Keeping AI cost effective in the move to cloud

AI is an incredibly promising and powerful technology, but certainly needs to be deployed as efficiently and effectively as possible in the short term to avoid the technicalities becoming time consuming in the long-term.
Published on
April 5, 2024

AI startups are the pioneers of the new digital age. Analysts estimate that the global AI market could grow by a factor of ten $15bn (2021) to $150bn by 2028, and in the UK, spending on AI technologies could reach £83bn in the next twenty years. 

As trailblazers, many AI startups are cloud native, making use of cloud’s scalability, agility and compelling performance-price to greatly accelerate their time to market. However, some do start with small, experimental tests on servers in-house, and eventually graduate to cloud environments. 

This migration can be tough, but it’s necessary - AI applications usually rely heavily on GPUs, which are a big investment for any startup or lab. Delivering this level of high performance at scale is often out of the question from a CapEx point of view, especially for a start-up. 

Moving to the cloud can help startups to remove technology bottlenecks without upfront investment in hardware. However, there are a number of things that teams should be aware of and prepare for if they are to make the migration to cloud as painless and productive as possible. 

  1. Portable Platforms: Working with platforms like Docker or Kubernetes from the get-go can greatly help before and after migration. Even before migrating to cloud, working with containers will allow you to replicate application components as well as any dependencies, and run it in an identical way. Having an AI application running in a container also means that it’s easier to configure during the migration process, because all its dependencies will be moved with it. 
  1. Credible Clouds: It’s important to do your research on different cloud providers. For example, do they have the right security accreditation for your needs? AI applications can often handle private user data, whether it’s a simple chatbot in retail banking or complex healthcare analytics systems, for example, so it’s crucial that this data is handled, stored and protected appropriately. 
  1. Costly Clouds: Price can be a complex issue when buying cloud. In theory – and often in practice – cloud offers a cost-effective way to pay for what you’re using on an opex rather than capex basis. However, moving data into (ingress) and out of (egress) can be a costly experience, and although large providers may offer discounts to start with, these discounts can become less compelling as time goes on. Startups should make sure they read the fine print and understand the whole price package – not to mention the contract duration - before signing up with a provider. 

“Thanks to OVHcloud solutions, our costs savings are roughly 50%,” Vitalii Zurian, CTO & Co-Founder at Leetify

  1. Sensible Sustainability: Sustainability is a must for any modern business today. The Wall Street Journal recently revealed that AI search queries can be seven times more compute intensive than a standard search engine. AI application developers should be considering sustainability from the start, and also trying to understand how their partners manage recycling and e-waste. 

  1. Love your Latency: From a technical perspective, latency considerations are crucial; chatbots and other real-time systems need to respond instantly to users. This means that both code and infrastructure must be fast, and developers and deployers should aim to shave off every possible milli-second. This also means that compute resources, for example, are as close to (or in the same place as) the data being handled! 

  1. Master your Monitoring: Once the application has been deployed, it’s key to keep monitoring performance. Demands change, and upgrades to an AI app can mean that it performs differently in its environment to when it was first deployed. Working with open standards, in an open-source cloud environment such as OpenStack, can often make this less challenging. 

AI is an incredibly promising and powerful technology, but certainly needs to be deployed as efficiently and effectively as possible in the short term to avoid the technicalities becoming time consuming in the long-term. 

If AI developers can plan carefully, choose their partners well, and streamline their processes when they move applications to the cloud, then they will considerably increase their chances of successful re-deployments, staying sustainable, keeping costs down and end-users happy. And given how tough things already are for small, growing companies, this is a very worthwhile goal indeed.

Companies referred IN ARTICLE: