- CIOsurge
- Posts
- AI-specific cloud spend grew 165% this year
AI-specific cloud spend grew 165% this year
Plus: how one security team went from RFC to production in three sprints, and the $500 billion cloud bill behind it all.
Zack Tembi
August 10, 2026
AI-specific cloud spend grew 165% this year
Powered by Single Fin
Welcome to this week’s edition of CIOsurge!
This week:
Shanief Webb, CISO at Headspace, on shipping production infrastructure with junior engineers in three sprints
Token costs are mounting, and leaders are revising their AI plans instead of pulling back
Enterprise cloud spending hits a $500 billion run rate, with AI-specific services growing 165%
Let’s make this week a game-changer.
Stay sharp. Stay ahead.
RFC to Production in Three Sprints: How AI Rewrote One CISO's Build vs. Buy Math
The build versus buy calculation is changing under our feet, and my conversation with Shanief Webb, CISO at Headspace, gave me the most concrete example I have heard of what that shift looks like inside a real security team.
I asked Shanief what leadership is asking for before green-lighting a new tool. His answer was that the scrutiny has intensified. "Now there's a lot of spend associated with the different AI tools that get procured. So the question then becomes, well, okay, you spend all these money on this AI tools. Well, what value are we getting back from it?"
But the more interesting part was what happened to his own team's build calculus. "We're definitely seeing like a shift in the build versus buy conversations as a result of leveraging AI to build for us more. We can now build our own logging pipeline architecture on our own in less than one quarter with junior engineers, just because AI kind of closed a technical gap for them." The number that stopped me: "We went from RFC to production infrastructure in just three sprints."
He was careful to scope the claim. Not every team using AI has an engineering skill set, so outcomes will vary. But where the skills exist, the math on buying a tool versus building the capability has genuinely moved.
Then we got into the part nobody budgeted for, which is what the building costs to run. I shared my own experience with the agent infrastructure at SingleFin, 14 agents and 60-plus sub-agents, and how quickly you learn that a nascent cron job running too often is just burning tokens. I have killed a lot of them. One of the agents I built does nothing but monitor cost per task and flag anything expensive so I can check whether the output justifies it. Frequency is the biggest culprit. You build something to run hourly and it turns out weekly was enough.
Shanief is thinking along the same lines. His team has AI build scripts so the scripts do the repetitive data pulling instead of the model, and they use tooling that trims the input before it reaches the AI. His prediction: "at some point we're going to shift into like token analysis and token spend and how do we optimize for that as well."
The takeaway for technology leaders: AI just made building competitive with buying for teams that have engineers, and junior engineers can now ship infrastructure that used to require seniors. The catch is that you have traded a predictable license fee for a variable token bill. Instrument your spend per task and audit job frequency now, or the build savings evaporate into inference costs.
Token Costs Are Mounting. Leaders Are Revising Plans, Not Pulling Back.
Rising token costs are forcing enterprise leaders to rework their AI strategies, but an EY survey found that many businesses are charging ahead rather than retreating as the bills mount. The cost pressure has pulled FinOps, a practice originally built to manage cloud spend, squarely into AI territory, and the Linux Foundation's newly launched Tokenomics Foundation is now developing frameworks to help enterprises get AI spending under control at scale. Tokens have become the unit of measure and the pricing mechanism for AI use, and while they account for only a portion of total AI spend, they are the most easily metered layer, which makes them the natural starting point for cost discipline.
Revising plans while charging ahead is the right instinct, but only if the revision is real. The teams getting burned are the ones treating token costs as a line item to absorb rather than a signal to redesign. Shanief's point in this edition is the playbook: script the repetitive work so the model only touches what needs intelligence, trim inputs before they hit the API, and audit how often your jobs actually need to run. The bill is not the problem. The bill is the feedback. The problem is architectures designed when nobody was counting.
The $500 Billion Cloud Bill Behind the AI Boom
Enterprise spending on cloud infrastructure passed $143 billion in the second quarter, a 43% year-over-year growth rate, according to Synergy Research. That follows 11 successive quarters of accelerating growth during which the market doubled in size, pushing total revenues for the trailing twelve months past $500 billion. AI is driving most of the incremental growth, with AI-specific cloud services growing 165% year over year, and the gains are concentrating among the leading hyperscalers and a rising class of neoclouds. Nine rent-a-GPU neocloud operators now rank among the top 40 cloud providers by service revenue, with CoreWeave, Oracle, Crusoe, Nebius, and Nscale posting the fastest growth among second-tier providers.
Every token your teams burn and every agent you deploy rolls up into this number, and this number is compounding at 43%. Two things follow. First, the pricing power sits with the providers right now, which means your negotiating leverage comes from workload portability, so architect for it before your renewal, not during. Second, the neocloud tier is now real enough to be a genuine alternative for AI workloads, and a bid from one changes the conversation with your hyperscaler. The market doubled in eleven quarters. Assume your AI infrastructure line will try to do the same, and build the controls now.
💡 CIO Spotlights
Florian Roth named Chief Information Officer at Nokia
Appointed CIO of Nokia, overseeing global IT strategy and supporting enterprise transformation through digital innovation, AI, and modern technology platforms
Joins after nearly 21 years at SAP, most recently as President of SAP Business Suite, Customer Services & Delivery, leading the global customer experience organization
Served as SAP's Chief Digital and Information Officer from 2018 to 2024, with earlier leadership roles across global business operations, controlling, and cloud
On the move, in his words: technology creates its greatest impact when it empowers people, simplifies complexity, and enables innovation at scale
Nikos Angelopoulos named Group Chief Information Officer at Nedbank
Named Group CIO of Nedbank, one of South Africa's largest banks, effective September 1
Brings more than 30 years of international technology leadership across Europe, the Middle East, Africa, and the Americas
Deep expertise in technology strategy, digital transformation, cloud, data, and AI
Joins from MTN Group, where as Group CIO he led technology strategy, investment, and governance across 16 markets
Reply