When CloudZero’s CTO Erik Peterson joined the FinOps Weekly podcast in October 2025, he didn’t hold back. Instead of going on about the usual best practices of AI cost optimization, he posed challenges to how we approach AI spending.
From “zombie AI experiments” eating your budget to why you should stop apologizing for using AI, these 10 insights from the podcast are worth considering in how we approach AI FinOps.
(Watch the full podcast below and keep reading for more!)
1. The Zombie AI Apocalypse Is Already Here (And It’s Expensive)
The concept according to Erik
“I call this the ‘zombie AI apocalypse’. You know, because we’ve got all these half-baked ideas busy costing money and nobody’s really quite sure why [they’re] there.”
Erik’s most vivid metaphor captures a reality every organization faces: abandoned AI experiments that continue consuming resources long after their creators have moved on. These aren’t simply idle resources. They’re the undead AI, actively burning through compute, storage, and API calls with no clear business purpose.
Why this matters now
Unlike traditional EC2 instances you forgot to shut down, zombie AI workloads are harder to detect because they can look productive. They’re processing data, making API calls, and generating outputs, but they’re not doing actual, beneficial work.
When AI infrastructure costs are 10-100x higher than traditional compute, AI zombies can become exponentially more expensive than their shadowy predecessors.
What you can do
Conduct a “zombie audit” every 30 days:
- Tag all AI experiments with creator, purpose, and expected end date
- Monitor for workloads with declining or zero human interaction
- Set automatic kill switches for experiments older than 90 days
- Require business justification for any AI workload consuming >$100/day
2. Take The One-Hour AI Challenge (Your Career Depends On It)
The concept according to Erik
“I call it the one-hour challenge, which is: take an hour out of your day and just do everything during that hour using a set of AI tools.”
Erik’s one-hour AI challenge pushes us to understand the logarithmic change happening beneath our feet. Today’s professionals are right on the edge of an exponential upward (or downward, if you’re a pessimist) curve, and many may not even realize it.
Why this matters now
The technology gap between AI users and non-users is widening exponentially, not linearly. Every month you delay adopting these tools, you fall further behind. If you’re doing math, Erik asks, “why are you showing up with an abacus?”
As such, Erik’s challenge forces you to confront this in just 60 minutes a day. Get comfortable with AI because it’s not going away.
What you can do
Block one hour this week and:
- Use ChatGPT for ALL written communication
- Let Claude write ALL your code
- Use AI for ALL research and analysis
- Document what surprised you, frustrated you, and amazed you
- Repeat regularly to track the technology’s evolution
3. Stop Apologizing For AI (You Don’t Apologize For Calculators)
The concept according to Erik
“Stop apologizing for AI. We’ve all been watching presentations or folks up there, [saying during a presentation], ‘Oh, and by the way, AI generated my images’ and ‘AI did this’ and ‘AI did that’.”
We’re in a bizarre transition period where people feel compelled to disclose AI usage like it’s cheating. Erik argues this apologetic stance doesn’t carry well, and is wholly unnecessary.
Why this matters now
The window between “AI as novelty” and “AI as necessity” is rapidly closing. Companies still treating AI as optional are about to be disrupted by those who’ve fully integrated it. Your embarrassment about using AI tools is a competitive disadvantage. After all, you don’t see mathematicians “confessing” that they used a calculator to do their work. So why should you?
What you can do
- Remove all “AI-assisted” disclaimers from your work
- Judge output by quality, not by method
- Invest in AI training for your entire team
- Make AI proficiency a hiring requirement
- Celebrate AI wins publicly to normalize usage
4. The AI Cost Iceberg: Your $500 Bill Is Hiding $50,000 in Damage
The concept according to Erik
“These AI experiments that are happening in your business are icebergs. There’s a little piece floating above the surface. You might see an indicator in your bill. Oh, somebody’s spending some money on Vertex or Bedrock and there’s a little bit of AI spend over here, and it’s a thousand bucks or 500 bucks.”
