CloudNC featured in Metal Market - A long-term vision for manufacturing

March 3, 2020
CloudNC featured in Metal Market - A long-term vision for manufacturing


Many metal components are made by machining away material to deliver the finished part, but it is time-consuming to create optimal programs for the modern machine tools used. CloudNC is striving to revolutionize the business of machining at its one-year-old factory in Chelmsford, UK, where Richard Barrett met Theo Saville, co-founder and CEO, to discuss progress.

Theo Saville

Picture a solid block of high-quality stainless steel, which might be half a cubic meter or more in volume, costing many thousands of dollars. Now imagine that your task is to drill multiple holes in all six surfaces of that cube, of various precise diameters – some threaded and others plain – making sure in the process that some connect up to create channels, but also guaranteeing that others are separated from their neighbours by a pre-set minimum thickness of steel.

Such complex components are needed for applications in industrial-scale hydraulic or gas-mixer blocks, through which gases or fluids pass at high pressure. Finished component failure in use is clearly unthinkable.

The creation of components like these – and a myriad of others that are beyond the capabilities of foundry-casting or 3D printing technology – still depends on the milling, turning, drilling and grinding processes used to make them for decades.

While the speed, accuracy and reliability of the CNC machines that remove metal to create a finished part have steadily advanced in recent years, programming them to undertake a sequence of steps to arrive at the desired products has, until recently, remained the domain of skilled human programmers. While their experience will create a machining program to do the job, its creation can take many hours and the physical production of the part itself many hours more of machining time.

At the core of CloudNC’s business is its development of software using complex mathematics and artificial intelligence to consider the huge number of different sequences of machining steps that could be used to create a given part in order to find an optimal one, consequently reducing both the programming and machining time needed to make any given component.

CloudNC CEO Theo Saville said that even the best and most experienced human machine-tool programmers will not find the optimal order in which to use the tools. “They will not. It is too difficult. If you take the best CAM programmer in the world and give them three months to create the perfect CAM program for a component, that is something that now we can get pretty close to, but in the future we will be able to far exceed. Because anything that a human can do to make a CAM program go faster, our software can do every single time with no effort required,” he said.

“And then we can go beyond what people would think to do… That is where the opportunity comes from because there are so many ways of doing it,” he added.

The iteration cycle for improvements, which is what we’re all about, is incredibly fast”


The new light and bright industrial unit that CloudNC moved into a year ago is very different from the rather dark, cramped and oily image engendered by factory workshops of old. It serves as both a test-bed to develop the company’s CAM programming software and as a commercial supplier of metal components.

At the time of Metal Market Magazine’s visit towards the end of 2019, CloudNC had 11 CNC machines in operation and had more on order. After set-up, each drilling, milling and turning machine whirs and hums away unattended. While the machines’ human minders can observe progress through the transparent safety doors behind which each machine gets on with work, each workpiece is automatically positioned during machining while robotic arms take care of selecting and replacing the tools needed at each step of the sequence.

The company works with clients’ standard CAD files showing tolerances and any supplementary information. “We’d love it if everybody used model-based definition instead of 2D drawings, and then the tolerances are in the 3D file, but nobody really uses it yet,” said Saville.

Clients sometimes need guidance from CloudNC’s manufacturing engineers if they are requesting machining steps that are very difficult or impossible. “Sometimes you get threads that have gone all the way to the bottom of a hole and you cannot do that – it is basically not possible. Well, you can do it, but it is awkward, difficult and prone to error because you have to do it by hand – you cannot get it done by a machine,” Saville gave as an example.

Each part starts life as a metal blank on storage racks: squares, billet, cylinders, plates, rounds – some already water-jet cut if they comprise chunky, hard-to-remove material. The material is loaded on to trolleys with the right cutting tools, CAM program and inspection plan assigned before it is released to the shop floor for transfer to a machine, which is then set up to run that job, programmed and loaded. The first part produced is checked for quality, and occasionally sampled, before it is ready to be dispatched to the customer.

“We try to narrow that chain of events to be as tight as possible,” said Saville. CloudNC has a few different types of machine, including robo-drills and lathes.

“Robo-drills can make things like iPhone chassis for Apple. They are very good mass-production machines and are very quick. They are not hugely versatile, but are incredibly fast for what they do within their capabilities,” he explained.

