15 Feb 2022

Data centres and AI Start-ups. Start as you mean to go on.

Despite the pandemic, organisations engaged in artificial intelligence (AI) contributed more than £15 billion to the UK economy in 2020, and it’s predicted that AI alone could help UK GDP grow 10.3% by 2030. AI is big business, it’s getting bigger all the time and the UK is absolutely leading the charge with a comprehensive national strategy to boot. AI start-ups across the UK have grown 145% since 2010.

Supporting this meteoric growth with the right infrastructure is not only important for the start-ups involved, it’s important for all of us. Start-ups are the lifeblood of the economy. Most AI start-ups begin life as a brilliant idea after long hours of intense concentration, gallons of strong coffee, and very often culminate in a ‘free-trial’ slot in one of the popular public clouds.

It’s a tried and tested recipe. Going with cloud infrastructure instead of on-premises data management seems to make sense – instant access, initial low cost, seemingly infinite growth, a virtual sandpit for testing and trying out ideas and no need to worry about underlying infrastructure and hosting; and all those boring things you don’t want to be thinking about when you’ve just had that lightbulb moment.

However, what starts out as a great idea, doesn’t always look so rosy in the longer term as organisations scale from prototype towards production – with cloud pay-per-use costs spiralling exponentially, data egress being expensive and the ability to fine-tune infrastructure in the cloud being arduous. Add to that, the virtual support-ticket structure when things don’t go right – and they don’t always go right – you can end up wanting to be able to see, hear and touch your infrastructure and potentially press that big, red, reset button.

There has already been much discussion about the merits of AI businesses moving away from cloud and into colocation data centres, or using a hybrid of the two, and the pros and cons of this move have been widely debated within the AI community. But, with the bulk of new AI start-ups exhibiting a need for extreme agility with heavy data workloads, plus the certainty of a high level of control over their data management environments, a different question is now being asked by entrepreneurs: ‘Instead of beginning with cloud and then suffering as we scale, why don’t we cut out the middleman and go to colocation from the start’?

Sure, some start-ups may not have the cash to invest in GPU-accelerated infrastructure or an under-the-desk DGX workstation straight away, but the fact is, data centres are no longer only for the big boys working to hyper or industrial scale. It is now financially feasible for many nascent AI start-ups to begin their journey in a colocation environment - with bare-metal systems specifically designed for HPC and GPU-accelerated AI, machine learning, financial grid computing and the potential for unhindered scale built-in.

The idea of beginning operations with their compute clustered in one colocation space instead of relying on sub-optimal on-premises systems or using scattered servers in a virtualised cloud network, is showing signs of acceptance with a growing number of AI start-ups. Keen to avoid the high costs and hassle of managing their own on-prem, or negotiating cloud capacity bottlenecks and the issues of ‘cloud lock-in’, these forward-thinking companies are making the leap to colocation early in their evolution, or even from the get-go.

One such organisation is London-based Instadeep – now a global leader in AI powered decision making. After initially launching with a small number of DGX servers on premise, rapid growth brought pressures that required their impressive technical team led by Nacef Labidi to quickly adapt this strategy. In January 2021 it was our pleasure to help Instadeep establish a high density HPC cluster at Kao Data’s Harlow campus, giving them unbridled room for growth and the ability to operate this in tandem with cloud instances in perfect, fluid hybrid computing harmony.

Referring to their move to colocation, Karim Beguir, CEO and Co-Founder of Instadeep, said: “Instadeep is an EMEA leader in decision-making AI, hence it is key to deploy our supercomputing and AI hardware within a technically superior environment, capable of supporting the latest HPC and GPU-powered technologies. By working with Kao Data, we have found a data centre that offers world-class infrastructure, specialist HPC support and the scalability to help us to grow our industry-leading AI product platforms and solutions.”

Instadeep has recently raised new investment worth £100 million to expand their operations and they clearly demonstrate that colocation is a viable solution for AI start-ups very early in their life span. This idea will only grow, and instead of viewing data centres as a far-off vision, something only to consider when reaching production, AI start-ups would do better to see them as early-stage incubators for fine-touch, data intensive computations with the potential to nurture growth and success alongside whatever lower-level compute they also file through public clouds.

Moving forward, the question for tech entrepreneurs should no longer be, ‘can I afford to use colocation at an early stage?’, it must be, ‘in order for my company to scale at pace, can I afford not to?’.

You can read more about Instadeep’s remarkable journey and how Kao Data have been honoured to help them via our Case Study here. If you’re an AI start-up in a similar position, reach out and we’d be delighted to give you a tour across one of our purpose built HPC data centres.

Tom Bethell

Tom Bethell is one of Kao Data's Business Development Directors. With a background in IT Infrastructure; Tom has been working specifically within the areas of data centre colocation and high performance computing for a number of years.



Share

Other articles

3 May 2022

Commissioning – the first step in ensuring a sustainable data centre footprint

Read more
6 Apr 2022

Collaboration is key to customer implementation

Read more
21 Mar 2022

To cloud or not to cloud - that's not really the question!

Read more
Get a quote