The Day the AI Boom Showed Its Cracks
There's a number that should make every tech investor, policy maker, and CEO lose a little sleep: $750 billion. That's what Anthropic is reportedly contemplating spending on computing power. Not over a decade. Not in cumulative commitments across multiple partners. A single company, at a single moment in time, staring down a compute bill that would rival the GDP of Turkey.
And yet on the same day, Qualcomm announced worldwide job cuts after annual reviews. Oyo's parent Prism slashed its IPO fresh issue to $353 million — a retreat from its original ambition. DayOne, a data center operator, is considering delaying and shrinking its planned US IPO. The signals are screamingly contradictory, and most analysts are treating them as separate stories. They're not. They're the same story.
The global technology economy is experiencing a violent bifurcation: capital is concentrating into a handful of AI infrastructure bets while simultaneously retreating from broader expansion, cross-border growth, and workforce investment. The result is a system that looks productive in aggregate metrics but is hollowing out at the edges.
The Spending Paradox Nobody Wants to Name
Let's do the arithmetic that the AI cheerleading class refuses to confront. Anthropic spent nearly $9 billion on computing capacity in a recent period. Now it's contemplating $750 billion in total compute commitments. Nvidia is backing Pigment, an AI planning startup, with $250 million. Spark Capital is leading a $325 million round for ASI, an AI aviation company that beat Palantir for an FAA contract. Blackwall's valuation quadrupled to $2.2 billion on a $100 million raise. General Catalyst is pouring $320 million into Otter.ai.
This is not healthy capital allocation. This is a land grab dressed up as investment.
Meanwhile, Qualcomm — a company that should be the single biggest beneficiary of the AI-on-device revolution — is cutting jobs. Not restructuring. Not optimizing. Cutting. Because even in an AI boom, you need someone to tell you when you've hired too many people for a product cycle that's moving faster than your headcount can track.
The irony is almost too clean: the AI revolution is simultaneously the reason companies are spending trillions AND the reason they're firing thousands. DeepSeek open-sourcing a tool to cut AI costs on Nvidia Blackwell and Huawei Ascend chips is the market telling you that these compute prices are absurd. When a Chinese lab is essentially publishing a "discount coupon" for your infrastructure, your business model has a pricing problem.
The DayOne Signal
DayOne's potential IPO delay deserves more attention than it's getting. A data center company with 22 listed facilities — not all confirmed operational — is hesitating to go public. Why? Because the market is asking harder questions about whether the AI infrastructure buildout has a revenue model that justifies the capex. When the companies building the picks and shovels are scared to ring the bell, the gold rush may be entering its terminal phase.
The Great Repatriation Is Real, and Southeast Asia Feels It First
The HSBC Global Entrepreneurial Wealth Report 2026 dropped a bomb that most outlets buried: the share of entrepreneurs globally who want to expand internationally fell six percentage points. Six points. In one year. That's not a blip. That's a structural shift.
And Southeast Asia is feeling it acutely. The region's beloved $1 trillion digital economy target — a number that has been the North Star for a generation of founders and VCs — has now slipped five years. The 2026 GMV projection of $368 billion, up 17% year-on-year, is respectable in isolation. But the slippage tells you everything: the assumptions embedded in the original forecast — cross-border integration, regulatory harmonization, frictionless capital flows — are being systematically dismantled.
Taiwan blocking Grab's $600 million Foodpanda acquisition is Exhibit A. This would have been Grab's first expansion outside Southeast Asia, and it was blocked not on antitrust grounds but on what amounts to economic nationalism dressed in regulatory clothing. The message is clear: the era of "build regionally, expand globally" is over. The era of "build regionally, get blocked everywhere else" is here.
The IPO Retreat as Repatriation
Look at Jio Platforms opening bidding for its $3.2 billion IPO. It's an offer for sale — existing investors sell, Jio raises nothing new. This is not a company going public to fund growth. This is investors cashing out and the company accepting that its growth story no longer commands fresh capital at a premium. Oyo's parent Prism cutting its fresh issue to $353 million tells the same story. These aren't IPOs. They're liquidation events with a ceremonial bell.
The pattern is unmistakable: capital that once powered cross-border tech expansion is now retreating to domestic markets, to AI infrastructure bets, to safe-yield instruments. OCBC offering ringgit-denominated savings to Singapore customers — requiring a 5,000 MYR average daily balance just to earn interest — is a small data point, but it signals the kind of hyper-local financial pragmatism that replaces the old "open borders, open capital" ideology.
The Application Layer Is Getting Crowded and Thin
Zoom is adding AI controls and compliance tools. Salesforce is buying Braze for $5.6 billion. Visa is partnering with Bill and Billtrust on AI agent payments. Toku is launching a visual tool for building speech AI. Verba has 30,000 paying customers building AI video workflows. Genie Health raised $3 million to organize your scattered health data.
Every single one of these is a legitimate business. Every single one is also a commodity waiting to be commoditized further.
The Salesforce-Braze deal is instructive. A $5.6 billion acquisition of a marketing automation platform in a world where every company is building AI-native marketing tools from scratch. Braze shares jumped 51% on the news — which tells you the market thinks this is a great deal for Braze shareholders and a questionable one for Salesforce. You're paying $5.6 billion for a customer base that AI is about to make redundant.
The 81% of Singaporeans who plan to use AI for gifts, the Jobstreet data showing agentic AI as Singapore's hottest job skill, the $300 million going into AI planning tools — all of this confirms that AI adoption is broad and real. But breadth without depth is just a fashion cycle. The question isn't whether AI is being used. It's whether enough of that usage is creating durable, defensible value or just moving the mess faster, as one founder aptly put it.
The Agentic AI Employment Cliff
Singapore's Jobstreet data showing agentic AI as the hottest job skill should terrify anyone who thinks this transition will be smooth. The market is signaling that the jobs being created require the same capabilities that will replace them. You can't have a labor market where "agentic AI" is both the scarce skill and the automation tool. One of those is going to lose, and it won't be the technology.
The Bottom Line
October 22, 2026 is the day the AI boom's contradictions became impossible to ignore. Anthropic wants $750 billion in compute. Qualcomm is firing people. Southeast Asia's trillion-dollar dream is slipping further away. Taiwan is blocking acquisitions. Jio is cashing out instead of raising. The globalization thesis that powered twenty years of tech is in its death rattle, and the AI thesis that's supposed to replace it is consuming capital at a rate that will leave nothing for anything else.
The most important question in global business right now isn't "will AI work?" It clearly is. It's "can an economy sustain $750 billion compute commitments AND fund the cross-border expansion, workforce development, and application-layer innovation that actually create broad-based prosperity?" The answer, based on today's evidence, is no. Something has to give. And it won't be the AI spending. It'll be everything else.
The AI revolution is real. But revolutions that cannibalize their own base — the workers, the cross-border entrepreneurs, the application developers building on top — don't end in utopia. They end in consolidation. And consolidation, for most people, is just another word for fewer options.