A bubble popped already. Nobody in the Philippines felt it, because it happened in Seoul: South Korea’s KOSPI index lost 44 percent of its value in 40 days this summer, erasing $2.18 trillion, after investors fled Samsung and SK Hynix on fears that Big Tech’s AI datacenter spending would not pay off. The question is not whether there is an AI bubble. It is how many bubbles there are, which one pops next, and what happens to your job, your remittances, and your portfolio when it does.
Everyone from Ray Dalio to the European Central Bank has joined the debate, OpenAI’s own CEO has said the quiet part out loud, and the numbers on both sides are genuinely wild. This deep dive walks through what a bubble actually is, the bull and bear cases with real figures, the three trigger events that could pop it, and the honest answer on timing that nobody giving a conference keynote will tell you.
What a bubble is, and why this one is weird
A financial bubble forms when asset prices detach from the cash the assets can actually produce: tulips in 1637, dot-coms in 2000, subprime in 2008. The classic sequence is displacement (a new technology), credit expansion, euphoria, then the turn. By that template, AI checks every box: a genuinely revolutionary technology, hundreds of billions in debt-funded spending, and US market valuations the most stretched since the dot-com crash, with the Shiller price-to-earnings ratio above 40 for the first time in 25 years.
What makes this cycle strange is that the biggest bulls and bears agree on the facts. Sam Altman himself said in August 2025 that he believes an AI bubble exists. Jamie Dimon says AI is real but some of the money invested will be destroyed. The Australian Financial Review called it “the most anticipated example in history.” When everyone can see the bubble, the timing question gets harder, not easier: capital keeps flowing to whoever is actually building, and shorting a boom with real earnings has bankrupted more hedge funds than bad ideas ever have.
The bear case in numbers
The core problem is the gap between commitments and revenue. OpenAI has committed roughly $1.4 trillion in datacenter spending over eight years against about $13 billion in annual revenue, and expects operating losses of $74 billion in 2028 alone. Deutsche Bank analyst Jim Reid estimates OpenAI’s losses at $140 billion between 2024 and 2029. One estimate projects the company could run out of money by mid-2027 without new capital.
Then there is the circularity problem. Nvidia pledged up to $100 billion to OpenAI, money OpenAI spends partly on Nvidia chips; OpenAI buys from AMD and takes equity; Microsoft holds a giant OpenAI stake while OpenAI commits $300 billion to Oracle. Apollo’s chief economist Torsten Slok says AI profits are “being funded by investors rather than earned from customers.” Morgan Stanley pegs global datacenter spending at $3 trillion through 2028, about half funded by private credit, much of it BBB or junk-rated. The Bank of England and IMF have both warned investors are not properly priced for a crash.
The bull case in numbers
The bears’ facts are real, but so are the bulls’. Goldman Sachs’ chief equity strategist Peter Oppenheimer argues the rally is backed by actual profit growth, with forward P/E ratios well below dot-com peaks. Morgan Stanley calls bubble fears “misplaced,” noting the median big US firm holds about three times the cash reserves of past bubble eras. Unlike pets.com, today’s leaders generate real revenue and positive margins: Nvidia, Microsoft, and the cloud giants are among the most profitable companies ever created.
There is real demand underneath the speculation, too. The five largest cloud companies are pouring nearly $600 billion into capital spending because enterprise AI adoption keeps compounding, and Fed Chair Jerome Powell has noted AI datacenter construction is contributing measurably to GDP growth. The strongest version of the bull case: even if AI never produces superintelligence, the capex is building compute infrastructure the whole economy will rent for decades. Amazon’s warehouses did not pay off in 2001 either; they paid off for the company that survived.
The first crack already happened
Which brings us back to Seoul. In late June and July 2026, the KOSPI crashed 44 percent in 40 days, erasing $2.18 trillion, because more than half the index was Samsung Electronics and SK Hynix, and Big Tech’s short-term returns on AI infrastructure disappointed, sparking fear for the memory-chip demand that carried Korea’s market. Korean retail investors, who had leveraged into chip ETFs, were wiped out; the Financial Times interviewed investors whose life savings evaporated. The index has since rebounded about 30 percent, but the lesson stands.
A concentrated market can pop on a demand wobble even when the underlying technology keeps advancing. That is exactly the structure the US market has built: 30 percent of the S&P 500 sits in just five companies, the highest concentration in half a century. Michael Burry of Big Short fame is now publicly arguing that competition will compress Nvidia’s margins, and the ECB warned this week that a US tech correction of 20 to 30 percent is plausible, with European banks holding roughly €440 billion of exposure to US tech.
