Too Much Spend, Too Little Benefit?" encapsulates the debate. The report predicts over $1 trillion in AI CapEx spending with uncertain returns, questioning the transformative impact of generative AI. Experts like MIT's Professor Darren Osamoglu and Goldman Sachs' Jim Cabello provide a cautious outlook, highlighting the slow pace of transformative changes and the high costs of AI infrastructure.
Osamoglu estimates that AI will impact less than 5% of tasks in the next decade, boosting U.S. productivity by only 0.5% and GDP growth by 0.9%. Cabello doubts AI's ability to solve complex problems cost-effectively, comparing it unfavorably to revolutionary technologies like the internet.
Conversely, voices like Rohi argue for patience, emphasizing that AI's full potential is yet to be realized.
"I know it's trendy to dismiss generative AI as a bubble, but let's not forget that GPT-4 was released just a year ago. We're still in the early stages, and there's a lot more to unfold."
His call for patience underscores the broader debate about AI's long-term potential versus short-term skepticism.
Adding to this perspective, Joseph Briggs, a senior economist at Goldman Sachs, ironically from the same investment bank as Cabello, published a report in April 2023 highlighting AI's potential for job creation and economic growth. Briggs argues that advances in natural language processing and their integration into businesses and society could drive a 7% (or almost $7 trillion) increase in global GDP and lift productivity growth by 1.5 percentage points over a 10-year period.
The current investments, though substantial, are necessary for long-term gains. Short-term scepticism overlooks the strategic importance of staying competitive in AI development.
Conclusion
From the AI Diplomat editorial room, it's clear that the discourse on AI is enormous and complex, shaped by varying perspectives, biases, and timeframes. While scepticism about AI's immediate returns is valid, dismissing its long-term potential might be premature. The ongoing debate underscores the importance of a nuanced understanding of AI's evolving landscape, balancing immediate challenges with future possibilities.
Stay tuned to the AI Diplomat Insights Series for more in-depth analysis and discussions on the latest developments in AI.
