AI Budgets Get Real
The era of blank-check AI spending is over — ROI frameworks finally arrive
Two numbers define this week: $340B in enterprise AI spending (Gartner) and 45% of it with no ROI measurement (also Gartner). That’s roughly $153B in AI investment that companies can’t evaluate. We’re past the era of experimentation budgets — these are real capital allocation decisions being made without basic financial discipline.
The correction is coming from two directions simultaneously. From above, the SEC now requires AI disclosure in financial reporting — which means CFOs will demand the measurement frameworks that tech teams never built. From below, Google Cloud’s cost allocation tags give enterprises the infrastructure to actually track what they’re spending on AI at the use-case level.
Klarna’s one-year retrospective is the most honest data point in enterprise AI right now. Not a success story. Not a failure story. A nuanced operational reality: AI excels at volume and speed, fails at empathy and judgment, and the companies getting real value are the ones designing around that boundary rather than pretending it doesn’t exist.
Gartner: Enterprise AI Spending to Hit $340B in 2026 — But 45% Lacks ROI Tracking
Gartner's latest forecast projects $340B in enterprise AI spending for 2026, up 38% year-over-year. But the buried lead is devastating: 45% of that spending has no formal ROI measurement framework attached. Companies are writing checks they can't evaluate. The report recommends every AI investment above $500K should have a pre-defined measurement plan with baseline metrics captured before deployment. If you're approving AI budgets without this discipline, you're building a write-down, not a capability.
Klarna Reports AI Customer Service Results After Full Year
Klarna published its one-year retrospective on replacing 700 customer service agents with AI. The nuanced reality: resolution time dropped 25%, first-contact resolution improved by 15%, but customer satisfaction scores for complex issues fell 12%. They've since hired back 200 specialized agents for escalation handling. The lesson isn't 'AI replaces customer service' or 'AI can't do customer service.' It's that AI excels at volume and speed but still fails at empathy and judgment. Design your AI customer service around this boundary.
Google Cloud Introduces AI Cost Allocation Tags
Google Cloud launched granular AI cost allocation tagging, allowing enterprises to track AI compute costs by department, project, use case, and even individual model call. This sounds mundane but it's a game-changer for AI governance. Most enterprises currently have AI compute costs buried in general cloud spending with no visibility into which teams are spending what. This is the infrastructure that makes AI ROI measurement possible. If you're on GCP, turn it on immediately. If you're on AWS or Azure, demand equivalent functionality.
Meta Releases Llama 4 with Enterprise License — Open Source Gets Serious
Meta released Llama 4, and the enterprise story is the license change: organizations under $10B revenue can now use Llama 4 commercially with zero licensing fees and no usage reporting requirements. For companies above $10B, a nominal licensing fee applies but is a fraction of API costs for equivalent capability. This collapses the cost argument for proprietary models for a large segment of the enterprise market. If you're spending more than $50K/month on API calls, run the math on self-hosted Llama 4.
Boeing's AI Quality Inspection System Prevents $340M in Potential Defects
Boeing disclosed that its computer vision AI quality inspection system, deployed across 6 manufacturing facilities, identified defects that human inspectors missed — preventing an estimated $340M in potential warranty claims and rework costs in its first 18 months. The system doesn't replace inspectors; it runs in parallel and flags disagreements for senior review. When the AI and the human inspector disagree, the defect rate is 4x higher than when they agree. This 'disagreement signal' is the most underused pattern in enterprise AI deployment.
SEC Issues Guidance on AI-Related Disclosures in Financial Reporting
The SEC released interpretive guidance requiring public companies to disclose material AI-related risks, spending, and dependencies in their financial filings. Specifically: if AI systems influence revenue-generating processes, risk management, or financial reporting itself, the company must describe the systems, their governance, and the risks of failure or bias. CFOs who haven't involved their AI teams in financial disclosure processes need to start this quarter. The audit implications are significant.
Don't miss the next issue.
Get The Brief delivered to your inbox — long-form operator stories from founder-CEOs actually shipping AI.