Learning about machine learning

Written by Chris Fellows

Earlier this year, I attended an event in Bury, put on by Easy Compute, that highlighted opportunities for landowners to add a mobile data centre to their farms. This event caught my attention because, as a company, we spend a significant amount of money each year with Amazon and Microsoft on machine learning and AI processing. All of this processing is currently carried out in their data centres. The idea of having our own data centre on the farm, using our own AI processing power, is very appealing—especially if it can significantly reduce costs.

You might think you’ve never heard of a mobile data centre, but chances are you have. If you’ve ever heard of someone mining cryptocurrency at home, it’s essentially the same technology: a computer with a powerful graphics card capable of processing vast amounts of information. However, instead of using that processing power to solve problems and earn cryptocurrency, you can now rent out your processors to businesses running AI processing and get paid in cash. Alternatively, you can use the processing power for your own AI models, saving money—which is exactly what we’re considering.

Practical Applications for AI

For our business, AI is helping us understand the sentiment behind farmers’ opinions on a wide range of topics. For instance, we can gauge how 20,000 different farmers feel about organisations like the NFU, Red Tractor, or Defra, as well as brands like John Deere. This kind of insight is becoming increasingly important for marketing. Companies want to understand how farmers feel, but it’s impossible to manually read everything that’s written by farmers.

The event, hosted by Easy Compute, aimed to help farmers and landowners understand the technology and its applications in three key areas:

• AI and Machine Learning
• Blockchain
• Cryptocurrency

Power Requirements

The primary input for a mobile data centre is power—a lot of it. Having a good electricity contract, or better still, generating your own power via solar or wind, is the ideal starting point.

A mobile data centre consumes power in several ways:

1. Computing Equipment (Servers and GPUs): The servers, processors, and specialised hardware like GPUs are the biggest power consumers, handling high-intensity calculations and data processing.
2. Cooling Systems: Keeping the hardware at optimal temperatures is essential. If you only have a single machine, it can happily warm your office without additional cooling!
3. Networking Equipment: Devices like routers, switches, and firewalls require electricity to keep the servers connected to the internet.
4. Lighting and Security Systems: Though a smaller expense, lighting, surveillance cameras, and other security measures still require some power, particularly for 24/7 operations.

Depending on your setup, profitability is still achievable even with electricity costs of 35p per kWh. However, costs drop significantly if you have a better contract or generate your own renewable energy, such as solar or wind power, where rates can be as low as 7p per kWh. For farms already selling power to the grid, using it to operate a data centre could be far more lucrative.

Making Use of Heat

The primary “waste” output from a data centre is heat—but that’s only waste if you can’t use it. A single computer in your office generates the equivalent heat of a 2kW heater. Twenty rigs in a container will produce a considerable amount of heat, which farms can repurpose:

• Drying or Maintaining Grain: Heat from the data centre can be used to dry grain over time, reducing the need for traditional grain dryers.
• Heating Sheds or Greenhouses: Chicken sheds, grow houses, or other heated spaces can utilise this waste heat.
• Slurry and Methane Capture: Combining a data centre with systems like the Bennamann methane capture system is highly efficient. Heating the slurry generates even more methane, which can then be used to feed the generator that powers the data centre itself. This means operating a data centre would be basically free, using the methane reduces a farms carbon footprint as well.  However, I’m not sure most people think of dairy farms as the “perfect” place to have a data centre! But I think they just might be.

Diversification

We all know that multiple revenue streams for farming businesses reduces overall risk. Whether that be different crops, agri-tourism, AD or renewable energy. Mobile data centres represent a low-maintenance diversification option. Unlike holiday lets, you don’t need extra staff, cleaning services, or marketing efforts. This solution works particularly well alongside renewable energy.

While this technology has been used extensively in crypto currency mining historically, this would not be the main focus for me. Crypto mining relies heavily on the price of the currencies at any given time. It builds in more risk (although prices just keep going up).  Some people also have ethical issues with using power from the grid to produce a crypto currency, but there are plenty of companies that have made significant sums from coins like Bitcoin.

I remember stating proudly in 2013 that buying bitcoin was madness, at $8 a coin. Which clears up that I have made some bad decisions when it comes to crypto in the past. A £1000 investment in 2013 would now be worth £11 million today and you would be missing out on my writing skills. 

A Mobile Data Centre is like an Airbnb

During the event I quite liked this analogy.  It’s something many farmers are familiar with, having properties, tents or caravans to rent on their farms.  If you have a machine with 5 processors in it.  You effectively have 5 properties to let.  You can let them short term for a higher daily price or long term to a cheaper daily rate. But you aim is always to keep you occupancy as high a possible to get most reward from your property.  Having a data centre on your farm is just the same.  You want to keep it rented out as much as possible.

And in the same way Cottages for You will rent out your holiday cottages for you, there are websites lie Vast.ai and RunPod that help you rent out your processors for.  Companies like Easy Compute can help you with this process.

Also, when your processors aren’t being rented out for AI, you could use the spare time to generate some crypto currency.  If you are happy with the risk, you could still choose to do this. With the reward you receive being variable based on the price of the coin.  Some people are suggesting the price of bitcoin to rise to $250,000 while others are saying it will be worthless. While the total value of bitcoin in circulation, is worth more than the total value of the silver market, would you feel more comfortable owning bitcoin or silver?

Bitcoin price for last 3 years in USD

The Costs

With any diversification, it is all about the costs.  Investment versus reward.  Is the waste heat worth anything to you? What does you electricity cost? When considering a bigger mobile data centre or even a single rig with 4 processors, you need to consider it within the context of your farm, your power costs and your heating needs.

A single rig uses about 1000 units of electric a month. If your electric cost 12p per unit, a 4-processor rig will cost around £120 a month in electric.  This figure needs to be adjusted then for exactly what you pay, be this the price you pay or the price you export at. This could be as high a £240 a month or as low as £50 a month.

You set an hourly rate per processor to rent it. A typical rate is £0.40 GBP per hour. Ideally you then want to rent out 70-90% of the time. This generates between £800 – £950 a month in income. But like renting property, the more effort you put in, the better rental fee’s you can receive.  Consider these figures a baseline.

In this example you would generate between of £560 – £880 a month. Quite a big difference. But if you can also use the heat, that might save you some money somewhere else, so there may be other factors to consider.

The Hardware

If you have been thinking, “this sounds amazing, what’s the catch”  well we also need to discuss the cost of the hardware in the first place and the ongoing support to maintain the system.   A single 4 processor system on a fully maintained basis (which is important if you are new to the sector) will cost circa £25,000.  The good news is that is fully tax deductible, so can be offset against any tax bill. 

Also based on the figures above, a system over 10 years would conservatively generate AI processing revenues between £67,200 and £105,600.  There would of course be economies of scale on bigger data centres and the value of the heat has not been factored in.  

The Future

Demand for AI processing is rapidly increasing and is projected to grow over the next decade. If you’re looking for a forward-thinking diversification strategy, a mobile data centre could be the perfect addition to your farm.