Hello Betamax,
This week’s US midterm elections could determine how easily American AI companies can keep expanding in their home turf, and whether some of their spending could head to regions such as Southeast Asia.
Data centers have become a flashpoint over their electricity bills, water use, and impact on local communities. Voters in 10 US states will consider measures that could delay construction, ban facilities, or add requirements for operators.
 Made by Ulla/Tech in Asia with the help of AI The pressure isn’t limited to one political party. An analysis from The Information found that in five gubernatorial races, both Democratic and Republican candidates favor pausing new data center projects. These include Texas and Ohio, two of the biggest data center markets in the US.
Much of the debate comes down to who should pay for the additional power infrastructure these facilities need. Some candidates want developers to foot the bill rather than passing the costs on to households, while others are proposing to remove tax breaks and impose stricter environmental requirements on the facilities.
If building data centers in the US gets too complicated, could Southeast Asia pick up some of that demand?
That depends on whether its markets could get projects moved from MOUs to the ground. As we’ve reported previously, energy constraints and water shortages remain huge bottlenecks in parts of the region, and some data center projects haven’t left the planning stage.
The US elections will test public appetite for the AI infrastructure buildout, and depending on the results, it could give data center operators a stronger case for building abroad. But Southeast Asia should be careful what it wishes for.
Glenn Kaonang, journalist
WHAT ELSE WE’RE THINKING
1️⃣ Apple turns to Alibaba to close its AI gap in China
Alibaba’s Qwen is heading into selected Apple Intelligence features in China. Developer testing starts November 4, with a public release planned for December.
The timing matters for Apple, as it is competing against Chinese rivals such as Huawei, which have moved faster with rolling out AI features. Apple Intelligence has been largely absent from China because of regulatory hurdles.
Apple’s fortunes have improved. Its smartphone shipments in China grew 23% year on year in Q2 2026, giving it an 18% market share and putting it second to Huawei, according to Counterpoint Research.
But recovering sales doesn’t mean Apple can afford to fall behind on AI. Its partnership with Alibaba gives it access to local AI capabilities while helping it navigate China’s regulatory requirements.
Alibaba has something to gain, too. Getting Qwen into Apple Intelligence puts its technology in front of iPhone users without having to persuade them to download a separate app.
For all the talk about AI models competing to be the smartest, this partnership shows that even Apple can’t go at it alone in China.
2️⃣ Sam Altman’s next villain arc: your RAM
 Image credit: Ulla Sam Altman has a complaint about the memory sitting in your computer.
Discussing local AI in a recent podcast, he called it “quite terrible” for scarce DRAM to sit in individual machines that are used only occasionally. He proposes to pool memory resources in data centers, where they can be shared across users and put to work more often.
Conveniently, that’s also where OpenAI runs most of its models.
I only have 16 GB of RAM in the computer I’m typing this on, but as one Reddit user wrote, Altman will have to “pry it from my cold, dead hands.”
In case Altman needs a refresher, RAM has other jobs besides running AI. And even when it comes to AI, maximizing memory utilization isn’t everything. Plus, having enough memory to run AI models and software locally gives us something the cloud doesn’t always offer: privacy, independence, and control.
But who gets to decide what counts as a good use of scarce resources? OpenAI poured substantial computing resources into Sora, only to shut down the video-generation app after a short run.
Maybe Altman can take his complaint up with Nvidia. The company currently sells RTX Spark, a laptop chip with up to 128 GB of unified memory that is explicitly pitched for running AI locally alongside gaming and creative work.
Perhaps the AI industry should examine its own priorities before questioning how the rest of us use our hardware.
AN EXECUTIVE VIEW ON AI
In this edition, Tech in Asia speaks with Heather Paterson, vice president of Asia Pacific and Japan at legal AI startup Legora, which recently raised US$550 million at an US$8.8 billion valuation. Paterson shares how AI is changing document review, why junior lawyers still need apprenticeship, and what it takes to get people beyond using AI for basic drafting.
This interview has been edited for brevity and clarity.
 Heather Paterson, Legora’s vice president of Asia Pacific and Japan / Photo credit: Legora What does AI let lawyers do today that they couldn’t before?