The three triggers that could pop it
| Trigger | How it works | Early warning signs |
|---|---|---|
| The yield spike | AI capex is debt-funded; higher rates raise financing costs and crush the present value of distant profits | 10-year Treasury above 5 percent; the 10-year yield has been climbing and sits near 4.74 percent as of this week |
| The revenue miss | Enterprise AI renewal rates disappoint; hyperscalers cut capex guidance; the demand story cracks | Any Big Tech earnings call cutting 2027 capex; cloud revenue growth decelerating two quarters straight |
| The credit event | A major datacenter borrower defaults; private credit marks cascade; financing window slams shut | Junk-rated datacenter bond spreads blowing out; a failed refinance at a neocloud or SPV |
Of the three, watch the credit channel closest. CNBC’s analysis of the dot-com comparison notes that rising yields are the mechanism that popped past bubbles, and this cycle is more debt-financed than dot-com ever was: telecoms at least raised equity. When $3 trillion of datacenter spending leans on private credit, the marginal buyer of risk can vanish overnight, and valuations built on cheap money reprice violently.
So when does it pop?
The honest answer: parts of it pop on a rolling basis, and the timing of the big one is unknowable, but the window most cited is 2027 to 2028. That is when OpenAI’s own projections show losses peaking at $74 billion a year, when the current $1.4 trillion of commitments hits its steepest spending curve, and when the IPO wave meant to refinance it all, OpenAI at a reported $1 trillion target and Anthropic racing toward a $2 trillion filing, must land into whatever market conditions exist that year. If those IPOs slip or price weak, the private-market valuations propping everything up reprice fast.
Three scenarios, honestly labeled as scenarios. The deflation (most likely, roughly where WSJ and ECB land): a 20 to 40 percent drawdown in AI names over 12 to 18 months without a macro crisis, starting from a specific earnings disappointment, with the KOSPI crash as the dress rehearsal. The pop (possible): a credit event plus rising yields turns a correction into a 2000-style bear market lasting years. The melt-up (possible): revenues catch up to capex by 2028, the bubble deflates through earnings rather than prices, and 2026 looks like 1996 instead of 1999. Any analyst who claims certainty on which one we get is selling something.
What it means for the Philippines
This is not a distant Wall Street debate. The country’s flagship services industry is directly exposed: IT-BPM revenue hit $40 billion in 2025 with 1.9 million workers, and the industry’s own association has already cut its 2028 revenue forecast from $59 billion to $50.5 billion, explicitly citing AI as a risk. Generative AI is precisely the technology that automates the voice, transcription, and content-moderation work employing millions of Filipinos.
The second channel is remittances: roughly four million overseas Filipino workers, including tens of thousands in tech roles in Singapore, the US, and Europe, send home the dollars that fund consumption. A global tech downturn hits those jobs and those transfers. The Philippines has also been urged to diversify its economic base precisely because AI threatens the BPO engine. And for the growing number of Pinoy retail investors in US tech funds, the KOSPI’s retail wipeout is the cautionary tale: leverage plus concentration plus a demand wobble equals disaster, even when the technology itself is real.
There is an upside case too, and it deserves saying. If AI productivity gains are real, the same technology threatening call centers also creates premium work: the IBPAP’s own pivot is “from capacity to capability,” higher value per worker. The Philippines produced Thinking Machines, a Filipina-founded AI firm backed by Temasek-scale money, which we covered when Temus backed the enterprise AI play. FEU Tech is going AI-native with OpenAI, which we reported when the university partnership launched. The countries that win the AI transition will be the ones that train for it early.
The bottom line
Is there an AI bubble? Yes, in the specific sense that valuations, debt, and circular deals have outrun current revenue, and one corner of it has already burst in Korea. Will it pop all at once? History says bubbles deflate on their own schedule, usually triggered by rates or credit, and the most cited window for the main event is 2027 to 2028. The rational move for a Filipino reader is neither panic nor euphoria: no leverage, diversification, and a clear eye on the three triggers, yields, revenue renewals, and credit spreads, that will tell us the answer before the headlines do.
One comparison keeps this grounded: we have already run a live experiment on what happens when a speculative structure meets a demand wobble, and it is the same force behind the MicroStrategy collapse we analyzed when MSTR fell 76 percent and started selling its bitcoin. Bubbles do not announce themselves. They reprice, then the story arrives to explain the repricing.
