Your grandmother probably thinks AI is just fancy autocomplete. Your investors think itās the next industrial revolution. Both might be right. And thatās exactly the problem.
Welcome to the most expensive game of musical chairs in human history.
In October 2025, OpenAIāthe company that made you question whether your job is safeāsigned roughly $1 trillion worth of deals. Not over decades. Not in theoretical future value. One trillion dollars in commitments that locked together the biggest names in tech like a high-stakes game of Twister.
Nvidia committed up to $100 billion to OpenAIās data centers. AMD followed with tens of billions more. Oracle inked a $300 billion cloud contract. Each company took equity stakes in OpenAI while simultaneously becoming its customer and supplier.
Itās beautiful. Itās terrifying. And if youāre building anything in Southeast Asia, itās about to force your hand.
The Flywheel That Might Break the World
Hereās whatās actually happening beneath the surface of those press releases.
OpenAI needs computing powerānot just a lot, but an almost incomprehensible amount. Weāre talking 20 gigawatts worth of data centers. Thatās the output of 20 nuclear reactors, running continuously, just to train the next generation of AI models.
They canāt pay for this upfront. So theyāve structured deals where chipmakers like Nvidia essentially finance OpenAIās infrastructure in exchange for guaranteed orders. Nvidiaās money buys data centers filled with... Nvidia chips. Which OpenAI uses to train AI models. Which drives demand for more Nvidia chips. Which justifies Nvidiaās stock price. Which gives Nvidia more currency (in the form of valuable equity) to invest in... OpenAI.
See the loop?
Now multiply this across AMD, Oracle, Microsoft, and a web of cloud providers and startups. Everyone is simultaneously the investor, the customer, and the supplier. Capital flows in a perfect circle, each deal reinforcing the next, each rising stock price validating the previous bet.
This is either the most sophisticated value-creation flywheel ever constructed, or itās vendor financing on steroids.
The Cisco Parallel Nobody Wants to Talk About
If youāre over 35, you remember what happened to Cisco Systems.
Late 1990s. Internet boom. Cisco was the arms dealer of the dot-com gold rushāselling routers and networking equipment to every startup that raised venture capital. Their stock went parabolic. They briefly became the most valuable company on Earth.
Then came the vendor financing strategy. Cisco would invest in or loan money to internet companies... so those companies could turn around and buy Cisco equipment. Revenue exploded. Wall Street cheered. Cisco executives became billionaires.
Until the music stopped.
When the dot-com bubble burst in 2000, Cisco discovered that a huge chunk of their ārevenueā was actually just their own money cycling through customer companies. Those customers went bankrupt. Ciscoās stock dropped 90%. The playbook that seemed genius became the textbook example of bubble economics.
Nvidiaās $100 billion stake in OpenAI looks uncomfortably similar.
Is this time different? Maybe. AI is real in a way many dot-com businesses werenāt. ChatGPT has 200 million users. Companies are deploying AI in actual workflows, not just buying vaporware.
But hereās the uncomfortable question: How much of AIās current growth is real demand versus artificially inflated demand created by these circular financing arrangements?
Why This Matters for Southeast Asia (And Why You Have Less Time Than You Think)
While this trillion-dollar poker game plays out in Silicon Valley and Shenzhen, Southeast Asia is being forced to make a choice it didnāt ask for.
Do we join this ecosystem on whatever terms we can get? Or do we try to build our own capabilities knowing weāre years behind?
The honest answer: We need to do both. And we have maybe 24 months before the window closes.
Hereās why the timeline is so tight.
Right now, these mega-deals are still being structured. Standards are still fluid. The technology stack is still evolving. Thereās room for regional players to position themselves as integration layers, deployment partners, or specialized service providers.
But once these circular deals lock ināonce Nvidiaās chips only work seamlessly with Microsoftās cloud which only optimizes for OpenAIās modelsāthe interoperability window slams shut. Youāre either inside the ecosystem or permanently outside it.
