Posted on — Leave a comment

After OpenAI and Anthropic, Mistral goes further with a full APAC HQ in Singapore

Every big AI lab now has a Singapore story. OpenAI chose the city-state for its first Applied AI Lab outside the US. Last month, Anthropic opened an office here and signed an AI safety testing pact with the government.

Now Mistral AI, Europe’s best-known model maker, has gone a step further and made Singapore its Asia-Pacific headquarters.

The Paris-based company said on Friday it would more than double its regional headcount to 200 by 2027. Most of the new hires will work in science, engineering, AI applications and robotics. Their job is to develop, train and deploy what Mistral calls sovereign AI models.

Also Read: Sovereign AI starts long before the AI model

That is a big commitment. Mistral’s website lists more than 900 employees worldwide. If the Singapore-led team reaches 200, Asia Pacific could account for as much as a fifth of the company’s staff. That is a hefty share for a firm whose home market, and much of its political goodwill, sits in Europe.

Selling sovereignty with a French accent

Sovereign AI means a government agency or business hosts the key building blocks of an AI system on infrastructure it controls. Those building blocks include the chips, the data and the models. The system then answers to local laws and compliance rules rather than to a foreign cloud provider’s terms of service.

There is an obvious irony in a French company selling sovereignty to Asian governments. Mistral’s answer is open weights. It releases most of its models with their weights available. A government can download those models, fine-tune them and run them on its own servers without sending data to Paris or anywhere else. That is a very different offer from US rivals, whose most capable models are mostly rented by the query.

Mistral co-founder and CEO Arthur Mensch said the company wants Singapore to be a strategic hub. Its role, he said, is to help governments and businesses across Asia Pacific “to retain ownership of their infrastructure and intelligence, and translate these into lasting economic value and industrial competitiveness.”

Critics have argued that Southeast Asia cannot build sovereign AI on borrowed choices. Mistral’s model does not fully resolve that tension. A French model in place of an American one, however open, still leaves the region relying on someone else’s research roadmap, training data and chips. In practice, sovereignty comes in degrees.

Singapore’s security agencies got in early

Mistral is not arriving cold. It is already working with DSO National Laboratories, Singapore’s defence research agency, and with the Home Team Science and Technology Agency (HTX). Mistral describes these as some of its earliest model customisation projects anywhere in the world. Earlier this year it also signed a strategic partnership with HTX and agreements with several local technology firms. These include Singtel’s sovereign AI cloud business, RE:AI.

“With Mistral opening its new APAC HQ in Singapore, we look forward to even more seamless collaboration to help the Home Team keep Singapore safe and secure,” said Ang Chee Wee, HTX’s Chief AI Officer and Assistant Chief Executive (Digital & Enterprise).

Also Read: Why Singapore firms fear data sovereignty failures but remain underprepared

The Singapore office will also be an extension of the Mistral Science Hub. It will build sovereign models for three sectors: the public sector, high-tech manufacturing and robotics. Mistral said governments across the region could use its open-weight models to build, customise and deploy AI systems inside their own environments, within their own regulatory rules.

Robots, factories and a fight for scarce talent

The focus on robotics and manufacturing fits neatly with Singapore’s industrial policy. The city-state’s factories are riding the AI wave, though that optimism rests on a narrow base. Its robotics lead has also been read as a warning sign as much as a win.

Pee Beng Kong, Executive Vice President at the Singapore Economic Development Board, said the move would strengthen Singapore’s position as a hub for frontier AI innovation. That is the sort of line that now accompanies almost every AI ribbon-cutting on the island.

The harder part will be hiring more than 100 specialists. Agentic AI is already Singapore’s hottest job skill. Mistral will be fishing in the same small talent pool as OpenAI, Anthropic and Google. It will also be competing with Databricks, which has pledged more than US$350 million to Singapore over three years.

Wedged between American scale and Chinese open models

Mistral faces competition from three directions. US labs bring scale and the most capable closed models. Chinese open-weight models such as DeepSeek and Alibaba’s Qwen are cheap and widely used across the region. Southeast Asia is also building its own regionally tuned models, from AI Singapore’s SEA-LION to Malaysia’s sovereign AI push. Mistral sits in the middle: open, neither American nor Chinese, and with a European pedigree governments find easy to explain to voters.

That neutrality may be its sharpest selling point. As US-China tensions push governments to spread their bets, a vendor that belongs to neither camp can look like the safest choice for sensitive public sector work.

Also Read: Red Hat opens Singapore AI centre to pull enterprises out of pilot purgatory

Mistral was founded in 2023 by Mensch, who previously worked at DeepMind, and by Guillaume Lample and Timothée Lacroix, who both came from Facebook’s AI research division. In under four years it has become Europe’s AI flag-bearer. Now it has to prove it can be Asia’s too.

The real test is whether Singapore becomes a delivery hub that wins contracts in Jakarta, Bangkok and Hanoi, or stays a showroom. Governments that care about sovereignty also tend to care where their vendors’ engineers sit. Two hundred people in one city-state is still a long way from a real presence in ten markets.

The post After OpenAI and Anthropic, Mistral goes further with a full APAC HQ in Singapore appeared first on e27.

Posted on — Leave a comment

Southeast Asia’s robot vacuum boom puts local execution to the test

Vietnam’s robot vacuum market offers a striking signal of demand for household automation. Consumers spent US$92.3 million on the devices in the 12 months through June 2026, an increase of 137.6 per cent by value, according to GfK data reported by VnExpress. ECOVACS, Xiaomi, Roborock and Dreame occupied the top four positions.

That growth presents Southeast Asian hardware founders and distributors with a practical test: choosing devices suited to local homes, reaching buyers at a sustainable cost and supporting them after the sale. Those decisions will determine how much of the market’s expansion becomes a durable business.

Established manufacturers already set the pace

The competitive pressure is regional. In September 2025, Yicai reported, citing GfK, that leading Chinese brands held more than 80 per cent of Southeast Asia’s robot vacuum market. Retail sales revenue had grown nearly 40 per cent in the first seven months of that year.

Those figures describe a market in which established manufacturers already have considerable reach. A new entrant needs a clear reason for customers to choose it and a realistic account of what that choice will cost to deliver.

Vietnam also illustrates a shift within demand. VnExpress reported that the fastest growth came from models priced above US$577. Higher selling prices can create room for distributors and retailers, but they also raise the stakes when an expensive appliance needs attention.

For a regional hardware business, the useful calculation is the margin left after discounts, marketplace charges, delivery, returns and warranty provision. A growing category can still produce an unattractive business if those costs are underestimated.

More categories create different markets

Manufacturers are expanding beyond floors. IDC reported that global window cleaning robot shipments grew 70.4 per cent in 2025, while robotic lawn mower shipments increased 63.8 per cent. Robot vacuum shipments grew 17.1 per cent.

These are global figures. Their relevance to Southeast Asia depends on the households and properties each product can serve.

Some manufacturers frame this expansion around adapting existing engineering. ECOVACS CEO David Qian, in comments released for IFA Berlin 2026, described “reusing around 80 per cent of our core technologies and adapting 20 per cent to each scenario”.

For Southeast Asian distributors and hardware founders, this raises a practical question: how much of the operating model can travel with the technology? A common engineering base still leaves each category with its own customers, installation requirements and maintenance needs. The useful measures would include shorter setup times, fewer customer interventions, lower repair costs and reliable operation in local homes.

Also Read: Singapore’s robotics dominance is a warning sign dressed up as good news

Singapore makes the distinction tangible. Department of Statistics figures show that 77.2 per cent of resident households lived in HDB dwellings in 2025. A floor cleaning robot and a lawn mower therefore address very different customer pools.

A distributor considering outdoor robotics needs to establish the number of suitable properties, the cost of reaching their owners and the support required. Overall interest in smart homes provides too little information to make that investment decision.

