Trouble is brewing in AI paradise

Pranav, Tech in AsiaOct 28, 2026

The Prompt

Welcome to The Prompt, your guide to the biggest issues and developments shaping AI, with Asia front and center. Want full access to all our AI reporting? Subscribe here.

Hello [subscriber],

What a month! October 2026 may be remembered as the end of the four-year AI bull run that began with ChatGPT’s launch.

The bubble hasn’t burst, but it is looking stretched. Since mid-month, AI infrastructure stocks like CoreWeave, IREN, Applied Digital, and Oracle have taken a beating.

Image credit: Made by Ulla/Tech in Asia with the help of AI

On Monday, Elon Musk predicted “guppy-level death and consolidation” among more than 200 neoclouds, saying debt markets were closed to companies without substantial revenue independent of AI data centers.

“Some will make it through, but not many,” he wrote on X, adding to the pressure on stocks.

Musk may be exaggerating, but his warning captures investors’ concern: Can demand for AI compute translate into reliable revenue, and can customers keep paying?

For infrastructure providers, the customers they sell to could matter as much as how much they sell. Long-term contracts with financially strong customers could determine who gets funded and who gets left behind.

Asia’s neoclouds have less of a public-market presence than their western peers, but they will be watching. Billions are going into data centers. Who are they being built for?

Smaller AI labs and startups may face larger upfront payments, tying up scarce compute capital. Providers will increasingly need creditworthy “anchor customers” to finance expansion. To sustain the buildout, banks and telecom companies still experimenting with AI may need to become committed buyers.

Meanwhile, Ben Goertzel, who helped popularize the term artificial general intelligence (AGI), told us in an exclusive interview that he worries more about humans using pre-AGI systems who could trigger World War III than about AGI itself.

And into this storm comes Anthropic, apparently convinced it is a great time to go public.

On Monday, it published its IPO papers with an explicit warning that AI poses “catastrophic or existential risks to humanity.” In this issue, we explain what the listing could mean for Asia.

On a lighter note, our editor Melissa takes a look at Muse as it launches in Singapore today. A little cuteness might be just what we need amid all this AI doom and gloom.

Pranav Balakrishnan, AI reporter

TOP AI READ FROM OUR DESK

AGI pioneer Ben Goertzel thinks the AI race is too narrow

Digital collage by Ulla, photo courtesy of SingularityNET

What is AGI? When does AI become AGI? Are we already there?

There’s no clear-cut answer. Even Goertzel, who co-edited an entire book about AGI, struggles to pin it down when asked.

He may sit outside today’s mainstream AI conversation, but he’s still pursuing that elusive goal through SingularityNET, which develops open, decentralized AI tech. He also runs BGI Labs, which builds proprietary products on SingularityNET’s infrastructure.

WHAT ELSE WE’RE THINKING

1️⃣ Why Anthropic depends heavily on frenemies

The AI giant’s S-1 filings with the US Securities and Exchange Commission reveal growth unlike anything public markets have seen: Revenue jumped 1,300% to US$16.3 billion in the first six months of 2026.

Its net loss of US$80.7 billion largely reflected non-cash adjustments to financial instruments’ value, while operating cash flow was positive at US$3.9 billion. The IPO’s size remains undisclosed. International markets contributed just US$4.5 billion in revenue.

Its data centers are primarily in the US, and it is “pursuing additional data center capacity internationally.” Further office expansion is unlikely: Existing locations, including Singapore and India, meet immediate sales needs, says the company.

The filings also expose its reliance on rivals such as Google, Amazon, and Microsoft – all of which are its investors. Hyperscalers generated 48% of first-half revenue in the first half of 2026, up from just 11% in December 2024.

Anthropic calls these cloud providers “investors, customers, distribution partners, and competitors,” warning they can track its commercial contracts and prioritize their own models.

Arch-rival Meta is also reportedly among Anthropic’s biggest customers, projecting that it could spend up to US$10 billion per year. To reduce this dependence, especially when it comes to compute, Anthropic says it will use IPO proceeds in capital expenditure to build data centers.

