The digital skyline will be dominated by factories, artificial intelligence (AI) factories. That is if you believe the hype – but what is an AI factory, and how can organisations benefit from this industrial revolution of technology?
Factories mass-produce the same product over and over again. In doing so, they cut the cost of production and increase accessibility to the product.
Consumer goods such as shampoo and soap are no longer the preserve of alchemists, but are churned out by the millions by multi-national companies, with the same brand being available in every geography.
Software has been moving towards mass production since the arrival of the graphical user interface (GUI); however, in the enterprise, there has always been a need, or perhaps a misplaced conviction, that there were unique software requirements.
With AI rapidly entering the enterprise, organisations need an efficient way to develop and deploy AI before costs spiral, tech debt tips the balance sheet, and nobody in IT knows who the owner is – and what the source code is.
Enter the AI factory concept. “An AI factory is the desire and ability to have repeatable small pieces of work, with stable costs and access to skills,” says Duncan Anderson, former chief technology officer of Watson AI at IBM and co-founder of Barnacle Labs, a generative AI (GenAI) consultancy.
An enterprise AI factory differs from sovereign AI factories, which Nvidia describes with Panglossian hope as where “governments can create economic opportunities, drive scientific breakthroughs, address societal challenges, cultivate local language models with region-specific datasets, and establish leadership in the global AI landscape” – no doubt with Nvidia chips.
Business benefits
Enterprise AI factories, in effect, do for AI what low and no code technology platforms have done for business applications. Using templates and repeatable processes, they can create business outcomes that users need. Prasad Prabhakaran, head of AI with Esynergy, a data and cloud engineering specialist, says this ensures organisations can move on from having AI proof of concepts (POCs) that remains POCs, as it’s hard to measure the economic return.
Anderson says organisations have to be aware that an AI factory is ideal for large volumes of small AI work that is scalable. A further advantage is that these smaller outcomes don’t require major data engineering to take place up front.
An AI factory can produce outcomes using point-to-point data integrations, he says. This is important if AI is to deliver on its much-vaunted promise. Business advisory firm McKinsey claims AI could deliver an additional $200bn in annual value to the banking sector and $400bn to retail a year. This, it says, will be delivered through worker augmentation, with half of existing work activities being automated by AI between 2030 and 2060.
Intelligent automation of processes, delivered through agents developed in the enterprise AI factory, could deliver on these promises. Unlike its predecessor, robotic process automation (RPA), AI can intelligently decide on the routes a process takes, whereas RPA could only follow a defined path. Berlin-headquartered N8N is being adopted by technology and recruitment firms for this type of automation.
Prabhakaran says organisations must evaluate the risk and value creation, curate the models used by the factory and constantly monitor the data sources to ensure the automations created by it benefit the organisation.
Factory limits
An enterprise AI factory is not the be-all and end-all of an AI strategy, but merely a component. With an AI factory, organisations can build their own AI agents and not be dependent on the vanilla agents from major platform providers.
Ben Peters, co-founder and CEO of Cogna, an AI specialist for sectors such as utilities and physical engineering, believes organisations will use an AI factory to “fill in the gaps” and that factory-made AI will not disrupt the systems of record from the likes of Oracle, SAP and Salesforce.
Anderson at Barnacle Labs says the rise of the term “agent” is “not terribly helpful” as it often refers to a very small amount of AI being used by a platform. Organisations must understand, he says, that an agent is at the lowest level of complexity in terms of what AI can do in an organisation.
“The AI Factory concept breaks down as you go up through the scales of complexity,” adds Anderson. “As the work becomes more sophisticated and larger in scale, and is using leading-edge AI, then the capabilities change, and it is no longer repeatable.”
As ever with enterprise technology, it’s vital that the full range of business needs is understood and the technologies that can help meet those needs are mapped to that situation. Once that is understood, organisations can see where simple AI processes and agents can make a difference, and then if there is value in building an AI factory.
“An AI factory is limited in scope to discrete processes,” warns Anderson. “A factory is the antithesis of creativity. With AI, you need creative thinking about making a process better and more efficient, and how to get better value as an organisation.”
Skills factory
For those organisations that can benefit from their own AI factory, there is a labour cost. “You cannot succeed without the engineering skills,” says Prabhakaran.
Organisations looking to tool up an AI factory will need data engineers, risk management and ethics expertise, strategic and stakeholder management leaders, business analysts, solution designers, and machine learning and cloud infrastructure engineers – a big ask when IBM and consultancy firm BCG found that few organisations (just 6%) had begun upskilling their teams to meet the needs of AI.
The good news, according to Peters, is that some skilled roles will evolve as the AI factory concept takes hold. He believes prompt engineers will take on evaluation roles as the factory outputs change or reduce the prompt needs.
What does it mean for IT?
It cannot be assumed the AI factory will be built by the technology leadership of an organisation. What can be expected is that at some point, there will be an impact on IT and the infrastructure they provide to the enterprise. As Peters at Cogna says: “When you move something from human to software, you hopefully make it more responsive, but you also make it a responsibility of the IT team, so security and compliance are key, as with AI, there is a larger attack surface area.
“An AI factory will make software abundant, whereas before, it was throttled back by access to skills. Now, there will be a proliferation, but its quality will not be as good.”
This means IT departments will need to increase their ability to review, identify and mitigate vulnerabilities.
Prabhakaran adds that IT departments will need to become good at risk profiling, especially as with or without an AI factory, there is the growth of shadow AI in organisations. In theory, an AI factory should curtail shadow AI, but workers who are under pressure will always find a workaround.
Demand for GenAI is on the increase, a 50% growth in the three months to May 2025, according to a study by security supplier Netskope.
The Netskope Threat Labs cloud and threat report finds: “GenAI platforms expedite direct connection of enterprise data stores to AI applications with the popularity in usage creating new enterprise data security risks that place added importance on data loss prevention (DLP), and continuous monitoring and awareness. Network traffic tied to GenAI platform usage also increased 73% over the prior three-month period.”
Ray Canzanese, director of Netskope Threat Labs, believes this is due to a growth in shadow AI. “The rapid growth of shadow AI places the onus on organisations to identify who is creating new AI apps and AI agents using GenAI platforms and where they are building and deploying them,” he says.
To mitigate this risk, IT will need to be at the forefront of change management programmes that are tailored to the adoption of AI. Gartner advises that change management programmes need to focus on employee behaviour.
Factory cost
Although an AI factory will reduce the cost of developing and deploying AI across the enterprise, there will be a cost to the organisation from increasing the usage of AI. With an AI factory able to quickly deploy small repeatable automations, organisations will need to study their IT estate costs to ensure it delivers value. “AI is more expensive than RPA, and you can definitely automate something and make it more expensive,” says Anderson.
Gartner warned in a paper: “It’s also easy to waste money on GenAI, because costs can be so unpredictable. If you don’t understand how your GenAI costs will scale, Gartner estimates that you could make a 500% to 1,000% error in your cost calculations.”
These costs build on the increased spend Gartner has already identified in its AI in the enterprise survey, which found that in 2023, average spending was $2.3m on POCs and that AI will increase the cost of enterprise applications by at least 40% by 2027.
It remains to be seen whether an enterprise AI factory could offset the latter by decreasing the AI footprint of major applications. Prabhakaran of Esynergy, which has a Cloud FinOps practice, says organisations will need to be careful at managing the application programming interface gateways that AI uses to keep costs under control.
As with the cloud, the use of AI and an AI factory will require a strategic lens on the right workloads for the right business outcomes.
Source link