And if youāre outside? Good luck competing when your opponent has access to computing power you canāt afford, AI models you canāt replicate, and partnership networks you canāt penetrate.
This is the new digital divide, and itās being drawn right now.
The Robot Revolution Nobodyās Pricing In
If the AI investment loop was just about software and cloud services, we could debate whether itās sustainable. But thereās a second wave coming that changes everything: embodied AI.
Translation: Robots with AI brains, walking around in the physical world.
July 2025. Shanghai. World Artificial Intelligence Conference. Over 150 humanoid robots on display. Chinese companies selling working humanoids for $16,000. Some models as low as $5,900.
Morgan Stanley just published research projecting the humanoid robotics market could hit $5 trillion in annual revenue by 2050. Thatās twice the size of the global automotive industry.
Let that sink in. Weāre not talking about science fiction or distant futures. Weāre talking about a trillion-dollar manufacturing ecosystem that needs to get built in the next 10-15 years.
And Southeast Asia has a real shot at being a major playerābut only if we move now.
Why China Is Winning the Robot Race (And What We Can Learn)
Hereās the uncomfortable geopolitical truth: China is currently best-positioned to dominate āembodied AI.ā
Not because they have the best AI research (though theyāre closing the gap fast). But because theyāve cracked three things that matter more than pure technology:
1. Manufacturing ecosystem at scale. China can produce robots cheaper and faster than anyone else. Their supply chains for motors, sensors, batteries, and materials are unmatched.
2. Guaranteed internal demand. Chinese state-owned enterprises will buy domestic robots as a matter of policy. That gives Chinese robotics companies a market to refine their products before going global.
3. Strategic patience combined with tactical speed. Beijing identified robotics as a national priority years ago. Theyāre playing a 20-year game with 6-month sprints.
Meanwhile, American robotics CEOs went to Congress in 2025 literally begging for a national strategy, warning that without coordinated policy and investment, the U.S. will lose both the robotics race and, by extension, the AI race.
The robots are where AIās economic value gets captured. If you lose robots, you lose AI.
Where does that leave Southeast Asia?
The Strategic Non-Alignment Playbook
Hereās the move: Southeast Asia should become the Switzerland of the AI-robotics cold war.
Not in the sense of being neutral and boring. In the sense of being the place where East meets West, where interoperability gets figured out, where multiple tech ecosystems coexist and connect.
Malaysia is already doing this. They signed AI cooperation agreements with China while simultaneously licensing chip design technology from UK-based Arm and partnering with U.S. firms on industrial automation. Theyāre building relationships on all sides while developing domestic capability so theyāre not completely dependent on anyone.
Singapore is even more sophisticated. They use Chinese robotics for some infrastructure, Western AI for financial services, and invest heavily in their own research. Theyāre building genuine optionality.
This isnāt fence-sitting. Itās strategic positioning.
Because hereās what most people miss: The company or country that can integrate Chinese hardware with Western software with local applications becomes incredibly valuable. Youāre the translator in a world where two superpowers speak different languages.
But this only works if you have actual capability, not just diplomatic skill. You need engineers who understand both ecosystems. You need companies that can deploy and maintain robots regardless of where theyāre manufactured. You need software that works across platforms.
Building that takes time. Hence: 24 months.
What Founders Should Actually Do This Quarter
Enough strategy. Letās get tactical.
If youāre a founder or operator in Southeast Asia right now, here are five moves that matter:
1. Pilot robots now, even if theyāre imperfect.
Donāt wait for mature technology. If youāre in manufacturing, logistics, or warehousing, start testing robot deployment today. The companies that learn how to integrate robots with human workflows now will have compounding advantages by 2030.
The cost of being five years behind in operational knowledge will vastly exceed the cost of adopting imperfect technology today.
2. Build the integration layer, not the hardware.
Unless youāre exceptionally well-funded, donāt try to compete with Chinese firms on robot hardware or Western firms on foundational AI. Instead, build the software and services that make those technologies useful in Southeast Asian contexts.
