If the AI bubble bursts, marketing could take the first hit
"How do we know when irrational exuberance has unduly escalated asset values?" That question from a 1996 speech keeps echoing. Different decade, same pattern: hype, cheap money, and big promises. This time, it's AI - and marketing sits right under the blast radius.
According to reports, roughly $1.2 trillion in debt is tied to AI-related companies. That's a historic bet on future productivity that hasn't shown up across the broader economy. One economist estimated GDP growth outside data centers was just 0.1% in the first half of 2025. Remove AI infrastructure, and growth almost vanishes.
Meanwhile, industry leaders say "everybody's going to do super well." That's morale-building, not a plan. Markets don't reward group optimism forever. They care about cash flow, real demand, and durable business models.
Why marketers should care
Our stacks now depend on a small set of foundational models. Many tools are wrappers around the same APIs. That creates a brittle system: if access tightens, prices jump, or outages spread, your entire operation can stall.
Think about what relies on those services today: predictive audiences, content generation, creative testing, dynamic personalization, product recommendations, even media mix modeling. Pull the plug on the core model and watch the dominoes fall.
There's another risk: circular mega-deals. Vendors buy each other's capacity, book revenue, and use that "demand" to justify valuations. If financing tightens, consolidation hits fast. You could wake up to unsupported platforms, forced migrations, or emergency price hikes.
Early warning signs to watch
- Sustained API price increases, stricter rate limits, or new usage tiers.
- Frequent model changes that break prompts or outputs.
- Outages, degraded latency, or lower quality with no clear timeline to fix.
- Vendors delaying roadmap items or pushing long contracts with upfront payments.
- Legal or regulatory pressure that restricts training data or deployment.
- Data center supply strain that pushes costs back onto customers.
A practical hedge plan for marketing teams
Hope is not a strategy. Here's a concrete playbook you can start this quarter.
1) Map your AI dependency
- Inventory every workflow using AI: ideation, copy, images, video, targeting, scoring, analytics, personalization.
- For each tool, document the underlying model, provider, region, and contract terms. Ask vendors to disclose model dependencies in writing.
2) Build a multi-model option
- Create an abstraction layer so you can switch models with one config change.
- Support at least two base models for critical tasks. Keep a smaller open-source model as a last-resort fallback for essential flows.
3) Design graceful degradation
- Define tiered fallbacks: AI → rules-based → manual. Don't let campaigns go dark.
- Cache reusable outputs (product copy, FAQs, headline variants) and version them so you can ship even during outages.
4) Control your data moat
- Keep first-party data clean, centralized, and queryable without AI.
- Maintain a vetted knowledge base and style guides that models reference. If the model weakens, your data still carries the quality.
5) Cut vendor concentration risk
- Score vendors by financial health, AI dependency, and single-provider exposure.
- Negotiate SLAs with credits for downtime, exit clauses, and data portability guarantees.
6) Budget for volatility
- Model scenarios: +50% API costs, 25% throughput cuts, 48-hour outage.
- Pre-approve shift plans: reduce frequency, swap models, switch channels, or pause non-essential AI features.
7) Lock your prompts and patterns
- Version-control prompts. Document expected inputs/outputs with test cases.
- Add automated evals so you catch quality drift before it hits customers.
8) Prep your team
- Train staff to run the playbook without AI for 72 hours: briefs, copy, QA, and approvals.
- Run a fire drill once a quarter. Measure time-to-publish and error rates.
9) Legal and compliance ready
- Track content provenance. Keep logs of prompts and outputs for claims or takedowns.
- Document data usage consents independent of any model provider's policy shifts.
10) Procurement and contracts
- Require vendors to disclose model swaps and give 60-90 days' notice before material changes.
- Prohibit training on your data unless explicitly approved and compensated.
Your 30/60/90-day checklist
Next 30 days
- Complete the AI dependency inventory and vendor exposure map.
- Add a secondary model for your top two revenue-critical workflows.
- Version-control all prompts and create a minimum eval suite.
Next 60 days
- Negotiate SLAs and exit clauses with your top five AI-reliant vendors.
- Implement caching for recurring assets and set rule-based fallbacks.
- Run a 48-hour "AI off" drill for content and lifecycle campaigns.
Next 90 days
- Ship the abstraction layer so model switching is one-click.
- Finalize budget scenarios and pre-approved channel shift playbooks.
- Publish an internal AI change policy: when to pause, switch, or scale back.
If the boom turns to a bust
If access tightens or vendors fold, survival will favor teams that can keep publishing, keep buying media with confidence, and keep personalizing on basics while others reboot. That comes from redundancy, clarity, and the will to simplify.
One more thing: don't overfit your strategy to a single supplier's roadmap. Optimism is not a guarantee. As the 1996 warning suggests, exuberance eventually meets physics. Plan like it will.
Greenspan's 1996 speech is a useful reminder that sentiment cycles. Build your marketing machine so it runs either way.
If you want structured, vendor-agnostic upskilling for your team, this certification can help: AI Certification for Marketing Specialists.
Your membership also unlocks:
