Runway Intelligence is OpenMetal’s executive insight series for late-stage startups and their investors, exploring how cloud economics, infrastructure design, and operational strategy shape valuation, margins, and time to exit.
At current SaaS gross profit multiples, every dollar of avoidable cloud spend erases $24–25 from the exit price. Most PE tech due diligence doesn’t look at cloud cost structure. That gap is becoming expensive.
Key Takeaways
- At 24–25x gross profit multiples, $500K/month in avoidable cloud spend erases $144–150M from the exit price. Cloud cost structure is a purchase price variable, not an operational footnote
- Most PE tech DD frameworks don’t model cloud spend as a gross margin variable; acquirers are regularly inheriting margin problems that were identifiable and fixable before close
- Every SaaS acquisition target sits on one of two tracks: adding AI on hyperscaler infrastructure with ballooning COGS, or running legacy workloads on overpriced compute. Both tracks carry exit value exposure
- Gartner projects that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, with roughly a fifth of enterprise application SaaS spending exposed to agentic AI by 2030; the cost structure of the defensive AI investment matters as much as the investment itself
- Dedicated GPU bare metal at fixed monthly pricing removes the variable per-GPU-hour meter of hyperscaler on-demand, a cost-structure difference that flows directly to the gross-margin line buyers price at close
$500,000 per month in avoidable cloud spend doesn’t show up in a revenue model. It shows up at close, when the buyer prices the company’s gross margin and the seller realizes the exit value they projected never accounted for $144–150M in erasable infrastructure cost. At 24–25x gross profit multiples (the current average range for high-growth SaaS), cloud cost structure is not an operational detail. It is a direct input to the number on the term sheet.
Most PE tech due diligence frameworks were not built for this environment. They assess product differentiation, churn, net revenue retention, and engineering headcount. Cloud infrastructure spend gets a line in the income statement but rarely a framework for evaluating whether the cost structure is appropriate, avoidable, or about to get worse. As a result, acquirers are inheriting margin problems that were identifiable, and fixable, before close.
The Due Diligence Gap
The gap is structural. Standard tech DD evolved when cloud infrastructure was a commodity cost with limited variance: AWS, GCP, and Azure prices were broadly comparable, and the optimization opportunity was a rounding error relative to the revenue story. That assumption no longer holds.
Two dynamics have broken it.
- First, Gartner projects that 40% of enterprise applications will feature task-specific AI agents by the end of 2026 (up from fewer than 5% a year earlier), and that agentic AI puts roughly a fifth of enterprise application SaaS spending, about $234 billion, at risk by 2030. Every SaaS category is now under pressure to ship AI features or be repriced as legacy software. The infrastructure cost of that competitive response (GPU inference, AI-native compute, model hosting) is now a central variable in how much a SaaS company’s AI positioning actually costs to maintain.
- Second, Bain’s 2025 Technology Report published a framework for gauging SaaS displacement risk along two dimensions, the potential for AI to automate SaaS user tasks and the potential for AI to penetrate SaaS workflows, scored across roughly a dozen underlying factors. For a PE firm with exposure in those categories, the diligence question is not just whether the target is defensible; it is whether the defensive investment in AI features is being made at a cost structure that holds under buyer scrutiny. A company spending heavily on AI features while paying hyperscaler GPU markup is defending its market position at an avoidable cost penalty.
Neither of those dynamics is visible in standard due diligence unless the buyer specifically models cloud cost as a gross margin variable.
Two Tracks, One Problem
Every SaaS company being evaluated for acquisition today is on one of two infrastructure tracks, and both tracks carry cloud cost exposure.
Track one
Companies adding AI features on public cloud infrastructure. GPU inference on AWS or Azure billed at a variable per-GPU-hour rate translates directly to COGS that grows with every AI feature shipped. For AI-first startups, GPU compute has run at 40–60% of the technical budget in the first two years of AI buildout (GMI Cloud). A company that built its AI roadmap assuming hyperscaler pricing will hold is carrying cost structure risk that compounds with scale.