The service model is already local

Competitors’ support arrangements show where local businesses participate in the market.

Roborock’s Singapore support page identifies W Network as its service centre operator. Its Malaysian site directs customers to Hello Smart for after-sales support. These are concrete examples of local organisations handling the practical work behind an international brand.

Dreame’s Singapore warranty policy also makes the operating requirements visible. Coverage applies to purchases through authorised channels. Repairs are generally completed within 14 business days of drop-off or quotation approval, with revised timelines when required parts are unavailable.

Published policies establish service terms; they do not establish which brand delivers the best repair experience. They do, however, identify issues that matter to customers and operators: authorised sales channels, access to spare parts and the time a household spends without its appliance.

For businesses expanding across borders, each additional market requires a workable answer to those issues. Inventory, technician capacity and responsibility for warranty costs need to be settled alongside the sales agreement.

Also Read: Try Everything 2026 puts AI, robotics and cross-border capital at the centre

Where regional founders can build value

This creates potential openings for Southeast Asian businesses in distribution, authorised repair, spare parts logistics and customer onboarding.

The economics require scrutiny. A repair operation needs enough installed devices to support trained staff and parts inventory. An onboarding service needs customers willing to pay, or a brand partner that can justify funding it through fewer support requests and returns.

Local knowledge becomes commercially useful when it produces a measurable improvement: fewer failed deliveries, faster repairs, lower support costs or better customer retention.

Founders should therefore assess expansion using evidence from customers already served. Relevant measures include margin after channel and warranty costs, repair turnaround, parts availability and the frequency of customer intervention during the first months of ownership.

Investors can apply the same discipline. Shipment growth and a wider catalogue provide an incomplete account of execution. Service performance helps reveal whether a business can keep its promises as its customer base expands.

Southeast Asia’s home robotics opportunity is substantial enough to deserve that scrutiny. Before entering another country or adding another category, an operator should be able to show that its existing customers can be served reliably at a sustainable margin.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Southeast Asia’s robot vacuum boom puts local execution to the test appeared first on e27.

Posted on — Leave a comment

Know your tenant: US chip rules turn Southeast Asia’s data centres into compliance desks

For most of this decade, Southeast Asia’s data centre operators asked prospective tenants a short list of questions: how many megawatts, how long a lease, how soon. Since November 16, that list has grown, and the question that matters most is no longer how much power a customer needs but who ultimately owns it.

That is the clearest reading of a new note from BMI, the research unit of Fitch Solutions, on the US Commerce Department’s latest ruling on advanced AI chips. The rule requires licences for shipments of specified advanced semiconductors to Indonesia, Malaysia, Singapore, Thailand and Brazil. More pointedly, it stretches export controls beyond the chip itself to whoever rents the computing power it produces.

Also Read: The Nvidia clampdown is a warning for Southeast Asia’s AI boom

The ruling arrives in the wake of the Megaspeed affair, which put a Singapore-based Nvidia cloud partner under investigation over suspected chip diversion to China (the company denies wrongdoing) and has already cost Jensen Huang his job at Nvidia. Washington’s message to the region is blunt: prove the leak does not run through your server halls.

A 750-chip tripwire, measured in compute

The sharpest clause targets data centres outside the US that supply more than 12 million aggregate total processing performance (TPP), a yardstick of raw computing muscle, to companies headquartered in mainland China, Macau or other restricted destinations. Cross that line and the operator needs a US licence.

BMI puts the threshold at roughly 750 Nvidia H100-class chips, with capacity aggregated across linked sites. That last detail closes the obvious escape hatch: splitting a cluster across several halls to stay under the line will not work.

The goal is to shut the loophole that lets companies in restricted countries rent advanced American computing overseas without importing a single chip. Rules on direct chip sales to China are unchanged; what changes is oversight of the ultimate user. The ruling was slated to take effect on November 19, upon publication in the Federal Register.

Nvidia and AMD become the region’s KYC desk

The immediate burden falls not on governments or even landlords, but on vendors. BMI says chip and server suppliers, including Nvidia and AMD, face the most immediate obligations. They need licences to ship specified chips to the five countries unless a deal qualifies for an exemption covering US-headquartered customers, which comes with monthly reporting.

Suppliers must also extract information from operators on the identity and ownership of their tenants, including anyone tapping compute through the facility. Operators that refuse will be treated as needing a licence until the concerns are resolved. In effect, the world’s dominant AI chip vendor becomes Southeast Asia’s know-your-customer (KYC) officer.

The net reaches upstream too. Foreign foundries and packaging firms must treat advanced chips designed by companies in the five countries, or in other restricted jurisdictions, as controlled unless the designers hold the necessary approval. For a region that has built its semiconductor story on assembly, packaging and a fledgling design scene, that is fresh paperwork on every order.

Also Read: AI’s bottleneck economy is taking shape across Southeast Asia’s chip supply chain

The US Bureau of Industry and Security (BIS) expects about 100 additional licence applications a year, each judged case by case. A hundred files is modest for a region pouring billions into server halls. Washington appears to be aiming a rifle rather than a shotgun, but nobody yet knows how often it will pull the trigger.

Two data centre markets are taking shape

BMI expects projects built around Chinese tenants to take the heaviest hit, while those serving US tech firms shelter behind the exemption. Existing operators supplying above-threshold compute to Chinese-headquartered customers get 90 days from the effective date to secure licences, which puts the deadline at February 17, 2027.

This is the split the region has been edging towards for months. Johor’s data-centre future is increasingly being written in Washington, Kenanga Research’s reading of the rules tilts Malaysia’s growth towards American hyperscalers, and Indonesia is paying the compliance bill for Asia’s AI chip race.

BMI’s more interesting call is that affected customers will be broken up rather than locked out. It expects them to drift towards smaller, fragmented and heavily documented compute arrangements. Translation: less capacity per contract, more lawyers per megawatt.

One carve-out deserves scrutiny. Compute used solely for commercial recommender systems, the algorithms that decide what you watch or buy next, is excluded as long as operators meet reporting requirements. Capacity inside the US falls outside this licensing clause altogether, a reminder that the rule polices the periphery, not the core.

Regulatory credibility is the new tax break

Licence decisions will also weigh how well the five countries enforce their own export controls, nudging governments to tighten domestic frameworks. Singapore has already signalled it will explore trusted access arrangements for its companies.

That turns regulatory credibility into a competitive asset, much as tax holidays and cheap land once were. A country that convinces Washington its corridors are clean earns faster licences and calmer investors. Given the region’s chip-smuggling problem, that credibility will not come cheap.

Don’t bank on a Shenzhen reprieve

BMI expects the substance of the rule to survive a public consultation that closes in January 2027, though details could change. Pressure in Washington is running the other way: three chip-control measures are reportedly included in the Senate version of the fiscal 2027 defence bill.

Also Read: Why Southeast Asia cannot build sovereign AI on borrowed choices

The research house did flag two possible softeners: President Donald Trump’s meeting with Chinese President Xi Jinping in Shenzhen on November 19, and the January 10, 2027 expiry of the current US-China trade truce. Either could tempt Washington to trade concessions, and the administration has bent before, permitting case-by-case licensing for Nvidia’s H200-class chips in January 2026.

Still, BMI warns against assuming a repeat. Its advice for the first quarter of 2027 is to plan for full compliance by February 17 and treat any relief as upside, not baseline.

For founders, operators and investors weighing a stake in the region’s data centres, that is the real takeaway. Southeast Asia’s compute pitch used to rest on land, power and proximity. Ownership structures and the identity of whoever sits behind each rack now belong on that list, and unlike power shortages, they cannot be fixed with a bigger substation.

The post Know your tenant: US chip rules turn Southeast Asia’s data centres into compliance desks appeared first on e27.

