ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

Hello, fellow humans! Today, we’re going to try to look into the future. Of course, predictions are always wrong. As The Oracle tells us, “no one can see beyond a choice they don't understand. And I mean no one.” There are no inevitabilities; there are only humans making choices that are rational strictly from their own perspectives. But we’ll try to understand what decision-makers will likely do, based on the history of startups and tech in Silicon Valley, and what seems rational to that very narrow perspective.

What Comes Next in AI: Part 1 of 2

The fog of AI hype is slowly lifting, and as the road ahead comes into clearer view, we are all seeing that most LLM companies are burning billions with no path to profitability. The question keeping people up at night isn't whether there's going to be a correction—it's what that correction will look like and who gets hurt.

Just last week, Gary Marcus told Axios ”the return on investment has not been that high. The costs are very high. Nobody except Nvidia is making all that much money. The big players are losing money."

I've spent the last few months tracking the circular financing patterns in AI infrastructure, watching companies raise massive rounds while their unit economics get worse, not better. The math doesn't work. A $600 million fundraise at a $20 billion valuation doesn't change the fact that you're losing money on every API call. That doesn't mean that the AI is fake or that the capabilities aren't real, but we are heading for a market restructuring that will fundamentally change who captures value in this space.

This is actually a rare moment in tech—we can see the transition coming and position ourselves accordingly rather than scrambling after the layoffs hit. So let's walk through what's most likely to happen, who's vulnerable, and who's positioned to thrive. This is part one of a two-part series. Today we're diagnosing the situation and understanding where different types of tech workers stand. Next time, we'll get tactical about where the real opportunities are and how to position yourself to capture them.

The most likely scenario isn't a spectacular crash. It's something more mundane but equally consequential: a series of acqui-hires combined with asset liquidations, and it will go something like this.

A mid-tier LLM company—let's say one that raised $500 million and is burning through it at $30-40 million per month—realizes they can't make the unit economics work. They're competing with OpenAI, Anthropic, and Google, all of which have deeper pockets and better distribution. The board starts having uncomfortable conversations about runway and exits. They approach Microsoft or Google or Meta about an acquisition.

The deal gets done at a modest valuation, maybe 30 to 50 cents on the dollar for investors. The acquirer is mostly interested in the team—particularly the top researchers and engineers who've actually debugged training runs at scale. Those people get retention packages. Everyone else gets 60 to 90 days. Meanwhile, the GPU clusters get liquidated to cloud providers at a 40 to 50 percent discount. The acquirer takes some capacity for their own use, but most of it floods the secondhand market.

This doesn't happen to just one company. Over 18 months, this pattern repeats across 4 to 6 significant AI companies. Each one reinforces the pattern. VC funding for AI infrastructure drops 70 percent because the exit strategy just got validated as "get acquired by Big Tech at a discount or die."

Now here's where it gets interesting from a systemic perspective. Remember those circular financing patterns I mentioned? Nvidia has been investing in AI companies that buy Nvidia chips. When those companies get acquired at steep discounts or liquidate their assets, Nvidia's investment portfolio takes a hit and the expected revenue from those customers scaling up evaporates. This triggers a repricing across the entire AI infrastructure stack.

But this isn't a catastrophe. It's normalization. AI shifts from "land grab at any cost" to "prove your unit economics." Foundation models commoditize somewhere around GPT-4.5 or Claude 3.5 level capability. Improvements become incremental—better speed, lower cost, slightly better reasoning—rather than the qualitative leaps we saw from GPT-3 to GPT-4.

The companies that survive this transition aren't necessarily the ones with the best models. They're the ones with actual distribution, clear use cases, and demonstrated ability to generate profit per inference. In other words, the boring stuff that venture capitalists spent two years ignoring suddenly matters again.

Here's what this means for tech workers. The industry is going to stratify into three distinct tiers, and where you fall determines your next five years.

The first tier is the protected class—maybe the top 10 percent of people currently working in AI. These are the elite researchers who've published breakthrough papers, the engineers who've actually debugged training runs at frontier scale, and the people with deep relationships inside Big Tech. When acqui-hires happen, these people get bidding wars. Their compensation might actually go up because they're insurance policies. Big Tech companies acquire them partly for their capabilities, but mostly to keep them away from competitors.

