AI investment is accelerating faster than any technology wave before it. The pattern underneath it is older than most of us realize.
There is something deeply human about looking up.
We’ve always done it. We looked at the stars and decided we belonged there. We looked at the horizon and built ships to cross it. We looked at problems that seemed impossible and found ways to make them less so. That instinct, the refusal to accept limits as permanent, is responsible for most of what we’ve built.
It’s also responsible for a pattern that keeps repeating. And we never seem to learn it.
Every generation gets a technology that feels like the one. The thing that will finally close the gap between what we can do and what we’ve always wanted to do. Computers. Then software. Then the internet. Then big data. Each one arrived with genuine capability and genuine promise. Each one attracted enormous investment. Each one produced real gains, new industries, new efficiencies, new possibilities that didn’t exist before. And each one, eventually, ran into the distance between what the demo promised and what the real world required.
That gap never closed the way we expected. So we did what we always do.
We threw more fuel at it.
Escape Velocity
The dream underneath every one of these waves is the same. Productivity that permanently outpaces human cost. Technology that does the work so we don’t have to. A world where the hard problems are solved and we get to focus on the things that make us human. Call it escape velocity: the point where the thrust of technology finally exceeds the drag of human limitation, and we break free.
It’s not a foolish dream. It’s the engine of progress. The desire to reach it has produced antibiotics, satellites, search engines, and smartphones. I’m not arguing against it.
But here’s what I keep watching. Every time a new technology arrives that seems capable of getting us there, we stop looking around and start looking up. The rocket is so compelling that we forget to ask whether we’ve learned anything from the last launch.
We haven’t. Not really. And we’re doing it again.
I know the objection. This time is different. The capability curve on foundation models is steeper than anything we’ve seen before. The pace of improvement is unprecedented. And that’s true — the technology is genuinely remarkable. But “this time is different” is also what people said in 1999 about the internet changing everything, and in 2012 about big data unlocking insights we’d never had access to before. The technology was remarkable then too. Remarkable capability and a repeating pattern are not mutually exclusive. In fact, the more remarkable the technology, the more fuel we pour in, and the harder it becomes to ask whether we’re pointed in the right direction.
What We Didn’t Learn
The dot-com era taught us that capability isn’t the same as a business model. Companies built genuinely impressive technology and then discovered there was no sustainable path from the demo to the balance sheet. The ones that survived were the ones that eventually answered the question nobody had asked at the start: how does this actually create value for someone willing to pay for it?
The big data era taught us that having more information isn’t the same as making better decisions. Organizations accumulated enormous data sets, built sophisticated pipelines, hired analysts, and still found themselves debating which number to believe in the same meetings they’d always had. The insight didn’t automatically follow the data. Someone still had to ask the right question.
Each wave felt categorically different from inside the rocket. Each one ran into the same underlying gap: between what the technology could do and what the problem actually required.
AI is not exempt from that gap. It has just made it more expensive to discover.
The Fuel Problem
More than 142,000 U.S. tech workers lost their jobs in the first five months of 2026, a 33% increase over the same period in 2025, as four hyperscalers committed to a combined $700 billion in capital expenditure, nearly double what they spent the year before.1
The companies doing the cutting aren’t doing it because AI has replaced the people being let go. Most current layoffs are about saving money to invest more heavily in AI development. It’s a resource reallocation decision framed as an operational efficiency decision.2
In other words: we’re jettisoning weight to generate more thrust.
When you need to go faster and you’re running low on fuel, you start asking what you can throw overboard. Headcount is the heaviest thing on the balance sheet. So it goes first. The logic feels sound in the moment: leaner, faster, more resources pointed at the thing that matters. And there’s always the sense that escape velocity is close. Just a little more thrust. Just a little more investment. We’re almost there.
What we tend to forget is what we’re throwing overboard.
The people being cut aren’t dead weight. They’re the ones who know the customers. Who understand the systems. Who carry years of institutional knowledge that isn’t written down anywhere, not in a document, not in a line of code, not in a model trained on publicly available text. It exists in their experience, their judgment, their memory of why the last three attempts at this failed. When they leave, that knowledge walks out with them. And the next team will spend months rediscovering what they already knew.
I’ve seen this before. Every company has. We just don’t seem to remember it when the next launch window opens.
The Picks and Shovels Problem
There’s a wrinkle in the current cycle that makes it different from the ones before it.
During the Gold Rush, the people who reliably made money weren’t the prospectors. They were the ones selling picks, shovels, and denim pants. The prospectors funded the ecosystem. Most of them came home with less than they started with.