Posted on — Leave a comment

Indonesia exports its talent, Korea captures the value: The K-pop lesson Jakarta can’t ignore

Hearts2Hearts’ debut under SM Entertainment in February 2025 (an eight-member girl group designed from the start for global audiences, with members drawn from multiple countries) was one of the more visible signals of how Korean entertainment has reshaped its talent acquisition model over the past five years. The group’s structure illustrates the dominant pattern in late-2020s K-pop: agencies in Seoul recruiting talent from across Asia, training them inside the Korean idol system, and releasing them through Korean entertainment IP infrastructure.

The parallel emergence of Indonesian artists, including No Na and others navigating between the Indonesian market and broader Asian opportunities, illustrates the other side of the same phenomenon. Indonesian talent is increasingly mobile. Indonesian entertainment industry capacity to capture the economic value of that mobility has not kept pace.

After 15 years inside Indonesian risk functions, I have come to see this asymmetry as worth naming. The structural questions underneath are the same the country has faced in commodities, manufacturing, and increasingly in technology talent. Value capture from human talent flows in Asia is a real economic question, not a cultural one.

What the Korean model does

Korean entertainment, particularly the major-agency tier, has built one of the most sophisticated talent recruitment, training, and IP capture systems in any creative industry globally.

  • Pre-debut investment scale. The major Korean agencies invest meaningfully (multiple years of housing, training, language education, performance development) into talent before commercial release. The investment is funded by the agency, capitalised on the agency’s balance sheet, and recovered through revenue-share contracts that run for years after debut.
  • Cross-border recruitment scaled to industrial level. Hearts2Hearts is one of many groups in the current generation with members from across Asia and beyond. The recruitment process is institutionalised: auditions across multiple countries, scouting networks, partnerships with local entertainment companies. Korean agencies capture talent before competing local industries do.
  • IP and revenue infrastructure designed to capture compound value. K-pop revenue is distributed across music, performance, advertising, merchandise, fan engagement platforms, content production, and licensing. The agencies control most of these revenue streams structurally. The talent’s share, while meaningful, is a fraction of the total economic activity their work generates.

Also Read: After Megaspeed, Indonesia pays the compliance bill for Asia’s AI chip race

Why the Indonesian industry captures less

Three structural factors limit how much value Indonesia captures from its own talent.

  • The development infrastructure gap. The training, performance development, and content production capability available in Indonesia is meaningfully smaller than what major Korean agencies offer. Indonesian artists who reach a certain stage often face a choice between the smaller domestic infrastructure or moving to where the larger infrastructure exists.
  • The IP and revenue model gap. Indonesian entertainment industry contracts, revenue distribution mechanisms, and IP infrastructure have not evolved as rapidly as the underlying market. Even when Indonesian artists achieve regional or international visibility, the share of value captured by Indonesian entities is structurally smaller than the equivalent in Korea or Japan.
  • The fan economy infrastructure gap. The platforms that monetise fan engagement at scale (fan communities, merchandise distribution, content subscriptions, ticketing) are predominantly built by non-Indonesian actors. Indonesian fans of Indonesian artists frequently engage through platforms that capture value outside Indonesia.

Lessons learned

Five principles emerge from observing the asymmetry.

  • Talent infrastructure is economic infrastructure. The countries that build serious training, development, and IP capture systems retain economic value from their talent. The countries that do not export the talent and absorb the leakage.
  • The cross-border talent flow is structural, not preventable. Indonesian talent will continue to move where infrastructure is more developed. The question is not whether the flow happens; it is whether Indonesia builds the infrastructure that captures part of the value before the flow.
  • Revenue model design matters more than talent quality. Indonesia has produced and continues to produce world-class musical and performance talent. The constraint is not talent quality. It is the contract structure, IP framework, and revenue infrastructure that captures the value the talent generates.
  • Cultural export requires institutional infrastructure. Korean entertainment did not become dominant globally by accident. It is the product of three decades of deliberate institutional investment: agencies, training systems, content infrastructure, government coordination, IP frameworks. Indonesia’s equivalent investment has been more episodic and less coordinated.
  • The next generation can change the curve. The infrastructure gap is not destiny. New platforms, models, and institutional approaches can move Indonesia’s value capture position over a five- to 10-year horizon. But the move requires deliberate building, not passive consumption of whatever model others have built.

Also Read: Entering Indonesia? The tax issues startups underestimate

The macro stakes

Asia’s entertainment industry economics are reshaping faster than most policy frameworks recognise. Korean entertainment is now a significant export category for Korea. Indonesia’s larger population and market potential make the underdevelopment of its entertainment industry infrastructure a more consequential gap than the others in the region.

Hearts2Hearts and the cohort of Korean groups with multinational members will continue to emerge. Indonesian artists, including No Na and others, will continue to navigate between markets. The question is whether the institutional infrastructure to capture more of the value of Indonesian talent (for Indonesia) gets built deliberately in the next decade, or whether the asymmetry continues to widen. The talent geography question is not going to answer itself.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Indonesia exports its talent, Korea captures the value: The K-pop lesson Jakarta can’t ignore appeared first on e27.

Posted on — Leave a comment

Ecosystem Roundup: Singapore fintechs pour money into AI, but just 11 per cent run agents live

Singapore’s fintechs have embraced AI faster than almost anyone, but they have not handed it the keys.

A new Kapronasia brief, co-published with Airwallex and the Singapore FinTech Association, finds 86% of Singapore fintechs already deploy AI, against 65% of the city-state’s traditional financial institutions. Yet only 11% run agentic AI in live environments, although 42% say they are ready.

Workforce adoption scores just 39 out of 100, and nearly six in 10 firms say half or fewer of their staff use AI. Caution is sharpest where the stakes are highest: 59% report no AI at all in lending and underwriting. Money keeps flowing regardless. Some 94% expect AI budgets to rise next year, but only 42% achieved the 26–50% returns that 61% expected, and just 4% name net-new revenue as AI’s biggest payoff. Regulators are not the brake: 86% engage with MAS’s Project MindForge, yet only 9% have built controls mapped to it or IMDA’s governance framework.

The data is self-reported, and Airwallex features its own product as a case study. Still, the lesson for the region’s founders is clear: wiring AI into the plumbing is easy; trusting it with customers’ money is not.

Editor.


REGIONAL

Everful raises US$48.5M to turn overseas sourcing into AI software: Vertex Ventures SEA & India and January Capital co-led the round. The Singapore- and San Francisco-based startup claims a US$100M revenue run rate by year-end, though its unverified numbers point to low-ticket goods and thin margins.

Solana Foundation takes majority stake in Malaysia-born CoinGecko: The undisclosed deal follows a reported US$500M sale process. With 18M users and 180,000 API integrations, the crypto data platform now plans Solana-powered trading, raising questions about its neutrality as rivals pitch themselves as chain-agnostic.

Iterative scraps its SEA batch, takes founders to San Francisco: The Singapore accelerator will not run a Winter 2027 cohort. Its new SEA to SF programme pairs US$150,000–500,000 cheques with a month in San Francisco, a blunt verdict on where AI’s frontier now sits.

Botsync raises US$11M to take its Physical AI platform to the US: IvyCap Ventures stepped up from co-investor to lead. The Singapore robotics firm, now past US$20M in total funding, wants its SyncOS orchestration software to run whole factory floors, not just its own forklifts.

DBS lets institutions mint and redeem USDC directly with Circle: Eligible clients get 24/7 USD-USDC settlement, cutting reliance on secondary-market liquidity. DBS research says 88% of large corporations already deploy or pilot tokenised money; both firms sit in MAS-led Project BLOOM.

Tether, Further put US$10M into Singapore’s MetaComp: The extension lifts MetaComp’s Pre-A round to about US$45M, after a US$35M close in March. The MAS-licensed firm links fiat and stablecoin rails and processes over US$1B a month across Asia, the Middle East and Africa.