A robot designed for a Japanese factory doesnāt automatically work in an Indonesian palm oil plantation. Someone needs to adapt it. That someone could be you.
3. Make your pitch anti-fragile.
If youāre fundraising, assume it will take twice as long as you think and that 80% of pitches will fail. Thatās not pessimismāthatās the new baseline.
Series A deal volume is down 18%, dollars deployed down 23%, and median fundraising timeline has stretched to 20+ months. Build your financial model assuming you need 24-30 months of runway, not 18.
4. Get specific about your AI storyāor drop it entirely.
VCs are getting sophisticated about AI-washing. If you claim to be an AI company, youāll get grilled on model architecture, training data, and inference costs. If you canāt defend those claims technically, donāt make them.
Better to be a great logistics company that happens to use AI than a mediocre AI company trying to find a use case.
5. Map your stakeholder ecosystem before scaling.
For every market you want to enter, identify the regulators, incumbent players, and local partners who will determine whether you can actually deploy. Then engage them early.
In Southeast Asia, your ability to navigate complex stakeholder dynamics is often more important than pure technological superiority.
The Bet Youāre Making Whether You Realize It or Not
Every founder right now is making an implicit bet about the futureāeven if youāre trying to avoid making a bet.
If youāre building in AI or robotics, youāre betting that this wave is real, that the investment will eventually find profitable returns, and that thereās room for new players despite the trillion-dollar incumbents.
If youāre staying away from AI entirely, youāre betting that the hype will deflate, that most AI companies will fail, and that there will be opportunities in the aftermath for more traditional businesses.
Both bets carry risk. But only one bet has upside if youāre wrong.
If you bet on AI and it turns out to be overhyped, youāve still built capabilities in cutting-edge technology. You can pivot. Youāve learned. You have optionality.
If you bet against AI and it turns out to be transformative, youāve built capabilities in a world that no longer exists. Youāre starting from zero.
This is why the smartest founders I know arenāt asking āIs this a bubble?ā Theyāre asking: āHow do I build something that has value regardless of whether this is a bubble?ā
The answer? Focus ruthlessly on unit economics, real customer problems, and sustainable business models. Use AI as a tool, not a story. Build partnerships that give you leverage, not vendor relationships that make you disposable.
And move fastābecause in 24 months, the rules of this game will be set in stone.
The Uncomfortable Truth About Timing
Weāre at an inflection point that happens maybe twice in a generation.
The last comparable moment was probably the late 1990s with the internet, or the late 2000s with mobile. Decisions made in the next 2-3 years will shape the next 2-3 decades.
The leverage available to individual founders, operators, and investors right now is enormous. But you have to be willing to grab it.
That means accepting uncertainty as the baseline condition. That means making decisions with incomplete information. That means being wrong sometimes and adjusting fast.
The biggest mistake isnāt picking the wrong technology or the wrong market. The biggest mistake is paralysisāwaiting for perfect information that will never come, watching the window close while youāre still analyzing.
Southeast Asia has a genuine shot at being a major player in the AI-robotics revolution. But only if weāre willing to act while the rules are still being written.
The trillion-dollar flywheel is spinning. The robot factories are being built. The investment decisions are being made.
You can shape this, or you can watch it happen to you. But you canāt do both.
Whatās it going to be?
Kim Yeoh is co-host of Sea of Startups and writes about technology, strategy, and building in Southeast Asia. Kevin Brockland is her co-host and occasional voice of reason. Theyāre both trying to figure this out in real-time, just like you.
Subscribe to Sea of Startups for weekly insights that wonāt make you dumber:
š§ Spotify | Apple Podcasts | YouTube
š¬ Whatās your take? Are we in a bubble, a revolution, or both? Comments are open.
DISCLAIMER: All views expressed are personal opinions and do not represent any organizations mentioned. Content is for informational and entertainment purposes only and should not be considered professional, investment, or legal advice.
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit seaofstartups.substack.com
