Author: Ruthc

There’s a fundamental question that most leaders are avoiding right now, not because it’s too hard, but because the answer is uncomfortable.

If you don’t understand where scarcity lives in the current economy, and you keep focusing on supplying more into markets with less demand or that are already overcrowded, you’ve forgotten Economics 101.

Scarcity drives value. That’s not a theory, it’s the foundation. When an item is scarce, it commands a higher price as its value is derived from its rarity. As supply falls short of demand, consumers are willing to pay more to acquire it. When you flood supply into a saturated market, you don’t create value. You erode it. Price compression. Commoditisation. A race to the bottom that nobody wins.

So the question every leader should be asking isn’t what can we build? It’s where is the genuine scarcity?

Technology is becoming abundant. People aren’t

The numbers are striking. The cost of machine intelligence has fallen by roughly 1,000x in just three years, from US$60 per million tokens in 2021 to US$0.06 with Meta’s Llama 3.2. Intelligence, once one of humanity’s most scarce resources, is becoming ubiquitous, abundant, and essentially free.

AI tokens have seen steeper cost reductions than streaming services and legacy technologies combined, driven by advancements in computational efficiency and competitive market forces. Everyone has access. The barriers are collapsing fast. Technology, for all its power, is no longer the scarce resource.

What’s actually scarce right now? According to Deloitte’s 2026 Global Human Capital Trends report, demographic shifts and disappearing workforces are making human capacity itself a scarce resource, elevating the need to invest where humans create unique and irreplaceable value.

Trust. Judgement. Relationships. And yet most leaders are doubling down on the tool, not the gap.

Audit point one: Where are you adding supply, and is there still demand?

Before you ship the next feature, launch the next campaign, or hire for the next role, map it. Is the market you’re entering hungry or saturated?

The data on tech saturation is clear. While AI spending reached US$200 billion in 2024, nearly 65 per cent of organisations report their AI initiatives have fallen short of expectations. And according to Gartner, 30 per cent of generative AI projects will be abandoned after proof of concept by end of 2025, not because the pilots failed technically, but because the human and organisational foundations weren’t built.

Most leaders can answer the technology question cold. Almost none can answer the demand question with real data. If you can’t point to a specific, unmet human need, you’re not leading. You’re just producing.

Also Read: Singapore’s AI dividend will depend on what happens after the pilot phase

The trap most leaders fall into

Tech-first leaders ask: what can this do? People-first leaders ask: what do people actually need?

One is supply-side thinking. The other is strategy.

To truly thrive with AI, leaders must recognise that its power lies in augmenting, not replacing, human connection. The creativity, emotional intelligence, and judgement that only humans can provide will always be in demand.

The leaders who consistently win are the ones who start with where people are underserved, frustrated, or stuck, and then decide whether technology is even the right answer. They’re reading the map before they start driving. Most leaders right now are driving fast, with no map, and calling it innovation.

Audit point two: What would break if you removed the tech?

This is the revealing question. Strip out the AI, the platform, the automation. Does the value disappear, or does it hold?

If the answer is it disappears, you don’t have a leadership strategy. You have a dependency.

Real value lives in the human layer. Trust is the foundation of most successful organisations. In high-trust organisations, employees feel safe to take risks, express themselves freely, and innovate. In contrast, employees at low-trust organisations are often bogged down by office politics and infighting, more likely to withhold information and hoard resources because they don’t feel safe sharing them.

If your entire proposition collapses the moment the tool goes away, you haven’t built anything. You’ve borrowed someone else’s infrastructure and called it a business.

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

Direction over efficiency

There’s a version of tech adoption that looks like progress but isn’t. Only 48 per cent of AI projects make it into production, according to Gartner research, meaning more than half of all AI initiatives die somewhere between proof-of-concept and actual deployment. The research is consistent: this isn’t a technology problem. It’s a direction problem.

Efficiency without direction is just faster mediocrity. You can automate your way through an entire quarter and end up further from where you needed to be, just with better dashboards to prove it.

Audit point three: Are you solving for scarcity or comfort?

Most tech adoption happens because it feels easier, not because it closes a genuine gap. The proof is in the trust data: globally, 68 per cent of respondents say they are worried that business leaders purposely mislead people, a 12-point increase from 2021. Most leaders are starting with a trust deficit and will have to earn trust with consistent communication and transparency.

Ask yourself honestly: am I using this tool because it moves me toward something scarce and valuable, or because it feels like progress?

