{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreia35z25u3onrcnwvukuy7lyhjkquxhijpii472bbr5qult4ylx4hi",
    "uri": "at://did:plc:qz6ohvpdsdvv5kniizyfz25y/app.bsky.feed.post/3mgahc2ddqkz2"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreiauzaczrctbqoloichnrimmhut2flkoexa4txzuqek7exrippolkq"
    },
    "mimeType": "image/jpeg",
    "size": 6410215
  },
  "path": "/article/4140076/from-opex-to-capex-the-case-for-modular-ai-pods.html",
  "publishedAt": "2026-03-04T13:00:00.000Z",
  "site": "https://www.cio.com",
  "tags": [
    "Artificial Intelligence, Budget, Budgeting, Developer, IT Management, Roles, Software Development",
    "severance packages at Chegg",
    "$400 billion spent in 2025 alone",
    "software sovereignty",
    "In mid-2024, they laid off 1,800 people",
    "Intuit’s stock is in freefall",
    "the Financial Accounting Standards Board (FASB) issued",
    "ASU 2025-06",
    ".",
    "h",
    "as compressed to roughly six months",
    "Want to join?"
  ],
  "textContent": "If you want to see the immediate future of enterprise org planning, don’t look at NVIDIA’s stock price. Look at the severance packages at Chegg.\n\nIn late 2025, the education giant cut 45% of its workforce, leaving it with fewer than 500 employees, down from nearly 2,000 just two years prior. The reason? They were selling “rented intelligence” for homework answers. When AI made that intelligence free, their fixed-cost business model collapsed. They had built a massive permanent workforce to deliver a commodity that suddenly cost zero.\n\nChegg is the extreme case, but it’s a canary in the coal mine for every CIO.\n\nWalk into most Fortune 500 boardrooms today, and you’ll see executives misreading the landscape. They’re obsessed with the idea of the “hardware hangover”. This is the staggering $400 billion spent in 2025 alone on GPUs and data centers. They think “CapEx” (capital expenditure) means buying servers, but the real war isn’t over silicon; it’s over software sovereignty.\n\nMost companies are currently drowning in a rental trap. You pay monthly subscriptions for SaaS tools, you pay per-token for API calls and you hire full-time employees to stitch them together. If you stop paying the bills, the intelligence turns off. You own nothing.\n\nIn 2026, the smart CIOs are flipping this model, moving from renting capability (OpEx) to owning IP (CapEx). They aren’t doing this by hiring 500 new data scientists, they’re doing it with a modular, elastic workforce that builds the asset and then leaves.\n\n## What can we learn from Intuit about the trauma of skill swapping?\n\nThe fundamental problem with the 2023-2024 AI hiring boom was that it treated AI skills as static. In the AI-era, skills depreciate faster than hardware.\n\nWe saw this play out at Intuit. In mid-2024, they laid off 1,800 people (roughly 10% of their workforce) while simultaneously announcing plans to hire 1,800 new roles to support their AI transformation. Despite the symmetry, this wasn’t a swap; it was a painful structural reset. They had to shed legacy roles to make room for “customer-facing” and product roles that could leverage AI. To add insult to injury, now Intuit’s stock is in freefall because AI is “eating” their software.\n\nSo what’s the main takeaway? If you lock your strategy into full-time headcount, the only way to pivot is through trauma: layoffs, severance and cultural destruction.\n\n## Why fire and rehire when you can just bolt on?\n\nI learned the hard way that adding headcount doesn’t solve all staffing issues during my time as a talent acquisition leader in the technology sector. Having a core team with augmented contract support for high-flux hiring always served our budgets better than a monolithic team of full-time staff. The same now rings true across all teams, but most acutely for software engineering and AI/ML departments.\n\nThink of it as Gig 2.0 economics, where we define contractors as “capital expenditure catalysts” rather than temporary labor.\n\nTo explain it simply:\n\n  * **The old way:** You pay a salary forever to maintain a system.\n  * **The new way:** You pay a specialized pod to _build_ a proprietary system in 90 days. When they finish, the headcount cost goes to zero, but the asset stays on your balance sheet.\n\n\n\n## The accounting reality: Why uncertainty demands modularity\n\nFor years, CIOs hoped accounting rules would make it easier to capitalize their AI experiments. The reality of 2026 is exactly the opposite and that makes the modular model even more critical.\n\nIn late 2025, the Financial Accounting Standards Board (FASB) issued ASU 2025-06. While it simplified some aspects of software accounting, it introduced a “significant development uncertainty” hurdle. Put simply: If your project is “novel” or “unproven” (which almost all innovative AI is), you likely have to expense the costs until that uncertainty is resolved. You can’t capitalize it.\n\nThis is a dangerous trap for CIOs looking to hire.\n\nIf you hire 50 full-time engineers to build a novel AI agent, and FASB says the project is uncertain, you are forced to treat their entire payroll as an expense (OpEx). You are loading your P&L with fixed costs for an experiment. If the project fails, you destroy your EBITDA _and_ you’re stuck with the headcount. The board won’t like that.\n\nThe new way avoids this trap. By using contingent pods for the uncertainty phase, you treat the innovation phase as a sandbox:\n\n  * **Scenario A (failure):** The novel AI doesn’t work. You disband the pod. The expense stops instantly. No layoffs.\n  * **Scenario B (success):** The pod proves the concept. The uncertainty is resolved. _Now_ you can bring it in-house or capitalize the remaining build, knowing the asset is viable.\n\n\n\n## Understanding the elastic talent capacity ratio\n\nTo execute this, you need a new headcount architecture. I call it the elastic talent capacity ratio, and it involves a 70/30 split:\n\n  * **70% the core team:** These are your full-time employees. They are the product owners, the data governors, the strategists. They ensure the AI aligns with your company values. Their job is not to write every line of code; their job is to set the strategic direction and be stewards of existing infrastructure.\n  * **30% the bolt-on team:** These are high-performance pods who enter with a specific mission to build a specific asset.\n\n\n\nThis structure protects you from the market. When the economy dips, you dial down the 30% without touching your core. When the economy booms (or a new model drops), you dial it up without waiting six months for HR to recruit.\n\nCrucially, this also solves the “shelf-life” problem of AI talent. The innovation cycle in our industry has compressed to roughly six months. Today, you might need a pod specializing in ultra-low latency voice AI for a customer service overhaul. Next quarter, the priority might shift to predictive maintenance for your logistics fleet, or perhaps a completely new reasoning model that hasn’t even been released yet.\n\nIf you try to hire full-time employees for every new wave, you end up with a team built for the _last_ war. A modular approach allows you to swap in experts who live on the bleeding edge of that specific niche. You don’t need to own the skill forever; you just need to rent the expertise long enough to build the capability, capitalize it and then pivot to the next breakthrough.\n\n## Get ready to be nimble, things will only accelerate into 2027\n\nThe companies that win in 2026 won’t be the ones with the largest payrolls OR the most layoffs. They will be the ones who can assemble the best talent for the specific problem, solve it and scale.\n\nThe accounting rules, the market volatility and the pace of technology all point to the same conclusion: agility is the only asset that will appreciate.\n\nStop renting your future from big tech. Stop loading your balance sheet with permanent risk for temporary problems. Start using the Gig 2.0 approach to build an asset economy.\n\n**This article is published as part of the Foundry Expert Contributor Network.**\n**Want to join?**",
  "title": "From OpEx to CapEx: The case for modular AI pods"
}