Here’s what will be covered in today’s newsletter:

Now I am by no means qualified to answer this question, and if history has taught us anything it’s that no one really knows we’re in a bubble until after it pops. However, this is a topic captivating the world, sparking both excitement and concern in equal measure. Personally, I have no idea where AI will ultimately take us, but I do find it fascinating to sift through the arguments guiding both sides of the debate.

A Case for AI Being an Economic Boom:

Let’s start with the optimistic view and why many believe we are in the most important technological revolution of all time.

A case for AI being an enormous bubble:

Now for the flip side, where many voices warn of potentially the biggest economic collapse just threatening to happen.

Is This Going to be the Next Dot-Com Bubble?

Now one of the most common correlations people like to draw is between the rise of Generative AI and the Dot-Com bubble. Similar to the Dot-Com era, the AI revolution is drawing huge amounts of capital injection, enormous valuations and is technology based in potential.

However, there are key differences. During the dot-com era, billions were poured into laying fiber-optic cables and building network capacity in anticipation of internet traffic that didn’t yet exist. Much of that infrastructure went unused for years. By contrast, today’s AI infrastructure, particularly GPUs and high-performance compute capacity, is already under immense demand. In fact, there’s a global shortage of GPUs, with major cloud providers struggling to source enough to meet AI workloads.

Moreover, today’s leading AI firms are often anchored within established tech ecosystems with diversified revenue streams (companies like Microsoft, Amazon, and Google). This makes the sector less fragile than the early-2000s internet boom, when many startups were burning through cash without any sustainable model.

That said, the speculative fervor, especially around startups and model training, still bears resemblance to the late 1990s. As before, the underlying technology is likely transformative, but the timing and valuation expectations may be overly optimistic.

Conclusion:

Ultimately this discussion could go on for much longer with hundreds of more points to advocate for either perspective. However, If I were to hazard a guess, I lean toward believing that AI will deliver real economic uplift for decades to come and that this may be one of the most remarkable technological eras we will ever live through. I also believe, however, that we are riding a wave of excess expectations and overvaluation, especially in terms of how quickly results will arrive. In other words: the boom may well be real but it is most likely being powered by unsustainable finances.

For some more content on this topic (and undoubtedly more comprehensive) I found the following intriguing:

For the more perspective of the optimist perspective, listen to Jensen Huang, CEO of NVIDIA (the world’s most valuable company as of 2025) on the future of compute and AI. His belief in the “industrial revolution of intelligence” offers an interesting counterpoint to the skepticism.

For a deeper understanding of the risks posed by the circular economics of AI, this video provides an excellent breakdown:

For AI to be truly effective the most critical requirement is the access to high-quality, contextual data. It’s not just about millions of data points, but data that’s properly labelled and collected often enough to reflect how the plant actually behaves under different conditions. Luckily, large-scale mining, mineral processing and energy operations are already logging thousands of variables in their control rooms, making them perfect ground for AI to step in.

This week will cover one example of an application for applying AI to flotation circuits; to show how smart systems can build on existing process control to drive measurable value.

Flotation Optimisation:

Modern systems equipped with advanced sensors now generate real-time multivariate data on flotation performance. Typical input and output variables could include:

AI models such as artificial neural networks (ANNs) or reinforcement learning (RL) agents can be trained on this historical and live-streamed data to identify nonlinear relationships between these variables and key performance indicators such as grade, recovery, and energy consumption.

An example application might involve developing an AI-driven optimization function—for instance, a revenue-maximization model that dynamically balances concentrate grade and metal recovery. Once validated, the trained AI model can interface directly with the process control layer of the plant, making real-time setpoint adjustments to reagent dosing, air flow, or froth depth to sustain optimal metallurgical performance.

To ensure reliability, this approach is often deployed within a hybrid control architecture, where AI operates as a supervisory layer on top of traditional process control and that collaborates with process operators.

Implementing such systems requires close collaboration between process engineers, data scientists, and control specialists. Key challenges include:

Whilst this is a new area of development, many companies like Hatch are pioneering this field to move AI beyond proof of concepts into production-scale control environments that represent the future for optimised process control.

This newsletter seeks to engage and challenge the way engineers see AI and its potential for application in industry. Any thoughts, questions or arguments are welcome!