How vulnerable is Ukraine to AI Chip Shortages?

by Donald Cook

NOTE: My South Carolina public school education is maxed-out trying to wrap my head around this topic. I have spent over a month reading and watching videos trying to figure out the world around AI and AI chips. Some of what I write might be wrong. Some of it will be outdated. The general observations and take-aways will be strong enough, however, to endure some errors and omissions on my part.

***

How will Ukraine use AI?

At war, the issue of using AI/Machine Learning (ML) isn’t just about delivering accurate ordnance. It can do so much more. It’s important to note that Ukraine has been using AI/ML since the beginning of the full scale war. The US government and Palintir gave Ukraine a light version of Project Maven. 

Project Maven is designed to process imagery and full-motion video from drones and automatically detect potential targets. (Generally, Artificial Intelligence and Machine Learning can be used interchangeably. ML is a bit more specific than AI.)

Here are some Ukrainian military uses for AI you might not think about. Maximizing administrative tasks. Like invoices and payments. Maximizing logistics. AI could decrease the time it takes weapons and gear to get to the front. Creating pro-Ukraine propaganda. You’d be crazy to think this isn’t happening already. Russia certainly does it. Creating predictive models of Russian movements and attacks. AI is just a very complicated probability/prediction engine.

The use case most thought of when we imagine AI in Ukraine is of course UAV target acquisition over long distances or in areas where communication is being denied by Electronic Warfare (EW).

What kind of chips does Ukraine need?

Ukraine needs high-end and consumer tier AI computer chips. There are many kinds of AI chips. In all likelihood you already have at least one. Maybe more. Your new phone is an AI Edge Computing device. If your computer has a GPU, that is also considered by many to be an AI chip. 

AI chips differ from your normal CPU processor in many ways. The main difference is parallel processing. Your computer CPU processes information one math problem at a time. Where AI chips process multiple. GPUs differ from TRUE AI chips in that there is additional functionality built into the AI chip to process even faster. As well, a true AI chip runs more efficiently. A consumer level GPU array is huge. It has multiple fans to keep it cool and uses a LOT of electricity. Think Bitcoin mining and data farms.  A consumer level AI chip is much smaller. It normally needs only a heat sink, depending on the application and device. Without going full nerd, just know that the smaller/closer the transistors are etched onto the chip, the more efficient it is, the less electricity it uses, which means less heat generation.

Like in your phone, Ukraine will need to use AI driven Edge Computing to target distant objects or operate in an EW contested environment. Latency in communicating over long distances and EW mean that everything that a person usually does in acquiring and targeting the enemy must now be done onboard the UAV, on the Edge.  Edge computing is just another buzz word for computing done outside “the cloud”.  In a consumer environment that means it’s done on the user’s device. Like your phone. Or an IoT (Internet of Things) appliance in your home. Perhaps a smart fridge.

Specifically for military purposes this means that the information detected by the UAV will be processed onboard. It means that the data set necessary to obtain the target, the data set necessary to track the target, the data set necessary to strike the target all reside onboard the UAV. All of this information is processed at or near TRILLIONS of times per second by the AI chip. It can make these decisions because it was trained to. Likely, 100,000s of thousands of military images and 100s of millions of environmental images were used to create a model. This model can recognize an object, the object at any angle, a partially occluded object. It can gauge distance, closing speed, and the direction the object is traveling. It can do this because the photo/video/data has been “Annotated”.  Annotation in AI is the process of labeling your data. Think of it as metadata. Rather than a human neural network learning this information and developing long term memory and hand eye coordination skills a Machine Learning Convolutional Neural Network is used.

As I’ve already said, AI chips are special. Not just in what they can do. But also in how they are designed and manufactured.  Many companies design AI chips. Apple, Samsung, NVIDIA, META, Intel, TSMC.  All the chips are manufactured by one company.  In one place. Taiwan. GPU chips are made all over the world. Including China. But true AI chips are made in one hotly contested constantly under threat place. A tiny Island only 100 miles/160km from mainland China.

NOTE: A lot of chips are called AI chips because they do dual processing. Dual processing is a necessary but insufficient capability of an AI chip. For the speeds necessary to do demanding AI computations well the chips should be 8 nanometers or smaller. China currently can only produce 10 nanometer chips at scale. 20 nanometers is sufficient for some Edge Computing. But model training with massive data sets in the millions of data points requires either time, or true AI chipsets.

What does all of this mean?

The limiting factors of UAV size, weight, fuel/battery, and the necessity of Edge computing dictate that Ukraine use true AI chips in its UAVs. These chips are manufactured in one place. This limits Ukraine’s options for acquiring these chipsets in the numbers and in the timeframes necessary to deploy it effectively if China and Taiwan become embroiled in a hot war.

Possible solutions?

The US is trying to bring 8 nanometer and smaller chip manufacturing to the USA. But it’s not coming soon. In the short term, Ukraine should stockpile AI chipsets that it needs not only to deploy weapons but to also train models. Long term, Ukraine should use its current strategic advantage and become vertically integrated into the AI chip supply chain: design, training, and eventually manufacture. This is a big ask. It would not be easy. This would provide Ukraine not just an economic benefit because it is involved in the whole process of bringing AI chips to market.  It would also put Ukraine in a dominant position for the silicone/chip boom. Processing power/chips are the new “energy” commodity. What oil and gas are today (indispensable) chips will be tomorrow. Building chip capacity means you will get special status in geopolitics. Just like oil and gas producing nations enjoy now. 

Benjamin Cook

https://researchingukraine.substack.com

www.bpcusc.com