Расходы на облачные вычисления впервые за пять лет выросли из-за неэффективных задач ИИ

Cloud waste has been sliding gently downwards for 5 years. This year it turned around. Fitzroy IT's Tim Jenkinson and Daniel Robinson have the details, and the answers to your questions! Flexera's 2026 State of the Cloud Report, built on responses from more than 750 cloud decision makers, puts wasted cloud spend at 29%. That's the first rise since the survey started tracking it, and the finger points squarely at generative AI workloads that got stood up fast and never got tidied up. Which made for a useful backdrop to the second session of iTWire's executive Q&A live interview series, held on Thursday 27 August 2026, where I put audience questions to two people who spend their working lives inside other companies' AWS accounts. Here is the full event video, please watch, and read on below! Our guests were Fitzroy IT founder and CEO, Tim Jenkinson, and Fitzroy IT head of consulting, Daniel Robertson. Fitzroy IT was incorporated in 2011, runs out of an office in Fitzroy in inner Melbourne, and is an AWS Advanced Consulting Partner and Managed Service Provider. It won AWS Rising Star Partner of the Year for APAC in 2021, and says it has delivered more than 500 cloud deployments across 200-plus managed services clients. Jenkinson spent more than a decade in IT at JB Hi-Fi while the chain went from a couple of dozen stores to more than 120, then ran a remote support business before starting Fitzroy IT. Robertson has close to 30 years in software, including building DVD and Blu-ray authoring software in Silicon Valley for Hollywood studios, and more recently led the Datadog observability rollout for Tennis Australia, where 1,500-plus monitors watched over tournament systems at the Australian Open. He holds a computer science degree from the University of Melbourne. Here's what came out of 45 minutes of audience questions, with the numbers checked where I could check them. Handing over the keys without losing the house The first question set the tone: we don't have the in-house skills to run AWS properly, so how much can we safely hand over? "It's about having a trusted partner come in and take on the tasks that can be a little bit dry and just a little bit mundane, so that your staff and your internal crew can focus on what makes your business great, and we focus on the things that keep it compliant, safe, and performant," Jenkinson said. That framing came back later in the session, when someone asked whether a managed provider replaces an internal IT person. Jenkinson's answer was that the best engagements he sees are the ones where internal techs already exist and stay. "Everyone sticks to their knitting," he said. Internal staff own the platform the company sells to its customers, Fitzroy IT takes the back of house. For smaller organisations, he pitched fractional support, maybe 10 hours a month, as a cheaper alternative to a full-time hire. AWS secures the building, you secure the apartment The security question arrived early and stayed for most of the session, which is about right. Robertson pushed back on the idea that being in AWS makes you safe by default. "AWS secures the building, but you're responsible for securing your own apartment. If you choose to leave the window open, people can get in," he said. "AWS tells you how to close the window. They provide you with services that act as locks for the window. They even give you steel bars that you can put on the window. But it's your responsibility to do it." His practical suggestion for anyone who wants a baseline without hiring anybody: run AWS Trusted Advisor yourself. Robertson put it at about 15 minutes and around 56 checks, at no cost, producing a report you can measure future progress against. (Trusted Advisor's free tier has expanded considerably since AWS wired it into AWS Config, so the exact check count depends on your support plan. The point stands: it's sitting in your console right now.) He also made the case that a correct build decays. "If you're not actively monitoring the security of your cloud setup, the chances are that you're going to have issues, even if your cloud infrastructure was set up according to best practice. Things change over time," Robertson said. The messiest environment they've walked into I asked for the worst cloud environment they've ever inherited, and what it was costing. Jenkinson wouldn't name it, and softened the framing first: "There's no bad companies or clients. There's just busy ones." Then the detail. The client ran a large number of customer-facing services across several businesses. Under the hood, Fitzroy IT found 24 separate portals running at full availability, all day, every day. Some of those workloads ran for 2 weeks a year. Alongside that sat roughly 90 firewall rules that were creating real exposure. Robertson did the cost work. The client had come in because their bill had doubled with no clear explanation, sitting at around US$9,000 a month. Turning off services that were provably unused took about US$18,000 a year out of the bill. Moving the remaining workloads onto on-demand patterns, Savings Plans and Reserved Instances put a projected 30% reduction on the table on top of that. The method behind it is deliberately unglamorous. Two questions, then subtraction. "The first question is, 'What do you have?' and the second question is, 'What do you need?'" Robertson said. "What we often find is a lot of things that were set up with good intentions or for a short period of time that are still there, that are still running." Orphaned services. Idle services. Underutilised instances. He reckons