If you're in this tier, you probably already know it. You're getting pinged by recruiters constantly, your DMs are full of "quick coffee?" messages from VPs at FAANG companies, and you're sitting on multiple term sheets. Your positioning strategy is straightforward: negotiate for equity in the acquiring company plus significant retention cash. Your leverage is highest in the 90 days immediately after an acquisition announcement when the acquirer is most worried about talent flight.

The second tier is what I call the pivot generation—the middle 40 percent. These are solid ML engineers, infrastructure specialists, MLOps people, product managers who've shipped AI features, and applied researchers who are good but not groundbreaking. Your jobs are going to disappear, but your skills transfer if you reframe them correctly.

The hard truth: compensation is going to reset. Those $500,000 total comp packages for ML engineers are going to drop to $250,000 to $300,000, basically matching senior software engineering roles at Big Tech. The premium for "AI skills" is going to evaporate because the skills are becoming table stakes rather than specialized expertise.

But here's the opportunity that most people are missing: you need to shift from "building models" to "applying models to domain problems." The person who wins in this tier isn't the one with the deepest knowledge of transformer architectures. It's the person who spent 10 years in healthcare and then spent 2 years learning ML. That combination is actually more valuable than pure ML engineering because the models themselves are commoditizing.

This is where I think most of the opportunity actually lives, and we'll dig into this much more in part two. The key insight is that domain expertise becomes the differentiator when the technical capabilities become widely available. You want to be the person who understands both the business problem and has enough ML knowledge to be dangerous.

The third tier is the stranded population—probably the bottom 50 percent. These are people who got hired during the hype cycle without deep technical skills. Prompt engineers without programming ability. ML engineers who only know how to fine-tune existing models. AI ops specialists for infrastructure that's getting consolidated into cloud platforms. Marketing people who pivoted to "AI growth" roles.

The brutal reality is that these roles were created by temporary market dynamics, not sustainable demand. When the correction happens, these jobs disappear and the skills don't transfer particularly well. If your entire value proposition is "I work with AI tools," you're competing with literally everyone else who can learn to use Claude or ChatGPT. That's not a moat. That's not even a speed bump.

I don't say this to be cruel. I say it because if you're in this category, you have maybe 12 to 18 months to reposition yourself, and pretending the risk isn't real doesn't help anyone. The good news is that repositioning is possible, but it requires honest assessment of which direction to move.

The important thing to understand is that this stratification isn't primarily about technical depth. It's about transferability and sustainability of the value you create.

If your value depends on proprietary access to frontier models or on the specific infrastructure of an LLM company that's burning cash, you're vulnerable regardless of your technical skills. If you can articulate your value in terms of business outcomes—cost reduction, efficiency gains, revenue impact—rather than technical capabilities, you're more resilient. If you're building skills that transfer across the AI hype cycle—domain expertise, systems thinking, product judgment—you're positioned for a 20-year career rather than riding a 3-year wave.

The mistake I see people making is treating "AI expertise" as an identity rather than as a toolset. It's the same mistake people made with "mobile apps" in 2011 or "blockchain" in 2017. The people who succeeded long-term weren't the ones who built careers on the technology itself. They were the ones who used the technology to solve real problems in specific domains.

This is the fundamental reframing that needs to happen: AI is transitioning from product category to infrastructure layer. When that transition completes, the value doesn't disappear—it shifts. It moves from the people building the infrastructure to the people using it effectively to transform workflows and business models.

In part two, we're going to get specific about where those opportunities are. We'll talk about the four opportunity zones that are opening up during this transition, the concrete positioning strategies for different career stages, and why a market correction might actually accelerate real-world AI adoption. We'll also address the contrarian take: this could be the best thing that happens to AI as a field because it will force everyone to focus on actual value creation rather than fundraising and hype.

For now, the question you should be asking yourself is: which tier am I in, and what do I need to do to position myself for what comes next? Don't wait for the layoffs to figure this out. The people who thrive through market transitions are the ones who see them coming and move preemptively.

Next time: Where the real opportunities are, and your positioning playbook for navigating the shift.