SEA’s digital consumer loan book hits US$19.8B; Monee holds 45%: Indonesia leads with US$8.6B outstanding. Monee, Grab, GoTo and GCash control about 60%, and Momentum Works says the window for new independent players to reach scale and venture-level exits has closed.

Sokin goes fully live in Singapore after securing its MAS licence: Licensed as a Major Payment Institution in April, the payments firm now offers overseas businesses SGD accounts without local entities. It plans to double Singapore headcount by the end of Q1 2027.

Promoted staff in Singapore earn about 9% less than outside hires: Omni HR compared employees at the same employer and experience level, suggesting loyalty carries a pay penalty. For startups competing for scarce talent, cheaper internal promotions may cost them in retention.

Canadia Impact Fund makes first French bet with Choose and Connect: The corporate venture arm of Cambodia’s Canadia Group backed a free career guidance platform for teenagers, an unusual cross-border pick that skips the deeptech and AI plays typically headlining such deals.

INTERVIEWS & FEATURES

Johor’s data-centre future is being decided in Ohio and Washington: US voters spent the midterms arguing over who pays for data-centre power and water. Southeast Asia, now the overflow tank for compute, got press statements instead, while decisions shaping Johor are made abroad.

Cheques, bans and bell-ringing: 50 Asian AI firms that defined 2026: Capital arrived at a scale that would have looked like a typo three years ago, with DeepSeek reportedly raising about US$15B in one round. The list tracks the funding, bans and listings that shaped Asia’s year.

GITEX GLOBAL 2026 shifts its AI focus from the cloud to machines: The 46th edition opens with The GITEX Summit at Dubai World Trade Centre on 7 December, followed by a four-day expo at Expo City, with sharper focus on AI infrastructure and robotics.

INTERNATIONAL

Anthropic opens 33% above IPO price at over US$2T valuation: Shares opened at US$111.50 under ticker ANT, ahead of Meta and Tesla by market cap. With private placements, Anthropic expects about US$220B in cash, more than any company has held, one analyst said.

Anthropic’s IPO tests AI’s circular financing loop: Amazon and Google, both compute suppliers, hold stakes worth a combined US$307.9B at the IPO price. Anthropic has committed US$900B, mostly to compute; SemiAnalysis says it needs US$280B in revenue over four years.

Nvidia board rules out bringing Jensen Huang back as CEO: Chairman Harvey Jones said the board unanimously accepted Huang’s resignation, despite his offer on X to return. Interim CEO Colette Kress stays while the board weighs internal and external candidates.

SEC reportedly probes whether Nvidia’s disclosures misled investors: The civil probe, reported by The Wall Street Journal, follows Huang’s exit over an affair with an ex-Megaspeed executive. Nvidia’s stock has fallen more than 10% since, even as quarterly revenue doubled to US$115.8B.

Musk likens Huang’s ouster to benching Michael Jordan: At DealBook, Musk said he did not think Huang was disloyal to the US and would hire him. Anduril’s Palmer Luckey also backed reinstatement, while Nadella and Altman reaffirmed confidence in Nvidia.

Nadella says 60,000 buyouts aim to rebuild Microsoft around AI: He insists AI is not replacing engineers but reshaping their jobs into managing agents rather than doing the work. An 11-person Cowork team, he said, builds the factory that builds the product.

HKTDC brings 370-plus startups to Hong Kong’s IP and founder forum: The 16th Business of IP Asia Forum and 18th Entrepreneur Day run on 19 and 20 November, with IP financing and AI licensing on the agenda for founders scaling across Asia.

CYBERSECURITY

Google finds AI agents running break-ins with no human involved: Google Threat Intelligence Group’s latest AI Threat Tracker says software agents now scan, exploit and adjust on their own, turning human hackers into supervisors of automated intrusions.

UK tribunal orders government to confirm secret Apple backdoor order: The Investigatory Powers Tribunal rejected the Home Office’s national security argument. Apple refused the demand and pulled Advanced Data Protection for UK users; a hearing on releasable details is expected next year.

SEMICONDUCTOR

After Megaspeed, Indonesia pays the compliance bill for AI chip race: Maybank warns US licensing will thin Chinese-linked demand and hit pure-play AI operators hardest. Indosat’s Zankore GPU build-out faces delays, but its NeoCloud revenue target is under 10% of forecast 2027 revenue.

BMI: US rules make Nvidia and AMD gatekeepers of SEA compute: Data centres supplying over 12 million TPP, about 750 H100-class chips, to China-linked firms need licences by 17 February 2027. Suppliers must verify who owns each tenant, splitting the market into two camps.

OCBC: China-linked firms hold 34% of Malaysia’s data centre capacity: The share is 38.4% of Thailand’s approved capacity and about 70% in Batam. Vietnam, the Philippines and India could gain a relative advantage as they escape the new destination-based licensing rules.

Nvidia stops selling H200 chips to China after Huang’s exit: Interim CEO Colette Kress ended the Trump-brokered arrangement. Jefferies says new US rules could trim US$44B to US$62B from Nvidia’s data centre revenue next year, with SEA customers at risk.

Kenanga: 750-chip ceiling tilts Malaysia’s data centres towards the US: US hyperscaler demand should cushion the blow, but non-US operators without confirmed end-users face delays. Sunway Construction’s US$175M project with K2 is the main exposure, and it is nearly complete.

SkyeChip, SiPearl plan Malaysian JV for an Arm-based compute platform: SkyeChip would hold 51% of the venture, targeted for December 2026. The non-binding MoU, signed during Macron’s Malaysia visit, depends on an Arm licence and export control clearances.

AI

Mistral makes Singapore its APAC HQ, plans 200 staff by 2027: The Paris-based lab will build sovereign open-weight models for government, high-tech manufacturing and robotics. It already works with DSO and HTX, but must compete with OpenAI, Anthropic and Google for scarce talent.

Nadella: AI agents will reshuffle the web’s advertising economics: Ad models built on human eyeballs may shift to transactional deals. Nadella suggested that if Amazon keeps blocking agents from its site, rivals such as Shopify could win by letting them in.

Altman, Brockman reject a ‘gradual handover’ of control to AI: Brockman said chain-of-thought monitoring is weakening as models grow more capable. After the Hugging Face breach, OpenAI moved 25% of production engineers to security work.

LinqAlpha opens Tokyo office to court Japanese financial institutions: The AI research platform, which raised a US$22M Series A in July, already has a Singapore office. Its buy-side clients manage over US$5T, and a Bank of Japan survey found 90%-plus of institutions use generative AI.

THOUGHT LEADERSHIP

Dubai sells trust, not just capital, to SEA founders: Astrid Dang argues SEA founders should treat Dubai as a gateway to markets beyond their region, not merely a place to raise money, given the UAE’s decades-long build-out in trade, logistics and finance.

Indonesia exports K-pop talent while Korea captures the value: Hearts2Hearts, SM Entertainment’s multinational girl group, shows how Korean labels now recruit globally. Bobby Yulandika Putra argues Jakarta must build its own value chain rather than simply supply performers.

Vietnam’s robot vacuum boom puts local execution to the test: Vietnamese consumers spent US$92.3M on robot vacuums in the year to June 2026, up 137.6% by value, per GfK. ECOVACS, Xiaomi, Roborock and Dreame lead; Giovanni Puglisi says execution decides the winners.

Developer files complaint against 7 AI giants over ‘linguistic tax’: Ashish Airon says identical prompts in Indian languages can cost up to three times more to process through commercial LLM APIs, penalising enterprises building AI for non-English users.

AI should cut wealth management fees, not just fatten margins: AI savings let firms lower fees, serve smaller accounts or keep the profit. David Ng argues the benefits will not reach clients automatically; firms must decide deliberately how to share them.

