Last week’s market volatility served as a reminder that animal spirits remain alive and well. After months of AI stocks climbing with almost athletic confidence—vaulting over every earnings expectation, rumor, or leaked demo—the sector finally took a step back. The sell-off wasn’t surprising to analysts who have been warning about Silicon Valley’s circularity problem: each time Nvidia, OpenAI, Oracle, AMD and others invested in one another, their stock prices rose in unison. If the industry is going to justify the capital already invested, we need roughly $2 trillion in new value creation.

So the market did what markets do: it rationalized investments. A correction is not a crisis. Just because investors sobered up a bit doesn’t mean the underlying technology has lost momentum. But it does confirm what many strategists have been whispering—AI valuations have been running ahead of actual adoption. Goldman analysts have hinted at consistent overinflation, and Gartner estimates that fewer than 30% of companies are using GenAI in any meaningful capacity. When stock prices sprint faster than operational reality, gravity eventually appears and taps us on the shoulder.

Not quite. But we might be over our skis just a bit.

A true bubble implies fantasy-land valuations—companies with no revenue, no customers, and no clear path to either. The dot-com era is the classic example: a time when a startup could add “.com” to its name and watch its market cap triple overnight.

AI is different. The technology works. The infrastructure is real. And the economic impact—though wildly uneven—is measurable. Chips, data centers, GPUs, distributed compute, LLM training cycles, and enterprise APIs represent capital-heavy, tangible commitments. Billions are being spent on things you can touch, plug in, and accidentally burn your hand on.

We dispense business advice, and not investment advice. Shying away from making AI business investments because of a stock market reaction, would be an overreaction. So here is the best practice, figure out how to measure the ROI of your AI effort within your organization.

The Dot-Com Parallel: Relevant, But Not Fatalistic

Yes, the internet bubble burst. And yes, markets reacted violently between 2000 and 2002. But since then, the S&P 500 has returned roughly 9% per year on average. And ironically, many of today’s most important technologies were born from the excesses of that boom. Cloud computing, E-commerce, Smartphones, Social platforms- all built on the infrastructure funded by what, in hindsight, was an unruly period of irrational exuberance.

Meanwhile, the technology itself refuses to slow down. GPT-5 provides more sophisticated multimodal reasoning and the early contours of agent-like autonomy. Google’s Gemini has vaulted forward, with performance equal to or greater than ChatGPT. These technologies are already creeping deeper into workflows, showing up in customer operations, finance teams, codebases, and internal knowledge systems. Enterprise adoption may be slower than investor enthusiasm, but the trend line is unmistakable.

Layer in the buildout of global data center capacity, the explosion of enterprise APIs, and ongoing breakthroughs in model efficiency, and it becomes clear: markets may wobble, but the technological curve is still bending upward.

We are mere toddlers in AI adoption. If anything, we encourage our clients to pick up the pace and have more urgency in deploying AI.

Structured data is the orderly, rule-driven side of your information. It sits in clean rows and columns—ERP tables, CRM records, financials, inventory logs. Because every field follows a defined format, it’s easy to search, sort, validate, and analyze. This is the data you can plug directly into dashboards, forecasts, and BI tools without breaking anything. Executives trust it because it’s predictable and consistent.

Unstructured data is everything that doesn’t fit that mold—emails, PDFs, customer reviews, videos, Teams messages, transcripts, proposals, jobsite photos, even the text buried inside contracts. It’s rich with insights, but chaotic. To make it useful, you need AI and natural-language tools to interpret it, organize it, and connect it back to your structured systems.

Actionable takeaway: If you want reliable AI, you must fix your data house first. Build data integrity by standardizing naming conventions, cleaning up duplicates, enforcing metadata, tightening access rules, and eliminating “shadow systems” where people store their own versions of truth. Create one governed source of record for each major dataset. Once your structured and unstructured data align—clean, labeled, and findable—your AI deployments become dramatically more accurate, more trustworthy, and more valuable.

The hosts of Morning Brew Daily sit down with the president of the Federal Reserve Bank of Chicago Austan Goolsbee to discuss how the country’s central bank is managing the data blackout from the government shutdown.