Anthropic is also trying to keep Big Tech investors at arm’s length through its share structure: Class A for IPO buyers, a separate class with higher voting rights for founders, and non-voting Class N shares for Big Tech investors.

Keep your frenemies close, and their voting rights at zero, I guess.

2️⃣ Muse is fun. That doesn’t mean I trust it

Disclaimer: I was given free access to the Muse app ahead of its launch in Singapore today.

Meta’s AI assistant Muse surpassed 10 million users last week, just over a month after being launched.

As a casual AI user, I can say that Muse’s mobile interface is surprisingly easy to get used to. By day two, I’d already sent up a daily “morning briefing” of my growing list of tasks – it had grown to seven in just under a day, from alerting me to price drops to suggesting a Pilates studio near me. (I’d spent a good six months finding one, which says a lot about how decisive I am.)

Image credit: Made by Ulla/Tech in Asia with the help of AI

I hate that “cutewashing” is working on me. But it is.

The woolly, blinking, smiling robot, with slightly flushed cheeks – which Meta internally named Jollybot – looks like it can do no harm. I keep my defenses up by calling it “Muse” instead of giving it a pet name.

Muse is clearly smart and able to correctly understand local street names like Telok Ayer. Also, it accurately pulls up names of major electronics retailers to monitor air purifier availability in haze-cloaked Singapore. But it is still not purchasable anywhere, as of the 5:25 a.m. restock check, Muse updates.

Clearly, it’s a handy assistant in your pocket. Coming up with a gift idea? Searching for price drops? Shortlisting a new gym in the vicinity? Easy.

And when support for local services like Grab – along with other local ones – gets progressively rolled out, I can see myself using the assistant to book a ride for me, too. (Muse wasn’t able to access the Grab app at the time of testing.)

The bigger question is this: Can I trust it to handle higher-stakes tasks like sending messages on my behalf?

I stopped short of giving it access to WhatsApp. In fact, I wouldn’t have given it access to my social media accounts if it wasn’t required.

Despite what Mark Zuckerberg says about an encryption key that prevents the company from accessing certain information, a combination of other users’ experiences, large language models’ safety track record, and my general distrust of technology can’t help but give me pause.

I won’t lie, the morning recaps keep me on task, but I wonder if I really need more notifications lighting up my phone screen throughout the day.

Melissa Goh, senior editor

AN EXPERT VIEW ON AI

In this edition, Tech in Asia’s senior editor Melissa Goh speaks to Howard Wright, vice president of Nvidia Inception, the firm’s global startup program.

Nvidia needs no introduction, but perhaps less known is the role it plays in bridging startups, founders, enterprises, and investors. Across Southeast Asia, the firm works with almost 2,000 startups – out of 50,000 worldwide. It also works closely with VCs to source deals through a global VC alliance, though the company couldn’t share which.

Among the Singapore-based startups accepted into the program are robotics firm Sharpa, biotech company Hummingbird Bioscience, and deeptech startup SynaXG. Sharpa’s five-fingered robotic hands, known as Sharpa Wave, are integrated into H2 Plus, an open humanoid robot reference design developed by Nvidia and Unitree.

The program is free to join and doesn’t require founders to give up equity. Is there a catch?

Howard Wright, vice president of Nvidia Inception / Photo credit: Nvidia

This interview has been edited for brevity and clarity.

When it comes to the challenges that AI startups in Southeast Asia face, what would you say is the biggest hurdle? Access to compute, talent, or funding?

It’s not talent. There’s a deep bench of talent from researchers, academics, and startups. I think the entire world is ready for Vera Rubin and ready for more compute, and we’re trying to find ways to do that on a big scale and a small scale.

There are the monster Terafabs that Elon and others are building. Just last week in Germany, Jensen [Huang] announced the DSX pod – you can get your latest and greatest hardware and software in a pod, and, like a shipping container, you could put that right outside of your enterprise behind a sports arena, or right down by the Formula 1 track.

