Separate two questions that get mashed together. Are AI company valuations frothy? Very possibly - hundreds of billions in data-centre capex chasing uncertain returns, revenue multiples that assume flawless execution, and prominent investors including OpenAI’s own Sam Altman saying parts of the market are "overexcited." Is AI capability a bubble? No - the tools already write working code, pass professional exams, and handle real support queues, and companies are already reorganising hiring around that fact (entry-level routine postings down double digits; AI-capable juniors earning a documented RM 6,000-9,000/month premium in Malaysia). The dot-com crash is the template: pets.com died, the internet did not. Bubbles kill overvalued companies, not underlying general-purpose technologies.

What happens to AI jobs if the bubble bursts?

The dot-com precedent is specific and reassuring on this: after the 2000 crash wiped out ~78% of the NASDAQ, tech employment recovered and surpassed its bubble peak within a few years - because companies still needed websites, e-commerce, and software regardless of stock prices. The people hurt worst were those betting on speculative employers; the people who did fine were those with real, transferable skills who moved to the survivors. An AI correction would follow the same shape: hype-stage startups die, but the banks, telcos, and retailers deploying AI for cost reasons keep deploying it - a correction makes cost-saving automation MORE attractive, not less. Skills survive crashes; tickers don’t.

Should I still learn AI skills if it might be a bubble?

Yes - and the bubble scenario is, counterintuitively, an argument FOR learning rather than against. Run both branches. If AI keeps compounding: AI skills are obviously the right bet. If valuations crash: companies get cost-obsessed, cost-obsessed companies automate harder, and the engineers who survived the dot-com crash were the ones who could actually build - exactly the fundamentals-first profile. The only scenario where learning AI-era building skills loses is one where AI capability itself vanishes, and nothing in the technology’s trajectory - or in how businesses already depend on it - supports that. A bet that pays in both branches isn’t a gamble.

How is the AI boom different from the dot-com bubble?

Two big differences, one in each direction. In AI’s favour: revenue is real and immediate - the dot-com era ran on eyeballs and hope, while today’s AI leaders book tens of billions in actual sales and the tools demonstrably do economic work now. Against AI: the capital intensity is far larger - data-centre buildouts run into hundreds of billions, so if returns disappoint, the write-downs would be historic. Both differences point to the same practical conclusion: the technology is more obviously real than the internet was in 1999, while the financial froth around it may be bigger. Which is precisely why you bet on the skill layer, not the stock layer.

What should I do differently because of bubble risk?

Three concrete adjustments. First, when job-hunting, weight employers by AI usefulness rather than AI branding - a bank automating loan processing survives a crash; a thin wrapper startup with no revenue may not. Second, learn fundamentals underneath the AI tooling rather than tool-specific tricks: frameworks and vendors churn in a correction, but the ability to build, debug, and evaluate software transfers to whatever survives. Third, ignore both hype and doom in your feed and watch what companies do with money - hiring premiums for AI-capable builders and enterprise AI budgets are the signal; funding-round theatre is the noise.

All posts

Perspective · AI & your future · Careers

Is AI a bubble? The stocks might be. The shift is not.

You’ve seen both feeds: one says AI changes everything, the other says it’s tulips with GPUs. Both cite real evidence, which is why the argument never resolves. The resolution is to separate two questions everyone mashes together - and notice that the smart career move is identical in both scenarios. A coding school has an obvious bias here, so this piece argues from the crash scenario, not the boom one.

Deric Yee Updated 25 August 2026 10 min read

Two questions wearing one trench coat

“Is AI a bubble?” is actually two questions. Question one: are AI company valuations inflated?Quite possibly. The honest bear case is strong: hundreds of billions of dollars in data-centre capital expenditure chasing returns nobody can yet prove, startups valued at extreme revenue multiples, and - the detail worth remembering - OpenAI’s own CEO Sam Altman saying out loud in 2025 that investors as a whole were “overexcited” and parts of the market would burn people. When the people selling the boom warn about the froth, take the froth seriously.

Question two: is AI capability inflated? Here the evidence points the other way, and it isn’t vibes - it’s payroll. Models in 2026 write working production code, pass professional licensing exams, and handle live customer queues. More telling than benchmarks is what employers do with money: entry-level routine postings fell 15% in the US and 29% globally in a year, AI was the #1 named layoff reason in March 2026 (~25% of announced cuts), and Malaysian employers pay AI-capable juniors RM 6,000–9,000/month against RM 3,500–6,500 for traditional profiles - the full dataset is in our State of AI Hiring report. Companies don’t restructure hiring around a capability that doesn’t work. So the precise answer: the financial layer may well be a bubble; the capability layer is a payroll line item. History has a name for exactly this combination.

What the dot-com crash actually teaches

The year 2000 is the perfect controlled experiment, because both sides of today’s argument were right back then too. The bears were right: the NASDAQ fell roughly 78% from peak to trough, pets.com and hundreds of eyeball-valued companies evaporated, and trillions in paper wealth vanished. And the bulls were right: the internet did exactly what its evangelists promised- within a decade of the crash it had rebuilt commerce, media, banking, and communication, and the era’s giants (Amazon down 90%+ in the crash, Google IPO-ing into the rubble in 2004) were built or proven in the wreckage. The bubble was real AND the technology was real. Bubbles kill overvalued companies; they have never once killed a working general-purpose technology.

Now the part that matters for you - what happened to people. Tech employment fell after 2000, then recovered and surpassed its bubble peak within a few years, because crashed stock prices didn’t change the fact that every company still needed websites and software. Sort the individuals and a clean pattern appears: the ones hurt worst held speculative equity or worked at eyeball-stage startups with no real product; the ones who did fine - often better than fine - had genuine engineering skills and simply moved to the survivors, where a decade of demand was waiting. The crash sorted people by whether their value was priced in stock or carried in skill. That sorting rule is the entire practical content of bubble history, and it’s the same rule the spreadsheet era enforced on our parents’ generation - the frame we lay out in AI is the new Excel: the technology’s arrival was never optional; only your preparation was.