Within the range of equipment at its factory, CloudNC has 5-axis DMG Mori DMU 60Evo machines. “They are very advanced and can make basically anything that you want for any industry that you want,” Saville said.

At the time of Metal Market Magazine’s visit, an aerospace component was one of the items being machined, with an automated changeover of cutting tools in process. “The thing about these machines, and what makes them so good, is that they are unbelievably fast considering their size and the amount of weight they have to move around,” Saville observed. Each of the machines holds about 120 different tools, kept in a carousel for rapid access and storage.

It might take a machine ten hours to make a multi-channel gas-mixer block from stainless steel if programmed in conventional ways, “because material like that is so hard to cut,” Saville explained. “Our software will reduce that time substantially. I would expect that part to be produced in half the time if programmed by CloudNC, which means you can double the yield of the machine in a day, and eventually quadruple it. That is the value-add of the programs that we build – take an off-the-shelf machine and make it four to eight times more productive than you would achieve in a traditional factory. That means that we get to make more margin and our customers get lower prices.”

CloudNC’s biggest CNC machine is a DMG Mori DMU 95 monoblock, which can make parts almost a metre cubed, weighing over a tonne. The company has hundreds of different cutting tools at its machines’ disposal. “We can use them in any order, with any machining parameters, at any depth, speed and coolant, so it is a huge problem to work out the best way of using these to get to your target component,” Saville stressed.

“In terms of time saving, we’ve benchmarked the technology in its current state on supported components as being about twice as fast, so the cycle time is halved and you’d get double the number of components out of a machine in a day than you would have done. We will increase that by 3x and 4x and then, on some really complex components, probably 8-10x. But on average you’re talking about several times faster in the future.”

CloudNC has already produced hundreds of different types of components and thousands of units. Saville said that CloudNC’s factory is nowhere near full. “This is 11 machines and we could probably go up to 40,” he noted. The machines are leased rather than owned. Moving a machine requires a specialized forklift truck and recommissioning it takes a couple of days. The factory layout was reconfigured in late-2019 within the space of three days.

“This building will probably be full in less than a year, but there is a lot of space here,” Saville added.

About 40-50% of the components being made by CloudNC now are already amenable to automation – a percentage that Saville said will continue to grow as the company’s software increases its capabilities.

He was unable to be specific about which of the components that CloudNC has produced to date gave him the greatest sense of pride, beyond saying that certain aerospace and military applications are amongst them. The company has also made some components for satellites.

“When we want to make components that are outside the usual envelope of automation, we have very highly skilled people here to do things that software cannot,” he added.

We want the purchasing of metals to be completely effectively commoditized


Saville sees CloudNC’s work as part of a much broader, long-term mission for the future of manufacturing.

“When many people think about the next industrial revolution, they are talking about the industrial internet of things, sensors and putting machines on line. They are thinking about 3D printing and new manufacturing processes,” said Saville.

“We see it a little bit differently. The new industrial revolution will come from the full automation of all of the traditional manufacturing equipment types: lathes, milling machines, punches, presses, bending and cutting machines. All these different pieces of manufacturing equipment that many people already assume are already automatic are anything but,” he stressed. “They all require a highly skilled human operator.”

“If you set foot in a factory, that factory is going to be filled with people making very complex decisions on an hourly basis to keep it running efficiently,” he added.

“I think that in a hundred years from now this will all be completely autonomous – you will set foot in a factory and there will not be any people in them. The only question is how long it is actually going to take to achieve that.”

He compared that vision with the difficulties of trying to find a good supplier of a run of, say, 100 metal components now, noting that it could take as long as 5-10 days to get a quote. “It’s probably going to take 6-12 weeks for the parts to come, and possibly even longer if the factory is very good and in demand. We estimate that there is roughly a 1 in 10 chance that the order will not arrive on time,” he added, also noting that indicators suggest that one in ten orders would have a quality issue in it.

He said that the price is likely to be higher than it needs to be. “So I’m going to have a terrible customer experience, probably I’m going to be waiting a very long time, paying more money than I need to for a product that is lower quality than it could be.”

For Saville, the future of manufacturing should be: “I’ve got a 3D design. Probably some generative CAD designed it with me – or designed it completely automatically – and the whole time it was speaking to a manufacturing service, probably through an API (Application Programming Interface), to fathom out how much it is going to cost to make, and optimized to make it cheaper, better and in the right factories.”

