AI isn’t just code anymore. It’s compute. It’s tokens. And that spend is rising faster than most teams are ready for. After watching Marketing Against the Grain, I came away with a clear stance: token burn is the new cloud bill—and leaders who ignore it will torch their margins. As someone who’s built online businesses for decades, I’ve seen hype cycles. This one comes with a meter running.
The New Burn Rate No One Planned For
The clip hit hard because it was blunt about the scale. A leader admitted he wanted to pay a developer a fair base salary and then front another $250,000 a year just for token usage. That’s not a rounding error. That’s strategy. The hosts also cited soaring usage at the biggest players and across enterprises.
“Meta last year said they burned through a billion tokens in a single month.”
“The average enterprise, they’re burning through 13 times more tokens this year than they were last year.”
Let that sink in. A billion tokens in a month. Thirteen times more year over year. This isn’t about a few prompts gone wild. It’s a new line item that rivals engineering salaries and ad spend. And it’s growing.
What The Hosts Get Right
The Marketing Against the Grain crew are not scolding teams for using AI. They’re surfacing a reality: usage-based AI is addictive. Once you taste the speed and volume, you want more. So do your customers. That’s fine—if your unit economics make sense. It’s a problem when you scale inference before you’ve tied it to outcomes.
As a marketer and builder, I agree with their signal. Burn rates without guardrails turn into tax bills. You wouldn’t let your ad platform auto-spend without a cap. Why let your models do it?
Yes, You Should Spend—But With Rules
There’s a fair counterpoint: you can’t learn without spending. True. I’ve shipped plenty by paying the “tuition.” But education spending has a syllabus. Token budgets need one too. Here’s how I’d set it up for a product team or a growth org.
- Define cost per outcome. Tie tokens to revenue, leads, or hours saved. No metric, no scale.
- Start with smaller models. Upgrade only if results demand it. “Bigger” is not a strategy.
- Cache and reuse. Don’t pay twice for the same answer. Store results where you can.
- Batch jobs during off-peak. Latency matters less for back-office work than your wallet.
- Put hard caps on services. Alerts at 50%, 75%, and 90% of budget. Stop the fire before it spreads.
- Reward efficiency in reviews. Make token thrift part of performance, not an afterthought.
- Measure cost per conversation for chat tools. If a support bot costs more than a human, rethink it.
These steps aren’t theory. They’re the same controls we learned from cloud and ads. AI just makes the burn rate feel invisible because there’s no loud “publish” moment. It’s a trickle—until it’s a flood.
The Salary-Plus-Token Model
The “salary plus $250k token pool” idea is bold. I like the intent. It tells builders, “ship with AI at the core.” But I’d tune it. Tie that pool to milestones and usage quality. Spend more when cost per win drops. Spend less when it rises. And give teams a slice of the savings they create by tuning prompts, choosing the right model, or pruning waste.
Also, move from “how many tokens” to “which tokens.” Some tasks work best with embeddings. Some with fine-tuning. Some with local inference. Mix the stack. Don’t just push everything to the most expensive endpoint because it’s easy.
What Most Teams Miss
The mistake I see is treating AI like a limitless brain in the sky. It’s not. It’s rented intelligence with a usage tax. The smartest teams think like CFOs and product managers at the same time. They test, measure, and then scale where the math works.
Could you starve innovation by being strict? Sure. But waste kills speed, too. Nothing slows a roadmap like a surprise bill that forces a freeze. Guardrails don’t block progress. They make it repeatable.
My Take For Builders And Marketers
I’ve made a career spotting where digital leverage meets business sense. This is one of those moments. The hosts are right to flag the spike in token spend, and they’re right to talk about it as a choice. If AI is the engine, tokens are the fuel. Plan the trip before you floor it.
Here’s my challenge to you: audit your AI flows this week. Find one process where you can cut token spend by 30% without hurting outcomes. Cache, compress, or switch models. Then push the savings into the one AI feature that actually drives growth. That’s how you turn burn into lift.
Don’t let your margins die by a thousand prompts. Set the rules, tie spend to wins, and build with intent. The future isn’t who burns the most tokens. It’s who gets the most value per token burned.
