Top Disruptive Tech in the Next 5 Years
Most Disruptive Technology
I think Agentic AI is the most disruptive technological shift underway today and is one whose impact on society is still significantly underestimated.
Looking at previous computing inflection waves like the cloud, smartphones, PCs, the internet; they all share common traits: you have increased productivity, faster time to market and cost savings. However, this new wave of agentic AI is completely different as we are getting to the point where you can replace a human from end to end… You can replace a software engineer or a data scientist and that disrupts everything. We're starting to see Fortune 100 companies deploying 50k agentic LLMs to map out their supply chain and tell them how they can minimize costs & disruption. Instead of having 20 workers in that department, you only need 3, for example. Hence to get an idea as to how big this market is, you can look at employee revenues in the US alone which is $10T. The harder part of course is determining what percentage could be replaced, but any number would still lead to significant disruption.
Where Does Value Accrue?
A useful way to think about this opportunity is to analyze where value is likely to accrue?
At the infrastructure layer, the demand for compute, networking and memory continues to grow. Agentic AI relies on chain-of-thought techniques which require significantly more compute. And compute is important but we've seen over the past year that there are massive bottlenecks in networking and memory bandwidth which is constraining the performance of these same GPUs. To put some rough numbers on this, performance of GPUs has increased 50x in recent years but the performance of networking & memory have only increased by 3/4x. As a result, you're paying $35k per chip to have MFUs (best metric to measure the efficiency of LLMs) of 30-50%. That's not sustainable. Hence why it wouldn't be surprising to see networking outgrow GPUs over the next years.
The companies responsible for easing this constraint are the optical transceiver providers such as AAOI, COHR and LITE. There are also power limitations from building these clusters in a single location so you could interconnect different data centers to build one large cluster and that's stocks like CIEN or NOK (through the Infinera acquisition).
On another level, I also think we're moving from a world where we don't have enough compute, can't get access to GPUs to a world where we won't be supply constrained in GPUs.. This is a typical pattern in the semiconductor cycle driven by boom and busts: demand driven chips lead to oversupply and market shifts from scarcity to oversupply. Nvidia is not exempt from this, it is still a cyclical business despite the secular AI tailwinds.
And the answer for when that world does happen lies in the application layer. More precisely, I think we will find a lot of value in companies which have high quality data, high barriers to entry and can use this data to create value for its customers. Foundational models are trained on public data but there is still a lot of very valuable data sitting within the enterprise and that will be the crucial differentiator in creating value.
You can find these companies in many sectors such as healthcare, industrials, advertising…
Meta is an example. Their objective is to lower the marginal cost of content creation, enhance the personalized feedback loop and keep users engaged for as long as possible. Content formats will also evolve which will help as we've gone from pictures to videos and soon there will be interactive content created by AI, video games on Instagram reels… All that to keep users longer on their platform.
Agricultural companies like John Deere are also very interesting as they have evolved to become a leader in data-driven agriculture. Their equipment collects structured data on soil composition, crop yields, and machine telemetry and uses Agentic AI to make predictions which drives higher yields and lower input costs for farmers. At the same time they're also selling higher-margin software subscriptions, creating higher switching costs which helps their revenues & margins.
Investors are massively underestimating the impact agentic AI will have on society and especially how early we are in the adoption curve.