Ramp’s CTO, Karim Atiyeh, last quarter fired off a question to his finance team: How much are we spending on AI, and where’s it going? Days later, still no clear answer.
That’s the scene playing out across corporate America right now. AI spend—Ramp’s hottest new target—is exploding, up 13-fold in monthly token usage since January 2025 per their internal data. Finance pros chase shadows while engineers rack up bills on OpenAI calls and Anthropic inferences, all billed by the token in ways that defy traditional tracking.
Ramp, that $32 billion fintech behemoth from New York, just dropped a product to lasso this chaos. It’s pulling token-level data straight from AI providers into its expense platform. No more guessing. Think corporate cards meet AI observability, but for the bean counters.
Why Is AI Spend Finance’s Worst Nightmare?
AI costs aren’t like your standard SaaS subscription. Fixed fees? Forget it. Usage-based billing—tokens in, dollars out—means volatility on steroids. Ramp’s data shows heavy users’ costs jumping 50% in a quarter. One day you’re experimenting with a chatbot; next, it’s eating your OpEx.
Here’s the thing. Engineering teams live in the code, closest to the API keys firing off requests. Finance? They’re stuck reconciling invoices that don’t match usage logs. Atiyeh nailed it in his blog post:
“Last quarter, I asked our Finance team a simple question: how much are we spending on AI and where is it going? The analysis took days and still couldn’t provide the level of detail I wanted.”
That gap? It’s toxic. Engineers optimize for speed and smarts; finance for forecasts and categories like COGS or OpEx. Existing dev tools spit out latency stats, but who cares if your model’s blazing fast but bankrupting you?
Ramp’s betting big here. They’ve been guzzling their own AI Kool-Aid internally—generating code, testing it—and watched costs balloon. Average token spend? Thirteen times higher now. That’s not hype; it’s math screaming for visibility.
But wait—Ramp’s not just reporting numbers. They’re layering on financial smarts: cost per request, per day, broken by team, project, even user. Set budgets. Flag spikes tied to that new feature launch. Reconcile invoices against actual tokens burned. It’s finance controls for the AI era.
Does Ramp’s AI Spend Tool Actually Solve the Problem?
Look, we’ve seen this movie before. Remember the early cloud days? AWS bills hitting like freight trains, no one knowing why EC2 instances were partying all night. Tools like CloudHealth and Cloudability sprang up, promising salvation. Ramp’s pitching itself as that for AI— a new layer in the stack, finance-focused, not dev-only.
Smart move? Absolutely. AI’s no side hustle anymore. Atiyeh claims it’s ballooning past payroll for some firms. My take: He’s underselling. With models getting pricier per inference and multimodal stuff chewing tokens like candy, this could hit 10-20% of tech budgets by 2027. (Bold prediction: Watch Microsoft and Google bake similar tracking into Azure and GCP billing by EOY.)
Yet skepticism lingers. Integrations with OpenAI, Anthropic, OpenRouter sound slick, but what about custom models on AWS Bedrock or self-hosted Llama? Coverage gaps could hobble it. And classifying spend—COGS vs. OpEx—relies on human tags. Garbage in, garbage out.
Ramp’s dodging dev tools like LangSmith or Phoenix, which obsess over prompts and latency. Fair play; those don’t answer “Is this worth it?” Instead, Ramp whispers to CFOs: Control the cash.
Internal proof? Ramp’s own team used it to pinpoint waste. But for outsiders, it’s early. Fintechs love moats—Ramp’s corporate card data gives them an edge over pure AI plays. Still, if adoption lags, it’s just another dashboard in a sea of them.
The real edge: Timing. Enterprises are shifting AI from PoCs to production. Spend visibility isn’t sexy, but it’s the guardrail keeping hype from crashing into reality. Ignore it, and your Q4 earnings call turns ugly.
Ramp’s not reinventing billing. They’re translating AI’s alien economics—volatile, token-tied—into ledger-friendly terms. Providers bill usage; Ramp contextualizes it. Simple. Effective?
History says yes. Cloud cost tools saved billions in overprovisioning. AI’s parallel: Token optimization could slash 30-50% off bills without touching models. (Unique insight: This echoes the GPU rush of 2023—Nvidia laughed to the bank while firms overspent on idle H100s. Ramp could be the finops for tokens.)
Critique time. Atiyeh’s post spins internal pain as universal gospel—classic PR move. Sure, Ramp struggled, but giants like Meta or Google have custom dashboards. This shines brightest for mid-market firms without armies of analysts.
The Bigger Market Play
Fintech’s invading AI ops. Brex and Brex’s rivals lurk, but Ramp’s first-mover here. $32B valuation? They’re not playing small. Expect copycats—Stripe maybe, with its Atlas AI billing hooks.
For devs? Indirect win. Finance breathing down necks means tighter API keys, smarter prompting. No more “just try the biggest model.”
Bottom line: Ramp’s onto something real. AI spend’s the fastest-growing, least-visible line item. Track it, or watch it track you—to bankruptcy court.
🧬 Related Insights
- Read more: Hegseth Boots Army Chief Over DEI Promotion Fight
- Read more: OpenClaw’s Multi-Agent Fix: Escaping the Single-Agent Hallucination Trap
Frequently Asked Questions
How does Ramp track AI spend?
Ramp integrates directly with OpenAI, Anthropic, and gateways like OpenRouter, pulling token usage into its platform for spend breakdowns by team, model, and project.
Is AI spend really growing 13x in 2025?
Ramp’s internal data says yes for average monthly tokens; heavy users see 50% quarterly jumps due to usage-based, volatile billing.
Will Ramp’s tool classify AI costs as COGS or OpEx?
Yes, it adds financial context to help finance teams categorize spend accurately, bridging the eng-finance divide.