Someone wanting to buy that component will then already have a price, which they had for the whole time they were designing it, and a lead time. “I just need to click, and then instantly a factory – or a network of factories – is spinning up and pulling all of the materials and tools that they need further down this automatic supply chain into that factory to produce and get it to me in days, or even hours,” Saville envisaged.

He drew a parallel to other industries, saying that web hosting is a very good example. “Before Amazon web services, there was an incredibly fragmented market where there were very many players, it was expensive, a little bit unreliable and difficult to access, generally awkward and inconvenient,” he noted.

“Then Amazon came on to the scene with AWS. It is vertically integrated. It is extremely cheap, reliable, and incredibly flexible – if you want to spin up one computer, fine, if you want to spin up a thousand you can do that. It was better than anything that came before it and that new kind of service drove the consolidation of that industry to what we see today,” with Google and Microsoft as two more major players. “It has gone from an incredibly fragmented market to near total consolidation in the space of just over a decade.”

“I think that the same thing will happen to manufacturing over time,” Saville added. He believes it will probably be slower for manufacturing, but that ultimately the industry will consolidate, as driven by technology. “Technology is a great un-leveller in this kind of thing. There comes a point when certain technologies can be developed that are disruptive enough but also expensive enough that they can let a few key players dominate the field,” he explained.

Another comparison is with web-based hotel-booking sites that have undermined the pricing power of individual hotels. By contrast, “In manufacturing, because quality and reliability of delivery and price are so incredibly inconsistent from supplier to supplier, you can’t easily do a marketplace approach; because if you order the same part from ten factories, you’ll get ten different parts, levels of service and different prices,” he observed. “A marketplace with the current state of affairs is very difficult,” he added.

“Now, if all the machines were automatic and they all performed consistently, regardless of which factory you chose, then we would be in a perfect scenario to have a marketplace which had all of the pricing power. The whole industry would become commoditized and a lot more competitive, but I don’t see that happening without technology of the likes that we are building,” he stressed.


CloudNC generally aims to be 10-25% cheaper than its competitors, at the same sort of level of quality and service.

“So, we don’t aim to be 25% cheaper than ‘Fred in a shed’ knocking out washers on his 50-year-old lathe, but we do aim to be 25% cheaper than a reasonably advanced machine shop doing work for aerospace, or oil & gas or medical,” said Saville.

Noting that there are different tiers of the market, he said that ultimately, as the company’s technology improves, CloudNC will be able to be 25% cheaper than ‘Fred in the shed’, but explaining that the business is, at present, focused very much on aerospace-grade quality delivered extremely reliably at a price point that is below the competition.

CloudNC prefers orders to make medium- to large-batches of components. Its clients are in aerospace, automotive, defense, oil & gas and others, but he said, “Our sweet-spot is not so much in [an particular] industry – it’s the customer size. If the customer is in the GBP10-100 million revenue range and they outsource CNC machining, they are by definition our ideal customer.”


CloudNC is still at the investment phase of its development at present, “and it will be for years,” said Saville. “We bring on new backers around every 18 months as we step up in terms of what kind of company we are.”

CloudNC has already completed several successful rounds of fund-raising and will be raising more capital in 2020. The capitalization of the company was around GBP20 million at the time of Metal Market Magazine’s visit in late-2019 and Saville expects it to be nearer GBP50 million soon.

While acknowledging that some others have looked at the concept of using artificial intelligence in machining, Saville said that, “As far as we can see, excluding any stealth-mode start-ups, we appear to have been the only people in the world to have made progress on this technology.”

“It’s an incredibly challenging problem technologically – it is not something that you can solve with an army of software engineers just by throwing them at the problem and saying ‘go fix it’.” He said that progress has demanded a lot of deep R&D and computer science to push the boundaries of what computers are able to do in order to get traction on this problem.

“For the first three years and eight months of the company we could not realistically make anything commercially with our software. It’s only recently, within the past few months, that we’ve been able to start extracting value out of this,” Saville said in November last year.

“We started in this factory before the software actually worked. We knew it was going to take a while to work out how to build a manufacturing company at the same time as well as a software engineering company and then when they both became ready we could start melding them together, and that inflection point was hit a few months ago.”

Saville said that at the top end of the machine tool market, machine performance between different machine builders is relatively close. “They perform extremely well. They are relatively commoditized,” he said.