Singapore’s wealth onboarding bottleneck is the back and forth: Bank of Singapore’s 2025 AI tool shortened source-of-wealth report preparation, yet Kew Yoke Ling says the ultra-wealthy clients Singapore most wants still get stuck in the queue.

Cross-border payments need clarity, not just speed: An exporter needs to know what arrived, what was deducted and whom to call when figures do not match. Gaurav Shisodia argues the debate’s focus on speed and price misses what global businesses need.

Ranking on Google no longer guarantees a ChatGPT recommendation: A Petaling Jaya business owner found her firm missing from ChatGPT’s recommendations despite a first-page Google ranking. Adrian Wong explains why AI search visibility follows different rules.

A startup took founder bias out of its roadmap. Customers walked: Writing from a one-north incubator, Vincent Tan recounts how removing founder bias from planning cost his team its customers. He draws on years of watching the dot-com bust from San Francisco.

AI productivity gains hide a quiet professional identity crisis: Psychologist Dr Zoe Wyatt-Potage describes a Singapore fintech research lead whose reports once took hours, and examines how AI’s speed can unsettle people’s sense of professional identity.

Fix internal workflows before adding AI to your product: Most software firms ask how to add AI to what they sell. Daniel Tan argues the more practical first move is to fix how work moves inside the company.

Your AI startup may be a very fast nonsense factory: Self-taught builder and former paramedic Faheem Aizat Kamsan says vague instructions let AI build the wrong thing confidently, a lesson learnt building OnTheRice between 7 and 11 October.

AI will transform science without needing AI geniuses: Scientific progress has long been limited by human attention. John Millar argues AI will remove the attention bottleneck, much as electronic databases once transformed literature searches.

AI slop wastes money and carbon, and offers firms a chance to lead: Adam Goulston argues LLMs learnt slop from trite corporate writing, but AI slop carries business and climate costs that human drivel never did, giving firms a chance to lead on quality.

High-performing teams lead through conflict over AI adoption: A Writer and Workplace Intelligence study found 42% of executives say AI adoption is tearing their organisations apart. Rubyni Karun argues strong teams handle conflict rather than avoid it.

AI speeds up video production; handling feedback is the real test: Generating a shot is one step. Xiaohu Hou says the real challenge in commercial work is turning a brief into video while absorbing changing feedback against tight deadlines.

A former reality TV producer explains how she builds apps with AI: Reality TV ran on master spreadsheets, schedules, itineraries and endless emails. Joan Leong describes how that pre-production discipline now shapes the way she uses AI to build apps.

Value, not connections, is becoming the currency of business access: Joycefaith Ang argues AI and technology make networks less of an inherited advantage, weakening the old maxim that your network is your net worth.

America’s affordability election puts corporate America on trial: Affordability, from groceries to healthcare, defined the 2026 midterms. Shawn Balakrishnan argues the consequential question comes after counting: whether companies will face accountability on prices.

Crypto’s hype cycle is dead; four narratives are replacing it: Play-to-earn collapsed, metaverses stayed empty and PFP NFTs lost their premium. Jenny Zheng calls the fatigue structural rather than cyclical and maps the four narratives taking over.

Anthropic’s IPO is a sentiment gauge, not a crypto crash trigger: Bitcoin fell 2.39% to US$62,731 as total crypto market cap slipped to US$2.15T. Anndy Lian blames a familiar macro squeeze, not Anthropic’s offering.

The post Ecosystem Roundup: Singapore fintechs pour money into AI, but just 11 per cent run agents live appeared first on e27.

Posted on — Leave a comment

The fastest raise in Singapore may still be a resignation letter

Staying loyal to an employer in Singapore seems to come at a price. Employees promoted into a role earn about 9 per cent less than people hired into the same role from outside, at the same employer and experience level, according to a new report from HR and payroll platform Omni HR.

The Singapore and Malaysia Salary Guide Report 2027, produced with salary-transparency platform PAYGAP Asia, also finds that 28 per cent of new hires in Singapore leave within their first year. Two in three of those exits are resignations.

Also Read: Indonesia’s AI hiring gap is real, just not 28×

The timing is deliberate. HR teams are setting their 2027 pay budgets, and Singapore employers plan an average salary increase of 4.1 per cent, the lowest of six Southeast Asian markets, according to consultancy Aon.

Why a double-digit raise still loses to the market

On paper, a promotion pays well. The median raise that comes with one in Singapore is 12.1 per cent, more than double the 5.1 per cent median for a routine raise.

The problem is the starting point. Omni HR argues that a promotion raise is usually calculated from the salary the person was originally hired at, while an outside hire is priced at today’s market rate. If market pay has risen since the insider joined, even a double-digit bump can leave them behind a newcomer. The explanation is plausible, but the release presents it as Omni HR’s reading rather than something the data proves directly.

The cost argument is easier to make. According to the Singapore National Employers Federation (SNEF), 83 per cent of employers cite rising manpower costs as a challenge, the most common workforce concern. Outside hires cost more in salary and in recruitment fees, and the hidden cost of a bad hiring decision can be bigger still. Omni HR’s point is that paying promoted staff the market rate is the cheaper option.

That should resonate with startups, which already struggle to hire against banks, multinationals and Big Tech. Underpaying the people they already have only makes the problem worse.

The first year is where pay stalls

The report finds that new hires in Singapore wait a median of 13.9 months for their first raise. Nearly one in four wait well into their second year.

The freeze is not limited to newcomers. Twenty-two per cent of employees in Singapore have gone 12 months or more without any change to their pay, rising to 29 per cent among those with more than a year of tenure. That group is 3.9 times more likely to resign.

For employers, that is a costly leak. SNEF found that 59 per cent of Singapore employers rank attracting talent as their top HR priority. If more than one in four new hires leave within a year, much of that recruiting spend is lost before it pays back. Similar strains are building elsewhere in the region, with signs that the Philippines’ calm job market may be hiding a resignation wave.

Read the fine print before the headline

The report combines anonymised payroll and HR records from more than 200 employers across seven APAC markets with salary and sentiment submissions from PAYGAP Asia users. The release does not say how many Singapore employers or employees sit behind the 9 per cent figure, or which functions and seniority levels drive it. Self-reported salary submissions also tend to over-represent people who are already unhappy with their pay.

The 3.9 times figure needs the same care. It shows a correlation, not a cause, and Omni HR itself notes that employees without a raise are also more likely to be let go. A frozen salary may simply mark someone a manager has already given up on, a pattern consistent with how merit in startups is often decided before performance reviews begin. In those cases the missing raise is a symptom rather than the reason the employee leaves.

The report’s recommendations are sensible. Omni HR says employers should benchmark promotions against the market rate for each role, budget promotions separately from the merit pool at roughly 12 to 14 per cent, run a quarterly report of staff with no pay change in 12 months, and hold the first structured check-in with new hires within weeks.

The arithmetic is harder. A separate promotion pool of 12 to 14 per cent sits awkwardly beside an overall budget of 4.1 per cent. The money has to come from somewhere, most likely fewer promotions or thinner raises for everyone else.

“Employee turnover rarely happens overnight. Our findings suggest that changes in pay, or the lack of them, can reveal early signs of disengagement before an employee actually resigns,” said Brian Ip, CEO of Omni HR.

Ip then pitches Omni HR’s platform as the tool that brings payroll and employee data together to spot those risks. In that sense, the report doubles as marketing: it diagnoses a problem that its publisher’s software promises to solve. That does not make the data wrong, but readers should weigh it accordingly.

A crowded market selling the same dashboard

Omni HR has used research as a calling card before, from a report on AI readiness among APAC HR teams to its August study on manager behaviour, The Tiger Boss Problem. It also acquired MajuHR last year to add chat-native features.