Track two
Companies still running legacy workloads on general-purpose hyperscaler compute. EC2, managed databases, and standard cloud services at hyperscaler pricing represent spend that could be migrated to private cloud infrastructure at a fraction of the cost, but hasn’t been, because infrastructure optimization was never a board-level priority. For these companies, the avoidable cost is sitting in the income statement waiting to be found. The patterns documented in our analysis of recurring infrastructure audit findings show the same overspend categories appearing across portfolios: idle reserved instances, unoptimized storage, and general-purpose compute serving workloads that purpose-built infrastructure handles at 40–60% lower cost.
Both tracks have the same outcome at close: a buyer who identifies the cost structure problem prices it into the deal before the seller does.
Why the GPU Angle Makes This Urgent Now
The GPU pricing environment has changed the stakes. AWS GPU instance prices increased approximately 15% in early 2026, driven by enterprise AI demand and supply constraints (Amplix). Bare metal GPU alternatives, single-tenant dedicated GPU nodes at fixed monthly pricing, replace the variable per-GPU-hour meter of hyperscaler on-demand. That cost-structure difference flows entirely to gross margin.
For a portfolio company running inference at scale, the delta is not marginal. It is the difference between a gross margin profile that holds at 65–70% and one that bleeds into the 50–55% range that buyers are already associating with AI infrastructure mismanagement. As the infrastructure repricing cycle has made clear, the window to lock in fixed-cost infrastructure ahead of continued hyperscaler GPU price increases is narrowing with each quarter.
Pre-close, the diligence question is whether the target has already absorbed this cost, or whether the new owner will. Post-close, the question becomes whether there is enough time before exit to show clean quarters at the improved margin. The answer to both is better when the analysis is done before the deal closes.
How OpenMetal Fits Into Pre-Close Infrastructure Diligence
Once a due diligence review identifies cloud cost exposure, whether in AI inference spend or general-purpose hyperscaler compute, the follow-on question is whether it’s fixable before close and how quickly the margin recovery shows in the financials that anchor the purchase price.
OpenMetal’s dedicated bare metal infrastructure, available as single-tenant dedicated GPU servers with fixed monthly billing, is designed specifically for workloads where cost predictability matters. Fixed monthly billing replaces the variable per-GPU-hour pricing that creates COGS volatility on hyperscaler infrastructure. For portfolio companies running AI inference workloads, the difference is structural: dedicated bare metal GPU at fixed monthly pricing versus variable per-GPU-hour hyperscaler rates, a cost-structure difference that flows directly to gross margin at scale.
For operating partners, the same infrastructure model works across multiple portfolio companies with similar workload profiles. Stable compute and inference workloads move to fixed-cost dedicated capacity; experimental or bursty workloads remain on elastic cloud. The result is a hybrid posture that stabilizes the gross margin line most relevant to buyers while preserving the engineering flexibility teams need to move fast.
OpenMetal’s dedicated bare metal model also simplifies the diligence process itself: no proprietary lock-in, clean cost visibility for the acquirer’s technical team, and an infrastructure footprint that doesn’t create new complexity in post-close integration. For PE firms that have absorbed deal friction from opaque cloud contracts and committed-spend obligations, that clarity has its own value at close. Contact the OpenMetal team to review infrastructure cost exposure ahead of a process.
What This Means for PE Analysts
Cloud cost structure belongs in the pre-close technical due diligence checklist, specifically as a gross margin variable, not just a line-item review.
The practical additions are straightforward: model cloud spend as a share of gross profit, not just as a percentage of revenue. Identify what portion of infrastructure spend is on variable-rate hyperscaler compute versus fixed-cost dedicated infrastructure. For any company with AI features in market, quantify the inference cost per unit of output and compare it against dedicated alternatives. And for targets in Bain’s high-displacement-risk categories, assess whether the AI investment is being made at a cost structure that will hold when buyers run the same numbers.
The $24 problem is not a pricing insight. It is a due diligence gap.
Sources
- Gartner, “Gartner Says $234 Billion in Enterprise Application Software Spend Is at Risk from Agentic AI,” July 1, 2026. gartner.com
- Gartner, “Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025,” August 26, 2025. gartner.com
- Bain & Company, “Will Agentic AI Disrupt SaaS?” Technology Report 2025. bain.com
- GMI Cloud, “Where Startups Find Low-Cost GPU Cloud Computing Services,” 2025. gmicloud.ai
- Amplix, “What AWS’s GPU Pricing Shift Reveals About Cloud Cost Risk,” 2026. amplix.com