Table of Contents
Vish Nandlall frames paradoxical AI trends around ambition, competition, cost and measurement
While the direction of travel in the world of AI is still firmly toward hundreds of billions of investments into the underlying infrastructure, if you start to peel the onion, a number of paradoxes begin to emerge. In this episode of Signal & Noise, Vish Nandlall, formerly of Dell Technologies and currently an independent technology and strategy advisor, talks through these paradoxes and what they may mean for the trajectory of AI.
The four paradoxes Nandlall lays out are around ambition, competition, cost and measurement. One at a time now:
- Ambition: enterprises and governments are eager to derive efficiency and productivity gains through the use of generative and other forms of AI. However, and particularly in a governmental setting, implementation is difficult. Problems generally include moving from pilot to production and, more specific to federal use of AI, governance, policy and enabling the right user to access the right data, and then turn that into a useful workflow.
- Competiton: Oracle has been on a rip as it expand its business from structured databases and cloud infrastructure to standing up compute capacity for the likes of OpenAI. More broadly, we’re seeing the rise of the neoclouds who are rushing to deliver capacity for hyperscalers. But is that a sustainable business or a short-term sprint to an acquisition?
- Cost: Training costs are following the AI scaling laws – processing more data for bigger models means more compute and more cost. But is bigger really better? And how should we think about the rising cost of training juxtaposed against the plummeting cost of inference which is what the end user, who is ideally paying for that inference output, actually experiences. No one puts and query into ChatGPT and says, “Thank god they have access to so many Grace Blackwell racks.
- Measurement: This one also gets around user experience. Given the pace of development, are existing AI benchmarking tools a bit untethered from what actually matters to the people paying the bills?
Check out the full interview here. Spoiler alert: it certainly includes conversation of these four paradoxes, but it also has lots of other thought-provoking conversation.
As Nandlall put it in a LinkedIn post: “These aren’t just news items. They are seismic shifts. We’ve moved past the ‘magic’ of AI and into the brutal economics of an industrial-scale utility. The game is no longer just about having the best algorithm, but also about having the cheapest power, the smartest logistics and the savviest partnerships.”