Also Read: The AI layoff trap points straight at Southeast Asia

For employees, the message is uncomfortable but familiar: the fastest raise often comes from leaving. For employers, the fix depends less on dashboards than on managers who are equipped to have pay conversations before a resignation letter arrives.

Until promotions are priced against today’s market rather than an old offer letter, Singapore companies will keep paying a premium to replace people they could have kept.

The post The fastest raise in Singapore may still be a resignation letter appeared first on e27.

Posted on — Leave a comment

Your brand ranks on Google, so why doesn’t ChatGPT recommend it?

A business owner in Petaling Jaya asked me something recently that I keep coming back to. A customer told her she’d “asked ChatGPT for recommendations” and her company wasn’t on the list, even though she ranks on Google’s first page for her main keyword. She wanted to know what she’d done wrong.

Probably nothing. But the question shows a shift that many brands in Malaysia and the wider region haven’t caught up with yet. People now get answers, not lists of links. Being found means being named in those answers.

The argument: SEO isn’t dead, but it’s no longer the whole job

I don’t think AI search replaces SEO. ChatGPT, Gemini, Perplexity and Google’s AI Overviews still pull heavily from the open web. If your pages are thin, slow or unclear, you’re unlikely to show up in either place.

What changes is the unit of success. Classic SEO rewards ranking a page. AI answers reward being a clear, consistent, well-corroborated entity: a brand the model can describe confidently because many trustworthy sources say the same things about it.

The shift is early but real. The Reuters Institute Digital News Report 2025, which surveyed audiences in 48 markets including Malaysia, found that 7 per cent of people already use AI chatbots for news each week, rising to 15 per cent among under-25s. The same report lists Malaysia as one of the “social-first” markets, where people now get more news from social and video platforms than from traditional media. Discovery here is already moving away from the classic search results page.

Reuters Institute Digital News Report 2025: 7 per cent of people across 48 markets including Malaysia use AI chatbots for news each week, 15 per cent among under-25s, and Malaysia is a social-first market.

What AI assistants seem to rely on

From the work our team does, and from testing prompts for clients in English, Malay and Chinese, a few patterns stand out. These are observations, not guarantees. The platforms don’t publish their ranking logic.

  • Consistency across the web. If your company name, address, services and founding story differ between your website, Google Business Profile, LinkedIn and directories, a model has less to go on. The brands that get described accurately usually describe themselves the same way everywhere.
  • Third-party mentions over self-description. Your “About us” page is one source. A local news feature, an industry association listing or a reputable review page is a corroborating source. Answers seem to lean towards what several independent sources agree on.
  • Pages that answer questions directly. Content written as plain answers (“How much does X cost in Malaysia?”, “What’s the difference between A and B?”) is easier to quote than marketing copy. FAQ sections, pricing ranges and clear definitions help.
  • Language coverage. This matters in Southeast Asia. A Malaysian customer might ask in Bahasa Malaysia, Chinese or Manglish. If all your content is in English, you may simply be missing from answers to those queries.

Also Read: “AI amnesia” is quietly costing Southeast Asian brands their customers

You can test this yourself in five minutes. Ask ChatGPT, Gemini and Perplexity something like “What are the best [your category] brands in Malaysia?”, then ask again in Bahasa Malaysia or Chinese. Note which brands get named, how they’re described, and whether yours shows up at all. The results will vary between tools and between languages. That variation is the point.

Where this argument has limits

I’ll be honest about what we don’t know. AI answers vary from one session to the next. The same prompt can give different brands an hour later. Personalisation, location and model updates all move the target. Anyone selling a “guaranteed ChatGPT ranking” is overselling.

Measurement is also immature. We can track Google rankings precisely. For AI answers, the best most teams can do is run a fixed set of prompts on a schedule and log whether the brand appears and how it’s described. That’s directional, not a hard metric.

And for many Malaysian SMEs, the basics still bring the biggest return. A complete Google Business Profile and a fast, clear website will do more this quarter than any AI-specific tactic.

A practical checklist for the next 30 days

If you run marketing for a brand in Malaysia or Southeast Asia, this is what I’d do:

  • Write one canonical brand description. Two or three sentences covering who you are, where you’re based, what you do and for whom. Use it word for word on your site, social profiles and directory listings.
  • Audit your entity footprint. Search your brand name on Google, then ask ChatGPT, Gemini and Perplexity “What is [brand]?” and “Who are the best [category] in [city]?”. Note what’s wrong, missing or outdated.
  • Fix the facts at the source. Correct outdated addresses, services and names on the listings you control first.
  • Earn two or three independent mentions. Industry associations, local media, partner pages and credible directories. Relevance matters more than volume.
  • Publish answer-shaped content. Take the ten questions your sales team hears most and answer each one plainly on your site. Include price ranges where you can.
  • Cover your customers’ languages. At minimum, translate your core pages and FAQs into the languages your buyers actually search in, not with raw machine translation.
  • Add structured data. Organisation and LocalBusiness schema with correct name, address, contact details and sameAs links to your official profiles.
  • Set up a monthly prompt log. Twenty fixed prompts, run monthly, recorded in a spreadsheet. Watch the trend, not single results.

Also Read: Japan is moving into Southeast Asia faster than the West, and most brands haven’t noticed yet

An 8-step checklist for the next 30 days to help Malaysian and Southeast Asian brands show up in AI search answers.

The takeaway

Search used to be about getting a click. More and more, it’s about getting a mention you never see happen. The brands that win in AI answers across Southeast Asia won’t be the ones that game a new algorithm. They’ll be the ones that are easy to describe accurately, in more than one language, by more than one source.

That’s less glamorous than a hack. It’s also much harder for a competitor to copy.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Your brand ranks on Google, so why doesn’t ChatGPT recommend it? appeared first on e27.

Posted on — Leave a comment

If nobody knows you, nobody owes you their trust: How founders can build it anyway

Here is an uncomfortable truth about business: If nobody knows you, nobody owes you their trust.

An established company does not walk into a room carrying only its product. It carries years of reputation, relationships, customer references and a network built long before the current opportunity appeared.

A founder starting out has none of that. No financial backing behind the manufacturing. No existing relationships to sell into. No name that makes a stranger pick up the phone.

Imagine starting without connections, without capital, without anyone willing to vouch for you. How good can the product really be if there is no funding behind the development? Who is actually going to buy from an unknown name? That is how a brand dies, often before anyone outside the founder’s circle ever hears of it.

But hold on.

If this describes your business, the answer is not to give up before starting. It is to understand something most founders miss.

The real reason businesses fail is rarely the one they blame

The businesses that fail are not usually the ones with the biggest problems. They are the ones that never figured out what their actual problem was. 

A founder who knows exactly what is broken, whether that is distribution, credibility, capital or reach, is already most of the way to fixing it. Once a problem is named, it can be scoped, tested and worked through. The businesses that quietly disappear are often the ones still guessing at what is wrong.

This is the real disadvantage of starting without a network. It is not simply that doors stay closed. It is that a founder without connections may have fewer people to help them see clearly what the actual obstacle is, until it has already cost them the business.

But the world has changed

Whether or not you have anyone to lean on right now, that is no longer the deciding factor it once was.

A business can build visibility without inheriting it. It can collaborate with businesses that already have what it is missing, and generate real evidence of value where a name and a reputation used to be required.

This is where, in my work with founders, the conversation about access connects to something larger. A business needs more than contacts. It needs a brand that gives people a reason to notice it, clear positioning that gives people a reason to remember it, a business model that gives partners a reason to work with it, and the capability to deliver on what it promises.

Also Read: The Philippines trusts AI more than anyone in ASEAN. Its rulebook hasn’t caught up

Collaboration can create what you did not inherit

This is where collaboration becomes more than networking.

A partnership should solve something.