Run both branches of your decision

Branch A: no crash, AI compounds

The premium for people who can direct AI and ship real software keeps widening - the RM 6,000–9,000 junior band is the floor, not the ceiling. Learning the skills now is obviously correct; every month of delay is paid in a steeper catch-up later. Nobody disputes this branch. The whole argument is about branch B.

Branch B: valuations crash hard

Speculative startups die and funding theatre stops - but surviving companies turn cost-obsessed, and cost-obsessed companies automate harder, because by then AI is cheap commodity infrastructure (the models don’t un-learn to code when the NASDAQ falls). Demand shifts toward people who can build reliable systems on that infrastructure - exactly the fundamentals-first, post-2000-engineer profile. Skills appreciate in this branch too.

A bet that pays in both branches isn’t a gamble

That’s the resolution to the argument in your feed: you don’t need to know whether AI is a bubble, because the skill-layer decision is identical in both branches. The only losing scenario is one where AI capability itself disappears - and nothing in the technology’s trajectory, or in how thoroughly businesses already depend on it, supports that. What bubble risk does change is the how: prefer employers with AI usefulness over AI branding (a bank automating loan-processing survives a crash; a wrapper startup may not), and build fundamentals under the toolingrather than tool-tricks on top of it - vendors and frameworks churn in a correction, but the ability to build, debug, and evaluate software transfers to whatever survives. That’s not a marketing line; it’s our entire pedagogy, argued in full in vibe coding vs fundamentals - and it was formed by exactly this history: shallow skills die with their bubble; deep skills move to the survivors.

One more honest disclosure: a school that teaches AI-era development profits if you decide to learn, so discount our conclusion however you like - but notice the conclusion doesn’t depend on our optimism. It depends on the dot-com employment record, on what a correction does to automation budgets, and on hiring data you can check yourself in the salary guide. And the cost of testing the conclusion is zero: the free trial - one signup, no card - lets you start building the crash-proof layer tonight, with real projects and a live instructor session, before you spend a single ringgit. If you’re still weighing when to move, is now a good time to get into tech runs the timing question with the same both-branches logic.

FAQ

Is AI a bubble?
Separate two questions that get mashed together. Are AI company valuations frothy? Very possibly - hundreds of billions in data-centre capex chasing uncertain returns, revenue multiples that assume flawless execution, and prominent investors including OpenAI’s own Sam Altman saying parts of the market are "overexcited." Is AI capability a bubble? No - the tools already write working code, pass professional exams, and handle real support queues, and companies are already reorganising hiring around that fact (entry-level routine postings down double digits; AI-capable juniors earning a documented RM 6,000-9,000/month premium in Malaysia). The dot-com crash is the template: pets.com died, the internet did not. Bubbles kill overvalued companies, not underlying general-purpose technologies.
What happens to AI jobs if the bubble bursts?
The dot-com precedent is specific and reassuring on this: after the 2000 crash wiped out ~78% of the NASDAQ, tech employment recovered and surpassed its bubble peak within a few years - because companies still needed websites, e-commerce, and software regardless of stock prices. The people hurt worst were those betting on speculative employers; the people who did fine were those with real, transferable skills who moved to the survivors. An AI correction would follow the same shape: hype-stage startups die, but the banks, telcos, and retailers deploying AI for cost reasons keep deploying it - a correction makes cost-saving automation MORE attractive, not less. Skills survive crashes; tickers don’t.
Should I still learn AI skills if it might be a bubble?
Yes - and the bubble scenario is, counterintuitively, an argument FOR learning rather than against. Run both branches. If AI keeps compounding: AI skills are obviously the right bet. If valuations crash: companies get cost-obsessed, cost-obsessed companies automate harder, and the engineers who survived the dot-com crash were the ones who could actually build - exactly the fundamentals-first profile. The only scenario where learning AI-era building skills loses is one where AI capability itself vanishes, and nothing in the technology’s trajectory - or in how businesses already depend on it - supports that. A bet that pays in both branches isn’t a gamble.
How is the AI boom different from the dot-com bubble?
Two big differences, one in each direction. In AI’s favour: revenue is real and immediate - the dot-com era ran on eyeballs and hope, while today’s AI leaders book tens of billions in actual sales and the tools demonstrably do economic work now. Against AI: the capital intensity is far larger - data-centre buildouts run into hundreds of billions, so if returns disappoint, the write-downs would be historic. Both differences point to the same practical conclusion: the technology is more obviously real than the internet was in 1999, while the financial froth around it may be bigger. Which is precisely why you bet on the skill layer, not the stock layer.
What should I do differently because of bubble risk?
Three concrete adjustments. First, when job-hunting, weight employers by AI usefulness rather than AI branding - a bank automating loan processing survives a crash; a thin wrapper startup with no revenue may not. Second, learn fundamentals underneath the AI tooling rather than tool-specific tricks: frameworks and vendors churn in a correction, but the ability to build, debug, and evaluate software transfers to whatever survives. Third, ignore both hype and doom in your feed and watch what companies do with money - hiring premiums for AI-capable builders and enterprise AI budgets are the signal; funding-round theatre is the noise.

Back to all posts The Excel precedent →

Bet on the layer that survives crashes.
Skills, not stocks.

Whether the valuations hold or burn, the ability to build real software with AI appreciates in both branches. Start building it free tonight - one signup, no card, live instructor session included.

Start the free trial See the hiring data