So, from the full stack, worth hundreds of thousands of square meters of data centers, AI factories, the DSX pod, the DGX stations, all the way down to DGX Spark – what we’re trying to do is make sure that all of this is modular and folks can find sips of compute much easier than just waiting for an AI factory to be built, which takes multiple years.

Part of the impetus right now is making sure that if it’s sips of compute or if it’s as if you’re bathing in compute, we make sure that we find compute for the world.

Let’s talk about Inception. It’s free to join, and it doesn’t require startups to give up equity. What is the business case for it, and how do you measure success?

The goal of Inception, unapologetically, is to captivate the hearts and minds of the most important startups and the most important VCs and get them up and running on Nvidia technology.

It’s free because these startup founders need a co-traveler. We provide engineering support; we can help with their minimum viable product, their go-to-market strategy, funding, and board of directors.

We provide the resources for folks to go from Inception to unicorn or decacorn. In the 50,000+ startups that we have in the program, we are absolutely convinced that multiple trillion-dollar market cap companies reside, and it is our financial responsibility to help them find ways to succeed.

The business model is simple. We find them, we curate them. We embrace them, we nurture them. Eventually, they become the “AI natives” and the unicorns that are the single-verb unicorns that exist today. And they buy plenty of compute from Nvidia at that point. So think of us as that early-stage access to this startup ecosystem. Our horizon is much further out than this month or this week or this quarter.

We have mini metrics and mini KPIs. Revenue just doesn’t happen to be one of them. We do have specific quantifiable measurements, but none of them involve revenue. Jensen teaches us not to look at the score and play the game the right way.

If the entire ecosystem – including the original equipment manufacturers, original design manufacturers, neo-cloud providers, VCs, and accelerators – has additional access to compute, the entire ecosystem wins.

You mentioned dropping breadcrumbs for the companies that join Inception. How hands-on does Nvidia get?

We can be as hands-on as the startup would like us to be. So our solutions architects can sit shoulder to shoulder and put fingers on keyboards next to other CTOs and startups.

Sometimes it is a soft introduction from somebody senior in our leadership team to introduce them to an enterprise in the area. Other times it’s joining Teams or Zoom calls with potential partners. Sometimes it’s going to meetings on Sand Hill Road in the US with our partners. Sometimes it’s setting it up; sometimes it’s going with them.

We like to think that we meet the startup where they are, and we’ve done all combinations thereof. The goal is to be very proximate to their solution set.

So it’s simply a matter of a startup raising their hands and saying they need help with a particular partnership or customer?

Yes.

A startup that joins the program and starts building the business based on Nvidia’s AI infrastructure, hardware, or software naturally becomes quite dependent on Nvidia. Does Nvidia encourage the startups to also build or use other types of hardware and AI platforms?

Nobody is dependent on us. To go get compute anywhere in the world right now, folks can do it on our silicon, our software, or someone else’s. There’s no requirement that if you join Inception, you have to be dependent in any way on our hardware or software.

What does Nvidia look for in the AI startups that you choose to partner or collaborate with?

There’s no checklist or quadratic equation, but when we peel it back, there are a number of criteria that kind of fit, which are: Who’s on the C-suite, and are they repeat founders? Who’s on the cap table, and from whom have they raised, and how much? Does the product have defensible IP, or is it the 214th company trying to do something similar? In which vertical are they predominant?

There are four or five things that, as you continue to look at them, it goes, “Oh, these reveal themselves as unicorns,” over and over and over again. So we know what excellence looks like, but it is not in a nice, neat, quadratic equation.

Catch up on past issues of the newsletter here.

P.S. Don't miss out on the biggest tech news and analysis. Add [email protected] to your address book, contacts, or safe sender list. Or simply move us into your inbox.

Tech in Asia thanks our partners for supporting our newsletters. Sponsorship has no influence on editorial content.

You can also remove yourself from all Tech in Asia emails over here.