Once the real problem is identified, collaboration becomes one way to solve it. What that collaboration looks like depends entirely on what is missing.

If the problem is credibility, one route is proximity. A new business places its product or service in front of an audience that already trusts an established name, through a genuine partnership rather than simply asking people to believe in it, so they can experience the value directly.

If the problem is reach, collaboration can create access to a shared audience. Two businesses with complementary, non-competing offers can combine their strengths, allowing each side to reach people it may not have reached alone.

If the problem is capability, a strategic partner can provide expertise the business has not yet built internally. Rather than spending years building a function from scratch, a business can work with an established operator that already has the relevant capability, while focusing its own resources on what it does best.

If the challenge is credibility in a new market altogether, a local partner can provide market knowledge, relationships and commercial context that an unknown business would otherwise have to build from the ground up.

The value of collaboration, however, is not simply the introduction.

It is what the collaboration allows the business to prove.

A customer experiences the product. A partner tests the proposition. Feedback reveals what needs to improve. Real customers can generate authentic content and user experiences that demonstrate value. A successful pilot can become evidence. A completed project can become a reference.

Instead of asking people to trust an unknown business because it says it is credible, the business begins to give them something real to trust.

Tested. Experienced. Proven.

That is how collaboration can begin to change the starting equation.

There is no single model

The right collaboration depends on the actual problem, not on what looks impressive.

A consumer brand may need customer proximity and word of mouth. A B2B company may need industry relationships and distributors. A business entering a new country may need local partners who understand the market and already hold relevant relationships.

Different businesses have different problems.

So why should they all follow the same model?

This is why the right partnership only makes sense once the business understands its own problem clearly and knows what it needs to achieve.

Also Read: AI saves me hours, but I don’t trust it blindly

Building step by step

Strategise the framework. Build the right partnerships. Track the progress. Track the growth. Scale the business.

In my work with founders, growth tends to follow this kind of sequence.

Branding establishes how the business is positioned. Business strategy determines where it can compete and how it can create value. Digitalisation and Agentic AI can strengthen its ability to operate, improve productivity and scale where they address genuine business needs. Connecting with the right overseas partners creates targeted access when the business is ready. Continued business development turns those relationships into ongoing conversations and commercial opportunities.

The framework is not meant to be identical for every company.

It is about understanding where the business is today, identifying what is holding it back, determining what needs to happen next and moving up the ladder step by step.

Building the network you were not born into

Whether you have someone to lean on right now or not, it does not have to decide the outcome. You may not have been born into a network. You may not have inherited the capital, reputation or relationships that make the first few steps easier. But you can build them.

You can build the brand. Strengthen the business model. Develop the capability. Find the right collaborators. Test what works. Track the progress. Learn from what does not. And keep moving towards the next stage of growth.

You do not have to solve every problem alone. The right strategy can help you see what needs to happen next. The right collaboration can help you solve what you cannot solve internally. The right business development can turn a relationship into an opportunity.

That is when the equation begins to change. You stop asking, “Who do I know?” and start asking, “What have I built that people want to be part of?”

A collaboration becomes a customer relationship. A customer relationship becomes a reference. A successful project becomes proof. Proof builds credibility. Credibility creates new conversations. And new conversations create opportunities that were once out of reach.

Over time, you are no longer waiting for someone to open the door. You have built enough value, credibility and relationships to create doors of your own.

You may not have inherited the network. But you can build the kind of business that builds its own network around it. And when that happens, access is no longer something you were lucky enough to be born with. It becomes something you have earned, built and can continue to grow.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post If nobody knows you, nobody owes you their trust: How founders can build it anyway appeared first on e27.

Posted on — Leave a comment

After Megaspeed, Indonesia pays the compliance bill for Asia’s AI chip race

For two years, Indonesia has pitched itself as the place where Asia’s AI workloads would live: a huge domestic digital market, surplus power and a government eager to cut ribbons on hyperscale campuses. Washington has just added a line to that brochure: terms and conditions apply.

In a note published on Tuesday, Maybank Investment Bank warned that tighter US export controls on advanced AI chips could slow Indonesia’s bid to become a regional AI and data centre hub, mainly by thinning out demand from Chinese-linked customers.

The pain, it argued, will fall hardest on pure-play AI infrastructure operators, while telcos such as Indosat Ooredoo Hutchison should escape with bruises rather than broken bones.

Also Read: Jensen Huang quits Nvidia, and Southeast Asia’s chip hubs inherit the fallout

The trigger is an interim final rule issued on 16 November by the US Commerce Department’s Bureau of Industry and Security (BIS). It requires licences for exports of advanced AI chips to Indonesia, Malaysia, Singapore, Thailand and Brazil over fears of diversion to restricted destinations, effectively putting Southeast Asia’s chip hubs on licence.

Washington is now watching the compute, not just the crates

The rule is narrower than a blanket ban but sharper than many expected. Shipments to companies headquartered in the US, or whose ultimate parent is American, are exempt, subject to reporting obligations. So are commercial recommendation systems used by online retailers and social media platforms.

Data centres that provide significant AI computing capacity to Chinese or other restricted entities, however, will need approval. Crucially, the controls also cover remote access, an acknowledgement that in the cloud era a GPU does not need to leave Jakarta to be used by someone in Hangzhou.

Maybank called the framework stricter than anticipated because approval hinges on the country and the identity of the end-user rather than a whitelist of approved firms. Operators that built order books around ByteDance, Alibaba and their peers now face a case-by-case process with no published playbook. Facilities already serving restricted customers have 90 days to comply.

The irony is that cloud computing, long sold as the great borderless utility, has made enforcement easier. Rented compute leaves a transaction trail, and Maybank noted that this visibility has shifted regulators’ focus from where chips are shipped to who consumes the computing power. Neighbouring Malaysia is already redrawing its data centre map around the same logic.

Why Jakarta landed on the list

Indonesia’s inclusion is not random. Maybank linked it partly to the Megaspeed case, which has raised questions about how well the archipelago polices chip re-exports.

Shipping records show Megaspeed International Indonesia, set up in May 2024, imported Nvidia hardware, including H100 chips, between 2024 and 2026. Indonesian police reportedly raided its Jakarta office in early 2026 at the FBI’s request and found unopened server racks but no GPUs. A government official alleged the chips had been sent to China, possibly by private aircraft. Megaspeed denies wrongdoing and says it complies with US law; nothing has been proven.

Proven or not, the episode fits a pattern that has been building for months as the region’s chip-hub ambitions collide with its smuggling problem. Maybank expects Jakarta to respond by tightening tenant verification, monitoring remote access to computing resources and generally behaving like a jurisdiction Washington can trust.

Also Read: AI’s bottleneck economy is taking shape across Southeast Asia’s chip supply chain

That reassurance comes at a price. Heavier compliance raises costs and could cool Chinese-linked investment, narrowing the customer base for operators that lean on Chinese tech giants for AI training and inference. The longer-term worry, Maybank said, is that controls could creep from chip licensing into network segregation requirements, making facilities costlier to run and less flexible.

Indosat’s Zankore bet hits a speed bump

Among the telcos, Indosat looks most exposed because of its planned GPU deployment for Zankore, its AI infrastructure project. New GPU imports will now need US licences, adding time and uncertainty to a build-out targeting 200 megawatts (MW) in its first phase by the first half of 2027 and 1 gigawatt (GW) by 2029. Customer verification adds a second layer of friction.

Management is targeting about US$300 million in NeoCloud revenue in 2027. Maybank estimates that would account for roughly 8 per cent of Indosat’s forecast FY2027 revenue of 66.7 trillion rupiah (about US$4 billion), meaning less than 10 per cent of the top line is potentially affected.

The bank’s view is that Indosat remains a mobile and data business with an AI growth option, not the other way round. There may even be an upside: if rivals struggle to secure approvals for GPUs and customer access, a large telco with an established compliance function could find the field less crowded.