The companies executing the largest AI cuts are explicitly redirecting payroll savings toward AI infrastructure: data centers, GPUs, energy, and top AI talent. That infrastructure is being built and sold primarily by a handful of companies, the hyperscalers, the chip manufacturers, the model providers. They win whether or not the companies buying their services achieve escape velocity. The fuel gets purchased either way.
In 2025, enterprises poured $684 billion into AI. By year-end, more than $547 billion of that investment had produced no measurable results.3
The picks and shovels are selling. The gold is harder to find than it looked.
Conversational AI and the Receding Destination
One of the most instructive examples is playing out in customer experience right now. It’s not new. It’s the latest version of a familiar story.
Companies have been trying to replace human customer service with technology for decades. First with IVR systems. Then with rule-based chatbots. Then with NLP-driven virtual agents. Each generation of technology seemed like the one that would finally close the gap. Each one required more investment than expected. Each one revealed new layers of complexity that the demo hadn’t shown.
LLMs felt different. They could actually hold a conversation. They could understand context, handle ambiguity, respond naturally. The demos were genuinely impressive.
Klarna went further than most, deploying AI to handle the majority of their customer service interactions. Then they pulled back from full automation, acknowledging that for complex, high-stakes interactions, real people offer something the technology couldn’t match: empathy, understanding, and genuine service at the moments that mattered most.4 IBM invested over four billion dollars in Watson for Oncology before quietly scaling it back. The technology worked. The gap between what it could do in controlled conditions and what the real problem required turned out to be larger than anyone had modeled.
I’ve seen the same pattern in companies I’ve worked with. The financial services customer who doesn’t want to reset a password. They want to talk about their retirement account, their specific situation, their fears about whether they’re going to be okay. The cable customer who isn’t asking a generic question. They’re asking their question, with their history, their frustration, their context. LLMs are good at language. They’re not good at knowing you. And the more personal and high-stakes the interaction, the wider that gap becomes.
The response, when the gap appears, is familiar. Bring in a bigger vendor. Spend more. Hire consultants. Add another layer. The destination keeps appearing close. The distance doesn’t close. Just a little more fuel.
Are You Using Your Fuel for the Right Things?
The desire to reach escape velocity isn’t the problem. It’s the engine of everything worth building. The problem is what we sacrifice in the attempt, and whether we’re pointed at the right destination.
While companies are burning fuel chasing the moonshot, the real problems sit unaddressed. The friction in the customer journey that could be meaningfully reduced with a more targeted application of AI. The internal process that wastes forty-five minutes of someone’s day, every day, and could be automated with a fraction of the budget going into the next big bet. The place where AI could shift the customer experience in a way that is concrete, measurable, and real, not because it achieved escape velocity, but because it solved an actual problem.
These aren’t consolation prizes. They’re the things that build durable value. They compound. They create the foundation for the next stage. And they don’t require betting the whole ship on a destination that keeps receding.
Before your next launch, I’d ask you to do two things.
Look back. Pull up the last major technology initiative your organization ran. Not the AI one. The one before it. What was promised? What was delivered? Where did the gap appear, and why? What did you learn from it, and did that learning actually change how you approached the next one? If the honest answer is that it didn’t, that’s worth sitting with before you commit the next round of budget.
Then look around. Not at the sky. At the problems in front of you. Ask three questions. What are the most painful unsolved problems your clients have right now? Where in your business is time or money being lost in ways that are specific, measurable, and well-understood? And for each AI investment currently on your roadmap, can you write a single sentence describing what success looks like and how you’ll know when you’ve reached it?
If you can’t answer that last one, you’re not ready to launch.
The dream is worth keeping. The pattern is worth breaking.
We keep building the same rocket. Maybe it’s time we studied the last crash before we launch again.
- TechTimes, May 29, 2026 — https://www.techtimes.com/articles/317392/20260529/tech-layoffs-reach-142000-2026-profitable-companies-cut-jobs-fund-700b-ai-infrastructure.htm ↩︎
- CEO Reporter — https://ceoreporter.com/tech-firms-cutting-jobs-to-fund-ai-not-because-ai-replaced-workers/ ↩︎
- Folio3 AI, April 24, 2026 (citing RAND Corporation and BCG) — https://www.folio3.ai/blog/ai-project-failure-rate-stats
↩︎ - AnswerConnect, May 2025 — https://www.answerconnect.com/blog/business-tips/ai-customer-service-disasters/ ↩︎