When I joined Legora in June 2025, there were three fairly isolated tasks: research, review, and drafting. Now, we’re connecting those tasks with the knowledge and context an organization already has.
Due diligence in a merger or acquisition is a good example. You receive files from a data room, email, or SharePoint, and they’re often a complete mess.
Previously, on a sizable transaction, you would narrow the scope of your review. You might say, “I’m going to take your top 20% of supplier contracts, and everything else we’ll assume a risk against.”
With AI, you can bring all the files into that review, while still having human verification as part of the red flag report. You can ask the product to organize the files, give them a naming convention, and then run due diligence.
That gives clients more coverage of their risk. It changes what you can examine as well as how quickly you can do it.
How do you build trust in those answers when AI can make things up?
Trust comes through verification. Sometimes, when you’re trying to extract something like a change-of-control clause or termination provisions, our product will tell you, “I’m not sure of the answer.”
It will also show you why: perhaps the information doesn’t exist, or it has found something that needs you to verify it in more detail.
That’s how you build trust. Sometimes the useful response is to tell you where the uncertainty is and ask you to look more closely.
Among your Asia-Pacific customers, are you seeing AI reduce demand for junior lawyers?
I don’t see that trend yet. What I’m seeing is demand for people who have skills in both AI and law.
Around two-thirds of my employees in the region are lawyers or have been lawyers. It’s not a cheap strategy, but our legal engineers help clients move beyond basic drafting and review. We look for different specialisms, from litigation to M&A and real estate, so they can help clients explore how to use AI in those areas.
We’re also seeing clients wanting to develop legal engineering skills internally, bringing together people who understand law and technology. In the future, I don’t think it will be called legal engineering. It will just be part of a lawyer’s job.
But junior lawyers traditionally learn through research and first drafts. If AI takes on those tasks, how do they develop judgment?
I wouldn’t recommend that people just do everything using AI. You have to help them understand how to do these tasks without it as well.
I used to work at Intralinks, in the data room part of M&A. I remember talking to lawyers who had to find a change-of-control clause in 600 leases or agreements. You don’t go to law school to sift through documents looking for a clause. That should be done by AI.
But universities and graduate programs do need to adjust. How do you help people apply judgment when they have this powerful tool? Part of that is understanding how an answer was reached: “Here’s my answer, and here’s how I got there.”
Reviewing AI-generated work alone is not sufficient. You need to work with a partner or senior associate when putting answers together and making recommendations. You learn judgment and the things to watch out for from working with people who have that experience.
Can you give an example of a Singapore firm using AI to consider work it previously wouldn’t take on?
We worked with a law firm that was already using other legal AI tools. One capability it explored during our proof of concept was regulatory monitoring.
Previously, the firm wouldn’t take on that business because it was so time-consuming, and the advice was only current at one point in time. You could give a client a picture of the regulations affecting its business across different markets, but something might change the next day.
During the proof of concept, the firm started thinking about delivering more of that work and offering something different in Singapore.
What it needed to see was how our product differed from the tools it already used, and that we had a local team with legal backgrounds to support it. That led to a purchase, and we’re now working on embedding and rolling out the product.
What stops lawyers from getting more out of AI once they have access to it?
People are creatures of habit. You could spend however much on legal AI, but if you’re only doing review, translation, or simple drafting, you’re not accessing its full capacity.
You also have to keep asking what the technology couldn’t do three months ago that it can do now. The capability keeps changing. That’s part of what our legal engineers help clients understand.
A lawyer might remember drafting a provision before but have no idea which folder or client matter it’s in. Connecting AI to the firm’s document management system lets it find that context and bring it into the next piece of work.
To me, the top skill for a lawyer in the world of AI is being curious and having an experimental mindset.
If AI makes the work faster, what happens to the billable hour?
I really wish they would get rid of it. You go to a lawyer for a judgment call. That call might take less time with AI, but it draws on decades of a partner’s experience and the firm’s accumulated knowledge.
I think we could see a few scenarios. Hourly rates might go up to reflect that value, or firms might pass through the cost of using AI and offer clients different options for how the work gets done.
That’s partly an expectation. I think it’s happening in patches, including outside law, where professional services firms are adding a technology cost to invoices.
Catch up on past issues of the newsletter here.
|