Pure-play operators have no such cushion. Their revenues depend heavily on AI demand and access to advanced GPUs, so a licensing delay or the loss of a Chinese anchor tenant hits utilisation directly, a risk investors were already weighing in the region’s data centres before this week.

Jakarta’s dilemma: stay open, stay compliant

For Indonesia, the rule creates a genuine policy bind. It wants to remain a magnet for data centre capital while staying on the right side of Washington, yet stricter oversight could make it less attractive to Chinese-linked investors. Maybank said the effects could extend beyond operators to industrial estates and infrastructure providers courting the sector.

The country does hold one card. Its surplus power capacity lets developers secure electricity faster than in competing markets, a real edge at a time when power has become the binding constraint across much of the region.

US hyperscaler demand, which sits outside the licensing requirement, is also expected to hold up. That matters for a sector that drew US$1.2 billion in deals across Southeast Asia earlier this year, much of it betting on exactly this regional build-out.

Also Read: Why Southeast Asia cannot build sovereign AI on borrowed choices

The sceptic’s reading is that the rule may concentrate Indonesia’s AI infrastructure boom in fewer, US-aligned hands. Hyperscalers and compliant telcos get a clearer runway; independent operators serving Chinese clients get paperwork. Whether that adds up to a hub or merely a well-behaved outpost is a question Indonesia’s AI ambitions have yet to answer.

Maybank’s conclusion is measured: more scrutiny, not a shutdown. But in a market where speed to capacity is everything, a licence queue of uncertain length is a restriction of its own.

The post After Megaspeed, Indonesia pays the compliance bill for Asia’s AI chip race appeared first on e27.

Posted on — Leave a comment

Johor’s data-centre future is being decided in Ohio and Washington

American voters spent this month’s midterms arguing about data centres: who pays for the power, who loses the water, and who gets a say. Southeast Asia, which has quietly become the overflow tank for the world’s compute, never got that argument. It got press statements.

Worse, the two decisions that will shape Johor’s data-centre future over the next year were not made in Johor, Kuala Lumpur or Singapore. One was made by voters in Ohio and Pennsylvania. The other was made by the US Commerce Department.

Also Read: The true cost of AI is beginning to surface

That should bother anyone who cares about how this region is governed, not just how it is invested in.

What America just voted on

The numbers from the US campaign are striking. Campaigns spent more than US$380 million on TV ads mentioning data centres, according to AdImpact data. An NBC exit poll found almost 80 per cent of respondents opposed new data centres in their neighbourhoods, TechCrunch reported. On election night, Senator Bernie Sanders called for a national moratorium.

The anger had a bill attached. Wholesale electricity prices near some US data centres have climbed as much as 267 per cent in five years, according to a Bloomberg analysis cited by Fortune.

You can debate whether a moratorium is wise. I suspect it isn’t. But the point is that Americans got to debate it, loudly, with their votes. If a new Congress makes data centres harder to build at home, the capex will not vanish. It will look for somewhere quieter.

Southeast Asia has been the quiet place before

We have seen this playbook before. Singapore’s 2019 to 2022 moratorium did not abolish demand; it pushed it across the Causeway. Johor now holds roughly 80 per cent of Malaysia’s operational data-centre capacity, largely because of that pause.

The scale since then is extraordinary. Johor had 1,110MW of live IT capacity in 2026 and 8,542MW in the pipeline, per Knight Frank’s Data Centre Atlas. Malaysia’s energy minister told Parliament in a written reply that data-centre electricity use could rise from 10,544GWh this year to 73,274GWh by 2035, taking it from 7 per cent to 31 per cent of Peninsular Malaysia’s demand.

A 7x jump in a decade is not an industrial policy footnote. It is a change in what the grid is for.

Also Read: SoftBank and Grab put US$1.85B down on a 5GW AI bet in Malaysia

Singapore exported a resource problem to its neighbour, and its neighbour’s residents had no vote on Singapore’s moratorium either. That is the region’s first lesson: in Southeast Asia, data-centre externalities cross borders, but accountability stops at them.

Credit where it’s due

It would be lazy to pretend Malaysia has done nothing. In several ways, it is ahead of the US.

Data centres connected above 50kV have paid an ultra-high-voltage tariff since July 2025, and operators must bear grid-upgrade costs under a user-pays approach. Johor stopped approving less efficient Tier 1 and Tier 2 facilities. Data centres in the state pay RM5.33 (roughly US$1.30) per cubic metre of water, about five times a household’s first band. In February, Prime Minister Anwar Ibrahim moved to halt new non-AI data centres.

The sceptics also have a fair point on water. Johor’s data centres used about 9.07 million litres a day, roughly 0.6 per cent of the state’s consumption. Some viral claims blaming data centres for local outages turned out to be wrong.

So this is not a story of a reckless state. It is a story about how the rules got made.

Policy by announcement is not consent

Look at the sequence. In February, residents of Gelang Patah staged what was described as Malaysia’s first data-centre protest. Residents of Nusa Bayu sent a memorandum to the Menteri Besar’s office asking for more transparency on power and water. When silt cut supply to more than 455,000 Johor accounts in late 2025, a deputy minister said data centres had bought water with their own tankers.

The rules tightened after each flashpoint. That is responsive governance, but it is not participation. Residents learn about a hyperscale campus when the dust arrives.

The data problem makes it worse. Projections for data-centre water demand in Selangor and Johor run to 389 million litres a day by 2030, far above what operators report using today. Both figures can be true, since applications and actual use differ. But when the public cannot tell which number to plan around, distrust fills the gap.

Now Washington holds the pen

Then came 16 November. The US Commerce Department named Singapore, Malaysia, Thailand and Indonesia among countries posing a higher risk of diversion and now requires licences to ship advanced AI chips to them. Data centres anywhere also need a licence to sell large volumes of compute to China-headquartered firms.

Also Read: After Megaspeed, Indonesia pays the compliance bill for Asia’s AI chip race

Singapore responded by seeking “trusted access”. That will be negotiated between officials, as it should be. But it means the most consequential variable in the region’s data-centre economy, which chips arrive and who may rent them, is now set in a US regulatory process. Southeast Asian citizens are two steps removed from it.

Americans voted on whether to host data centres. Southeast Asians are being told, by Americans, on what terms they may host them.

What a ballot looks like without an election

No one needs a referendum on every server hall. But the region can borrow the substance of what US voters demanded without borrowing the gridlock.

Start with disclosure. Publish facility-level power and water use quarterly, not just state totals. Malaysia’s water-usage cap of 1.8 litres per kWh is meaningless to residents who cannot see whether their neighbour meets it.

Make consultation binding, not ceremonial. Projects above a threshold should face a public hearing before approval, with power and water impact studies released in advance.

Share the upside locally. If a campus draws on a district’s river, a share of its tax take should stay in that district.

And coordinate regionally. The Data Centre Association of Malaysia warned last year that higher tariffs could push investment to Vietnam and Thailand. That is exactly the race to the bottom an ASEAN baseline on disclosure and water standards would stop. If Johor, Batam and Chonburi compete on who asks the fewest questions, residents everywhere lose.

The vote we’re not having

Southeast Asia sold itself as the place where AI infrastructure could be built fast, cheaply and without fuss. That pitch worked. But “without fuss” often means “without asking”.

Also Read: Investors do not just fund startups. They fund predictability

The US midterms showed what happens when people who live next to the grid finally get asked: they push back hard, sometimes too hard. This region has a chance to ask them earlier, and more sensibly.

If it doesn’t, the debate will still happen. It will just happen in Washington, and in the streets of Gelang Patah, rather than anywhere it can be settled.

The post Johor’s data-centre future is being decided in Ohio and Washington appeared first on e27.