In other words, that innocent $1,000 monthly OpenAI bill? It’s triggering massive data transfers, storage explosions, and infrastructure scaling that happen under the surface; in short, invisible yet expensive. Erik’s cost iceberg concept is a great metaphor for the true total cost of AI adoption, especially in his example of an organization whose data science team moved their data from Switzerland to the US west and ended up with an explosion in S3 costs..
Why this matters now
Organizations are budgeting for API costs but ignoring the 10x infrastructure multiplier effect. One team’s “small” AI experiment can trigger cascading costs across your entire cloud infrastructure.
What you can do
Map the full iceberg for each AI initiative:
- Track data ingress/egress costs triggered by AI workloads
- Monitor storage growth from training data and outputs
- Calculate human time spent on failed experiments
- Include compliance and security overhead
- Build a “true cost multiplier” (typically 3-10x the visible costs)
5. Ask ‘Is It Worth It?’ Not ‘What Does It Cost?’
The concept according to Erik
“Let’s all stop asking what AI costs and start asking if it’s worth it.”
Erik’s fundamental reframe challenges the entire cost optimization mindset. The question isn’t how AI spending can be curtailed. Rather, it’s identifying whether that spending generates proportional value.
Why this matters now
Companies are simultaneously slashing cloud costs while massively increasing AI spend. This isn’t a contradiction. It’s managing portfolios.
FinOps, Erik says, needs to go beyond thinking about it as infrastructure management. When companies focus too much on cost reduction, they’re missing out on the more important strategy of value creation.
What you can do
Shift your metrics framework:
- Replace “cost per token” with “revenue per AI interaction”
- Track “value delivered” not just “resources consumed”
- Compare AI ROI to human equivalent costs
- Measure opportunity cost of NOT using AI
- Accept negative margins if growth trajectory justifies it
6. AI Already Has Your Credit Card (You Just Don’t Know It Yet)
The concept according to Erik
“This debate is just starting to bubble to the surface. Are we ready to give AI the credit card? Are we ready to let it do the purchase?”
Your AI is already making purchasing decisions through architecture choices, instance selection, and resource allocation. The only question is whether you’re monitoring it.
Why this matters now
AI-generated code includes infrastructure decisions that create long-term cost commitments. That seemingly helpful Copilot suggestion to use a larger instance type just committed you to thousands in monthly spend. The credit card handover has already happened. Most teams don’t realize it until the invoice comes.
What you can do
Implement AI cost guardrails:
- Review all AI-generated infrastructure code for cost implications
- Set spending limits on AI-initiated resource creation
- Create approval workflows for AI-suggested architecture changes
- Track “AI-influenced spend” as a separate metric
- Build cost awareness into your AI prompts
7. The Fastest Growing Feature Is Always The Least Profitable (And That’s OK)
The concept according to Erik
“As we add these new AI features, our customers are trying to track that from a unit economics perspective. And I go, ‘Oh, that’s fantastic. Well, how’s it going?’ And they go, ‘Well, it’s amazing. It’s the fastest growing, most widely adopted feature of our company ever.’
“And I [say], ‘You don’t sound so excited’. [They respond,] ‘It’s also the least profitable currently. And I’ve presented this to my management, but they’ve told me there’s no way we’re turning this off.’ Yeah — because it is the fastest growing, most exciting thing that they’ve done in a long time.”
Why this matters now
The race for AI features has created a profitability crisis nobody wants to discuss. Companies are trapped between customer demand and unit economics that aren’t working. Yet stopping isn’t an option.
But that has to be corrected at some point — in the meantime, strategic measures need to be implemented specifically for AI.
What you can do
Embrace the unprofitable growth paradox:
- Set different success metrics for AI features (adoption over margin)
- Build a path to profitability timeline (typically 18-24 months)
- Use AI losses to negotiate better enterprise deals
- Bundle AI features with high-margin products
- Track the retention impact of AI features
8. Your Margin Is My Opportunity (The Bezos Doctrine Returns)
The concept according to Erik
“There is a very active conversation in the VC community [that] the SaaS world needs to rethink its margin target. For most people for a long time, it’s been like, ‘Well, we want to hit an 80% margin.’”