that first pass often pays for itself inside a month. The ordering matters, and it's the part most organisations get backwards. Fitzroy IT's rule is waste first, commitments second. AWS Savings Plans advertise up to 72% off on-demand pricing, and a 1-year or 3-year commitment made before the cleanup simply locks in a discount on rubbish you were never going to need. What it actually costs a small business to get this wrong An audience member asked whether cloud security is an enterprise concern or something a small business needs to think about. Jenkinson's view: it's arguably more urgent at the small end, because the absorption capacity isn't there. "There are figures that say the average security breach costs a business upwards of 50k for an SMB. And so, if you are a small SMB, then that 50k could sink you," he said. A larger organisation wears the hit and keeps trading. That figure checks out, and it's climbing. The Australian Signals Directorate's Annual Cyber Threat Report 2024-25 puts the average self-reported cost of cybercrime for a small business at $56,600, up 14% year on year. Medium businesses averaged $97,200, up 55%. Large businesses averaged $202,700, up 219%. ASD's ACSC fielded more than 84,700 cybercrime reports over the year, roughly one every 6 minutes. Not-for-profits and the server in the cupboard A not-for-profit in the audience asked whether the equation changes when you run lean and hold sensitive community data. Robertson's answer was a story rather than a pitch. A not-for-profit client's server went down for about 30 seconds. He investigated and found an underlying hardware failure at AWS. The services were reprovisioned and the server was back inside a minute. "If that had been the NFP's server in their closet, they would have had to go out and buy a new server, hope their hard drive hadn't died, and restore from backups. If it had, they could have been down for a week," he said. He also flagged that AWS runs discounts and credit programs aimed at not-for-profits. Jenkinson added that Fitzroy IT discounts its own rates for NFPs and social enterprises, which he described as a deliberate focus for the business. The generative AI hammer problem The AI question was the one I most wanted answered: everyone's telling us to graft AI onto everything, so what has to be in place first so it doesn't detonate the budget? Robertson, who has been running AI workshops for Fitzroy IT clients, went first and didn't hedge. "The thing that has to be in place first is a clear problem that you're trying to solve. The best way to blow budget is to do AI for AI's sake," he said. "We have a saying at Fitzroy IT, which is: when you've got a Gen AI hammer, everything starts to look like a nail." He said the harder conversations are the ones where he has to tell a client generative AI is the wrong tool for what they're describing. Once there's a real problem, his checklist is privacy, security and data governance. What data are you handing to the model? How is it protected? Do you have written assurance from the provider that your intellectual property stays yours? For the build itself, he was specific about the guardrails: - A sandbox account isolated from production - Only the data you actually need for testing, nothing more - Tagging on every resource so spend is attributable - Budget alarms configured before the first token is spent That last one is what stops a proof of concept quietly becoming a 5-figure surprise. (I noted during the session that you can't spell "nail" without AI, which felt clever at the time.) Flexera's data suggests the guardrails aren't widespread yet. Generative AI use among surveyed organisations hit 81%, up from 72% a year earlier and 47% in 2024, with 45% reporting extensive production use. Security and compliance was the top AI challenge for 53% of respondents. The waste number went up in exactly the year the AI workloads went mainstream, which is not a coincidence anyone at Flexera is trying to hide. When the setup that got you here starts taxing you One audience member had grown from 10 staff to nearly 100 and was still running the AWS configuration that got them started. When does that become expensive? Jenkinson said there's a point on every growth curve where a sufficient setup turns into an ongoing tax collected quietly in the background. Robertson took it somewhere more interesting, which is the cost that never appears on the AWS invoice. "If you grow from 10 people to 100 people and you don't change your environment, there's a different kind of cost. It's not a cloud spending cost, but it's an opportunity cost, because your 100-person company is not trying to solve the same problems as your 10-person company was," he said. His example: overnight batch jobs that creep longer as headcount and data grow. The job that used to finish before 9am now finishes at 11am or midday, and everybody who needs that data starts the day sitting idle. "Those people are sitting there as idle resources, not being able to effectively do their job," Robertson said. His fix leans on the fact that current-generation AWS compute, storage and networking usually buys back the performance for a marginal change in cost. Which brings us to the most quantified example of the session. Lovisa, 65% faster batch jobs, and where the bottlenecks actually were Fashion jewellery retailer Lovisa migrated close to 30 servers off on-premises infrastructure to AWS with Fitzroy IT, across more than 20 countries and time zones. The headline result is 65% faster batch processing. The underlying numbers are better: total processing time dropped from 2,723 minutes to 958 minutes, with individual