In an increasingly volatile world, Karl Havard, CCO, Nscale, says relying on large, multinational public cloud environments will be seen as a riskier option for an increasing number of businesses.
Years ago, many CEOs and boards fell in love with the promise of the public cloud.
The allure was enticing: lower costs, greater agility and access to cloud platform services that provide opportunities for innovation. For a while it seemed like the perfect solution for modern businesses.
However, as more companies adopted cloud services and data became increasingly valuable and abundant, the reality began to diverge from expectations.
As a result, businesses are increasingly revisiting private cloud today.
Reconsidering cloud strategies for the AI era
The driving force behind the move to public cloud was cost. Most companies bought into it in the hopes of achieving significant savings by renting server space rather than owning them outright. It also meant they jettisoned any maintenance or upgrade costs. However, according to a report from McKinsey, only 55% of European businesses are satisfied with their cloud investments.
At the same time, businesses also began to gain a transformative understanding of how they should use and store data. Even before the GenAI boom, there was an increasing need for data privacy and security ushered in through GDPR regulations and rising cyberattacks. These concerns have not abated – as organisations are becoming increasingly aware of international regulations like the PATRIOT Act that put their data at risk, we’re starting to see businesses take more of an interest in where their data is stored.
Growth in the private cloud services market is set to surge from $92 billion in 2023 to a staggering $405 billion in 2033 according to a report from Future Market Insights. And in a 2024 report on workloads from IDC, more than 80% of respondents expected to see some level of cloud repatriation in the coming year.
Whilst not solely down to the boom in AI, it is in large part responsible for this dramatic swing back to private cloud. AI holds a tantalising promise of immense productivity gains, but these advancements also come with important considerations on how it should be applied. Concerns remain around data security, privacy and the risks of accidentally leaking intellectual property or confidential company data with the negative impact this can have on a company’s reputation.
To compete today, businesses need cloud infrastructure optimised for AI that doesn’t incur unexpected usage surcharges or negatively impact other business applications.
However, public clouds are not proving sufficient for this and businesses also aren’t adequately scoping out what they need for their AI projects.
Research by RAND found that the failure rate is higher than 80%, despite $675 billion being spent on cloud computing last year. Although it is common to make use of the public cloud to run models, in part due to easy scaling capabilities, businesses are only just starting to recognise that it isn’t a one-size-fits-all solution.
The retreat to private cloud
As well as having the infrastructure required to train and run the most advanced LLMs, businesses also need to safeguard training data. Unsurprisingly, companies do not want valuable, corporate data floating around inside a public AI model.
Beyond AI performance, if cost was a driving force behind the move to public cloud years ago, then it is also a huge factor back to private cloud today. This is because private cloud offers predictability of cost. It is quite challenging to predict the cost of AI compute on public clouds – in addition to the cost of training a model, there is a potentially endless money sink in the form of paying for usage fees.
Alternatively, private clouds provide more predictability because the infrastructure is dedicated. According to Forrester’s Infrastructure Cloud Survey in 2023, 79% of roughly 1,300 enterprise cloud decision-makers said their firms are implementing internal private clouds, which will use virtualisation and private cloud management. Even though not all of these companies will be applying for AI, private cloud offers the certainty in expenditure necessary to justify large investments into cloud computing capabilities.
However, despite their advantages, private clouds are not without their challenges. For instance, specialised hardware is required for large-scale AI operations. This can be cost-prohibitive and require extensive power and cooling systems at a time when sustainability is a necessary consideration for businesses. Furthermore, it doesn’t last forever. The two major GPU providers, Nvidia and AMD, release new components annually, and businesses can expect to renew their investment in GPUs as the compute requirements to train future models increase.
At the same time, the financial impact of running this type of hardware is yet to be fully understood on account of the relative infancy of large-scale AI. AI requires power beyond what is traditionally expected for data centres – the average server rack consumes 130KW of power and there can be thousands of racks present in one facility.
Scalable, sovereign and secure cloud infrastructure
Over the next year, we’ll see enterprises move to private clouds, particularly for AI, and hyperscalers will change what they offer to reflect this demand.
Building a scalable, sovereign and secure cloud infrastructure that is capable of handling the requirements of large-scale AI is necessary for businesses looking to use the technology to gain a competitive advantage.
However, this transition must be a cost-effective and sustainable one.
Particularly as the world becomes more volatile, companies will increasingly value ownership of their AI and cloud infrastructure. Organisations are waking up to the need to own their infrastructure in regions with stronger data protection and privacy laws and they are placing greater emphasis on retaining full control over that infrastructure.
As a result, relying on large, multinational public cloud environments will be seen as a riskier option for an increasing number of businesses.