But while they are all good in terms of hardware, he sees plenty of room for improvement in simplifying their ease of use through better software. “Ultimately what we want to produce here is the perfect factory blueprint. You just rent empty spaces, fill them with machines and our system, and then it’s done and making money and it’s scalable. It doesn’t require much to expand at incredible speed then.”

CloudNC doubled in size over 2019 to 80 employees. “Ultimately this company is nothing without its people, so we have built the recruitment engine that brings in the very best people in the world and keeps them happy, keeps them developed and focused on the right things that they are retained – because you can’t solve this problem if your staff are changing over every week,” said Saville.

“I sometimes think that it would be easier if we were just a software engineering company,” he smiled. “This is one of the barriers to entry. If you want to solve this problem, you have to build software and you have to own and operate factories – that in itself is hard, but doing those at the same time, whilst scaling them up at the same time, that is incredibly challenging.”

“The management is important: it is critical to the success of a company,” he stressed. “I think that sometimes people look at technology and think that technology will save them, but then the company is nothing but its tech. No, if you take a tech company and you remove all the people the technology becomes worthless,” he added.

There are two sides to CloudNC’s factory: the production teams, who actually run the machines, generate revenue and satisfy the company’s customers, but also the software team and the continuous improvement teams.

“The iteration cycle for improvements, which is what we’re all about, is incredibly fast,” said Saville. “We can get towards that future perfect factory layout and perfect factory systems as quickly as possible, and supporting them is the software team. When it is not enough to rearrange how a process works physically, we can layer in digitally over the top of it,” he explained.

“It is a never-ending journey – the search for manufacturing perfection – but I think that we will be well beyond world class at that point,” he added.

What attracts customers to CloudNC now? “Being able to come here and see that we are visibly different to anything else that they have seen in this industry really draws them in – that and the price. The image and the price brings them in, and then the quality and reliability keeps them here,” Saville answered.

He added that many people in the industry already believed in the approach that CloudNC is taking even before the company existed: “It should be automatic, it should be faster, it should be better, it should be cheaper,” and they asked: “Why is the industry not like this?” “We are very impatient. That is what drives us,” he concluded.

CloudNC’s factory floor serves as both a test-bed to develop the company’s CAM programming software and as a commercial supplier of metal components


CloudNC machines standard alloys of aluminium and steel, and occasionally titanium. “We will be getting into aerospace grade nickel-based super-alloys over time,” said Saville, noting that the company has also machined copper and brass. “Ultimately it’s about what our customers want.”

He said that the company wants to do business with its supply chain upstream in the same way that CloudNC wants to provide business to its customers – in an automated and flexible way. “We want the purchasing of metals to be completely effectively commoditized, so that buying metal is like buying something off Amazon. I know that it is going to turn up. I know that the price is very good. I know that the quality is going to be perfect as well.”

Saville said that these things do not always happen. “The most important thing to us is the same thing for our customers, which is that material arrives on time and it is correct.”

In CloudNC’s experience, material is sometimes delivered that has been cut in a way that actually makes it very difficult to process, or it is not the size ordered. “There are good suppliers out there, but they inevitably have longer lead times as they are the ones that do it right,” said Saville.

“We will get there as an industry, but it’s a shame that they are not there yet. It makes it harder for us to deliver on the promises that we give to our customers if we are unable to get the material in time. I would like to be able to get any material within 24 hours in the right size required.”

CloudNC also does a certain amount of processing of customers’ own material. “If you start working with the very high-end aerospace and automotive companies, it becomes increasingly likely that they will give you the material [to work with] because they don’t want the surprises,” Saville added.

Material quality can also be poor. “For example, we’ve had experience of billet turning up with immense porosity inside it – you start machining it and a huge hole opens up!,” Saville recalled.

He noted that, for the entire industry, 5% of the revenue of a typical workshop goes to the cost of scrap. “If a part wanders out of tolerance, you could make 100 scrap units before you realise what is happening – if you don’t have the right sensors in place to check it.”

He added that since software does not make mistakes, it will not create scrap components. “More than 99% of the time, a scrapped component is down to human error. A broken machine is down to human error. The machines do not go wrong very often. If a cutting tool implodes, that was probably a human error – it is very rare that there was a defect within that cutting tool. It all comes back to people and their decisions,” Saville concluded.

CloudNC featured in Metal Market Magazine – published February 2020.

CAM Assist

Your AI co-pilot. Automate CAM in seconds.