overnight tasks improving between 30% and 90%, and no disruption to store or online sales during the cutover. I asked Robertson what actually needed fixing. "We started by looking at why the batch jobs were taking so long to run and where the bottlenecks were. And what we found was a number of bottlenecks, a surprising number. We were seeing delays with CPU. We were seeing low memory conditions. Even the networking infrastructure between the machines from which the data was being gathered was causing issues," he said. All 3 of those are cheap to fix once you're in a cloud: bigger instance types, more memory, faster interconnects. The expensive part was finding them, which is the recurring theme of the whole session. The published case study lists EC2 Dedicated Hosts (using Microsoft licence portability), Transit Gateway, DirectConnect for hybrid connectivity, load balancers and CloudWatch dashboards, delivered across 4 phases starting with a single-region proof of concept. Is our data in Australia, and does it have to be? Short answer from Jenkinson: yes, and depending on your sector, yes. AWS has run its Sydney region since 2012 and opened a second Australian region in Melbourne on 24 January 2023, with 3 availability zones under the API name ap-southeast-4. "Depending on what business you're in, then you may actually have to have it sovereign in Australia. We would always recommend that you do, unless you're a global business that wants to have it syndicated around the world," Jenkinson said. "AWS is very transparent in that way. You know exactly where you're working and in which data centre you are." The Essential Eight is being retired, and it still matters today I asked what a time-poor business should check this week instead of reading the Essential Eight and the AWS Well-Architected Framework cover to cover. Jenkinson's list was application control, patching, checking your security configuration, and backups you have actually tested by restoring them. That last one catches people constantly. Plenty of organisations have backups. Fewer have proof that a restore works. He also flagged, correctly, that the Essential Eight is on its way out. "That's a government framework that will be superseded by a new version of it in the coming couple of years. But it's not one of those things where you should wait for the new framework, because the new framework will foster everything from the existing one, and it's important right now," Jenkinson said. The detail behind that: ASD is consulting on an "Essentials" series to succeed the Essential Eight after 8 years, starting with Essentials for Enterprise IT and expanding into areas such as operational technology. It shifts from prescriptive technical controls toward a principles-based, threat-informed model, and is explicitly designed to be compatible with existing Essential Eight programs. ASD technical expert, Jayden Cooke, has said the new guidance "moves away from relying only on prescriptive technical controls". ACSC head, Stephanie Crowe, framed the change as guidance needing to evolve alongside modern threats. Fortinet CISO, Cornelius Mare, was less diplomatic, saying the Essential Eight is showing "its age" in the 2026 threat environment. Robertson's version of the same argument was shorter. "In the old days they used to say, 'If it ain't broke, don't fix it,' but that's a recipe for disaster in 2026," I noted. How to buy cloud help without signing a blank cheque An audience member had been burned by open-ended consulting bills and wanted to know how a small business avoids a repeat. Jenkinson's answer was to pay properly for discovery so the delivery can be fixed price. "If a sufficient amount of budget is given to a discovery piece, then the project work that comes out of it can be effectively given at a fixed cost. So the cost of not doing the discovery is that you end up in these T and M deals that can blow everyone's budget," he said. For a small to medium business, he put discovery at roughly 3 to 4 hours across 1 or 2 sessions, with infrastructure engineers brought in afterwards for a deeper look at what's actually running. Enterprise engagements take longer. Both of them also pushed hard against big-bang projects, arguing for sequenced work where each chunk delivers a visible outcome before the next one starts. A$20 billion of AWS spending, and whether a 30-person business feels it Amazon has committed A$20 billion to expanding Australian data centre infrastructure through to 2029, and AWS says it has trained more than 400,000 Australians since 2017. I asked whether a 30-person business will ever feel that, or whether it's enterprise news. Jenkinson called out the awkward maths first: a third of cloud spend going to waste while the hyperscalers pour billions into new capacity is a strange pair of facts to hold at once. His argument is that the investment flows downhill. "The rising tide lifts all boats. Small businesses then leverage all those same things, and the cost diminishes over time as well," he said. Robertson made the more concrete point. New AWS services routinely launch in US East 1 first and reach Australia later, if at all. Sustained local investment shortens that gap, which means Australian organisations get access to newer capability without giving up data residency. "To know that AWS is committed in the Australian market to providing that degree of investment means that if you go with them, you're going with somebody who's going to stay the course," he said. Multi-cloud, hybrid, and the UniSuper lesson Someone on Microsoft 365 asked whether they're already on the cloud. Yes, said Jenkinson, whose team deploys across