The AI revolution is forcing a fundamental reconsideration of SaaS economics. The 80% gross margin gospel is dead. Erik suggests AI-powered companies might operate more like services businesses than software companies. CloudZero CEO Phil Pergola stressed this point as well in a keynote at SaaS Metrics Palooza 2025.
Why this matters now
VCs are already recalibrating expectations. Companies clinging to traditional SaaS margins while competitors accept lower margins for AI-powered growth will lose. The market is being reset.
Erik puts it plainly: “You have companies or other ways of thinking that go all the way back to where Jeff Bezos was years ago when he was starting Amazon, where he said, ‘Your margin, my opportunity’. Where he basically said, ‘I just want to run the business exactly on the line, and invest everything back in, and just focus everything on growth.’”
What you can do
Recalibrate your business model:
- Model scenarios at 30%, 50%, and 60% margins
- Focus on absolute dollar growth over percentage margins
- Rethink pricing to capture AI value creation
- Consider usage-based pricing over seat-based
- Prepare investors for the new reality
9. Three Steps To AI FinOps Maturity: Change Yourself First
The concept according to Erik
“Change yourself, understand the icebergs, and then get into some of the tactical things. Prompt caching, better tracking of how agentic workflows are talking, getting into the A/B testing of new technology as it rolls out. But do it in that order.”
Erik’s three-step framework flips traditional FinOps on its head. Don’t start with cost optimization. Start with personal transformation, then visibility, and only then, optimize.
Why this matters now
Teams rushing to optimize AI costs without understanding the technology are making expensive mistakes. You can’t optimize what you don’t understand, and you can’t understand from the outside.
But there are things you can do. “You’ve got to build in for change, because every month seems like there’s going to be a new model that you might want to try to experiment with. You might want to have to upgrade to something new.”
What you can do
Follow Erik’s progression:
Step 1: Change yourself (Months 1-2)
- Complete the one-hour AI challenge weekly
- Build AI into your daily workflow
- Learn the technology by using it
Step 2: Find the icebergs (Months 2-3)
- Map all AI experiments in your organization
- Identify hidden cost multipliers
- Build comprehensive visibility
Step 3: Optimize Tactically (Months 3+)
- Implement prompt caching
- Rightsize model selection
- Optimize token usage
10. The Sam Altman Bet: Building The Machine That Tells You How To Profit
The concept according to Erik
“Sam Altman was asked at a conference, ‘How is OpenAI going to turn a profit?’ And he said something to the effect of, ‘I honestly don’t know how we’re going to turn a profit. We haven’t really figured that out yet.’”
This meta-approach to AI economics suggests we’re building systems whose business models we don’t yet understand. The AI will eventually optimize itself, including its own economics.
Erik echoes this sentiment: “I do know that I’m building the system that will one day tell me how, and then I’ll just go do it. Once the system tells me how AGI is going to [do it], and [once] OpenAI has produced the model that is actually smart enough, [Altman] knows at least his his bet is that the system he’s building is going to then one day inform him how how to go about it.”
Why this matters now
Every company is making the same bet: that AI advancement will eventually solve the unit economics problem. It’s simultaneously the riskiest and most necessary bet in business today.
What you can do
Embrace strategic ambiguity:
- Build for capability first, profitability second
- Track value creation metrics alongside costs
- Prepare for multiple pivots in your AI strategy
- Invest in flexibility over optimization
- Document learnings for when the model “tells you how”
The Bottom Line: Don’t Just Optimize, Start Transforming
Erik’s insights reveal an uncomfortable truth: traditional FinOps thinking doesn’t work for AI. While we’re counting tokens and optimizing prompts, we’re missing the forest for the trees.
The companies that win the AI race won’t be those with the lowest cost per token. They’ll be those who understood earliest that AI changes everything, from how we work to what margins mean to whether profitability even matters in the short term.
Your action item isn’t to reduce AI costs. It’s to ensure those costs are worth it. And if you’re not sure whether they are, you’re already asking the wrong question.
The zombie apocalypse is here. The icebergs are floating. The credit card has been handed over.
The only question is: Are you still showing up with an abacus?
The Cloud Cost Playbook
The step-by-step guide to cost maturity
