In 15-20 years, Intel will likely be the highest valued company in the world. Hear me out
1) AI isn't going anywhere. It is being widely adopted in many industries and is only improving in sophistication and application areas. This trend will only continue. We have been here before with the Internet, I remember the skepticism about the Internet being a fad and having little use for regular people very well
2) AI data centers are unsustainable. We all know this. The problem is that our conventional AI models will always require more and more power given their synchronous mode of computation. Put simply, our neural nets that run on GPUs perform many useless or redundant computations. Pruning/trimming parameters from these networks will not address this problem of scalability. We are nearing the ceiling already
3) The only way forwards is neuromorphic. Neuromorphic computing is a non-conventional method of asynchronous computation inspired by biological processes. Our brains and nervous systems are incredibly low power systems as energy is only consumed when computation occurs in a biological neuron. Neuromorphic computing seeks to replicate this by using asynchronous processing and neuron/synapse/learning rule models closely replicating the processes in mammalian bodies. We are talking milliwatt-scale levels of power consumption for some tasks
4) Neuromorphic computing is fundamentally different to conventional Von Neumann computing. We have to learn and develop these systems into maturity. It is an emerging technology that is rapidly approaching maturity
I'm in neuromorphic research for work and just attended a conference. Here are some of the things I saw:
\- A fully working, extreme low-power LLM built on a spiking transformer network. This is one of the biggest breakthroughs in the technology I've seen since I entered the space professionally. The structure of the spiking transformer network was also naturally mapped to structures in the brain. I cannot emphasise the magnitude of this enough
\- Biologically-integrated sensors for autonomous medical emergency detection before the emergency can occur. Again, extremely low-power, implantable, small, adaptable to the body's signals
\- Applications towards autonomous mapping, image processing, security, space... Anywhere you might want to deploy an intelligent, adaptive device for long periods without interference
But what about quantum/other emerging approaches to computation? Well, for AI there is simply no more suitable framework for its implementation than neuromorphic given its direct inspiration from mammalian nervous systems and brains. We look at the best example of a computer in nature and we implement it in hardware and software. Quantum is it's own thing and will tackle different problems
Risks? Photonics. Right now, the power bottleneck for neuromorphic devices is the splitting and multiplexing of electrical signal lines. Splitting the signal paths for a laser takes no resources. Photonics could be used in neuromorphic chips and it's possible that a company other than Intel gets there first. It is also possible that Intel collaborates with photonics firms to improve their line of neuromorphic boards (Loihi, currently available to researchers only). By the way, Loihi 3 hasn't been announced officially but is in development. Intel are continuing to invest in the technology, and researchers continue to break through conceptual barriers using it
TL;DR: if you've got 10-20 years to hold whilst neuromorphic technology reaches maturity, buy the fuck out of Intel. Buy buy buy. It's gonna go absolutely dambusters. It's near an ATH? So was Sandisk in Nov '25. Means absolutely dick if the stock and company is solid
Happy to answer any questions about the industry as best as I can