Microsoft and Google as well as AWS despite the firm's AWS focus. On multi-cloud, Robertson reached for the case every Australian IT leader remembers. "Multi-cloud is a valid strategy, particularly for things like disaster recovery. We only need to look at the example of somebody like UniSuper, who lost their whole Google Cloud environment and were only able to recover because they had a multi-cloud strategy in place," he said. The May 2024 incident saw a $125 billion superannuation fund's entire Google Cloud subscription deleted through what Google described as an unprecedented misconfiguration. Backups held with a separate provider are what saved it. Robertson's caveat: plenty of organisations are multi-cloud by accident rather than design, because different teams picked different providers, and Fitzroy IT often consolidates those back to one. "We don't have a prescribed answer. We listen and we design something for you that makes sense," he said. Flexera's numbers back the messiness. 73% of surveyed organisations run hybrid environments, with AWS at 83% active usage and Azure at 79%. Lock-in, and the question to ask before you sign I asked what a buyer should ask a prospective cloud partner, and what answer should end the conversation. Jenkinson's test is ownership. A provider should never hold your cloud account hostage, and removing them should be documented and simple. Robertson put the philosophy plainly. "We consider ourselves replaceable. We keep good documentation. We put in a single point of control. That's how we access their account. We show people upfront how they can remove that single point of control," he said. "That's not because we want to be removed, but because that's best practice for a partner." On what it costs to leave a provider who has made leaving hard, Jenkinson said the genuinely obstructive cases are rare and expensive, but most exits are simply a matter of the provider having drifted to a different business model. His advice before signing: check how long they've been around, and read the case studies. Where Fitzroy IT sits in the Australian AWS field The Australian AWS partner field is crowded, and Fitzroy IT competes across most of it. Versent, AC3, CMD Solutions and Telstra Purple are the names that surface most often on large transformation programs, Cevo holds AWS Premier tier, and Itoc has a long Queensland-based AWS pedigree. The reference list is where Fitzroy IT makes its case – consulting-first, and hands-on. Tennis Australia ran 1,500-plus Datadog monitors across tournament systems through the Australian Open, in front of a million-plus fans, with Robertson leading the rollout. Lovisa migrated close to 30 servers off on-premises infrastructure across more than 20 countries and time zones. Future Super came with a Kubernetes migration that cut batch job times by 75%. These Fitzroy IT clients are running enterprise workloads with enterprise consequences attached. What changes between a client that size and a 30-person business is the shape of the engagement, not the depth of the work – and Fitzroy IT handles both, with aplomb. On the cost side, the tooling market is equally busy. Flexera One, Apptio Cloudability, Vantage and ProsperOps all sell cloud financial management, and AWS ships its own Cost Optimization Hub and Compute Optimizer at no additional charge. Those products surface the numbers. Somebody still has to walk the account, work out which of the 24 portals nobody needs, and take responsibility for switching them off. That's the part Fitzroy IT sells, and it's a consulting problem more than a software one. Here's the figure that reframes the whole 29% waste number: 63% of organisations in Flexera's survey already have an established FinOps team, and 71% run a cloud centre of excellence. The waste is happening inside organisations that have staffed up to prevent it. Having the function is one thing. Having somebody with the time to act on what it finds is another, and that gap runs the full length of the market. What to do this week Stripping the 45 minutes back to actions a business could take before Friday: - Run AWS Trusted Advisor. It's free, it takes about 15 minutes, and it gives you a baseline. - List every running resource and ask whether you still need it. Turn off the orphans before you commit to any Savings Plan. - Restore a backup to somewhere safe and confirm it works. A backup only counts once a restore has been proven. - If you're experimenting with generative AI, move it into an isolated sandbox with tagging and budget alarms today. - Check where your data physically sits, and whether your obligations require it to sit in Australia. Jenkinson's closing pitch leaned on longevity rather than technology. "We've had front row seats to so many business challenges over the years. We've grown with this whole cloud process over the last 15 years, and we really just know what we're doing," he said. Robertson kept his to a sentence: "We're a company that genuinely care and genuinely want to help, and we've got the experience to do it." Fitzroy IT is offering iTWire readers a free 15-minute discovery session with no obligation, covering what you're running, what it's costing, and where the risks sit. Details are at fitzroyit.com/talk-to-us, but just scan the QR code below, and get connected with the team! Next in the series, on Thursday 3 September 2026, I'll be joined by iCXeed co-founder and chief customer officer, Ryan Rayner, talking about the customers who leave without telling you why. Registration is open now, and audience questions drive the session. If you didn’t get a chance to watch the video above, here it is again: Sources