People worldwide seek hope and trust as foundational leadership traits. Understanding these needs is what helps leaders build relationships and inspire others through changes in the world of work. The leaders who can anchor to that, and act on it, are the ones building where others aren’t looking.

The audit

Run these three questions quarterly. Not annually. Not when things slow down. Quarterly.

Where are you adding supply, and is there still demand? What would break if you removed the tech? Are you solving for scarcity or comfort?

Demand. Dependency. Direction.

That’s the audit. And right now, most leaders are failing it, not because they lack intelligence, but because they’re leading with the tool instead of leading with the people the tool is supposed to serve.

The map comes first. Always.

—

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 Are you leading with people? Or are you leading with tech? appeared first on e27.

DePIN (Decentralised Physical Infrastructure Networks) crowdsources real-world resources like compute, storage, and connectivity via blockchain incentives. RWA tokenisation brings tangible assets (real estate, commodities, infrastructure) on-chain for fractional ownership, liquidity, and global access. Together, they address SEA’s challenges: high infrastructure costs, limited banking reach, and volatility in local currencies.

Global tokenised RWA (excluding stablecoins) surpassed US$36 billion by late 2025, with forecasts reaching trillions by 2030. DePIN adds real utility: devices generate verifiable data or yield that becomes tokenised RWA.

Global alliances driving convergence

Western alliances set benchmarks. The Superintelligence Alliance (ASI, merging Fetch.ai, SingularityNET, Ocean Protocol) focuses on decentralised AI but connects to RWA through partnerships like SuperWorld. This integrates ASI’s AI stack with virtual real estate, enabling monetisation of physical-digital interactions via tokenised assets, DePIN hardware, and RWA rewards, democratising access to infrastructure yields.

These alliances standardise protocols, attract capital, and reduce fragmentation, turning AI-DePIN data into investable RWA.

SEA’s regulatory momentum and emerging alliances

Vietnam’s 2026 framework marks a shift. The Law on Digital Technology Industry (effective January 1, 2026) recognises digital assets, backed by real-world underlying (excluding securities/fiat). Resolution 05/2025/NQ-CP launches a five-year pilot for issuance/trading, emphasising compliance, cybersecurity, and tokenised assets for practical use.

This creates opportunity for RWA in infrastructure, real estate, and green assets. SEA builders leverage it with alliances tailored to local needs: mobile onboarding, affordable hardware, cross-border utility.

Also Read: Bitcoin dominance at 58.5% and the 55% line that still blocks altseason

Examples include Depin Alliance (depinalliance.xyz), uniting DePIN builders for standards and adoption. It hosts events like the Yacht Party in Ho Chi Minh City, a luxurious networking gathering with global VCs, DePIN leaders (IoTeX, Aethir, Witness Chain), and founders, fostering collaborations on RWA convergence.

Insights from leading alliances in Asia

I was previously CMO at U2U Network, a DePIN-focused Layer-1, which was involved in organising Depin Alliance conferences and side events across Asia. From that position, onboarding and fragmentation were the problems that came up most often when builders, investors and regulators met at these events. Alliances in this space tend to be judged on whether they reduce those two frictions, so it is worth asking what they have actually changed for members.

Niche approaches like “opening association as a service” (templates, events, support for sector alliances) help scale fast. In SEA, where RWA attention grows (tokenised commodities, private credit), alliances capture momentum by linking physical networks to liquid markets.

Challenges and path forward

Fragmentation across chains causes pricing gaps. Hardware costs and volatility deter contributors. Regulation varies (Vietnam pilots cautiously, Indonesia securities oversight) but clarity boosts confidence.

In 2026, alliances will mature DePIN-RWA. Global ones provide standards; SEA ones localise for inclusion. The convergence promises practical benefits: tokenised infrastructure shares, verifiable yields from physical devices. Builders who connect communities early will lead the wave.

—

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 DePIN and RWA alliances in 2026: Global momentum meets SEA opportunities appeared first on e27.

Jenni is an academic research and writing platform used by more than six million researchers, who have written over 15 million papers on it. The US-headquartered company says it passed US$10 million in annual recurring revenue (ARR) this year and is profitable, having grown almost entirely from revenue after raising only a small angel round.

While ChatGPT remains the first port of call for many researchers, Jenni has deliberately stayed narrow, focusing specifically on academic work. Its citations link to real papers rather than invented ones, users can trace every AI-generated claim back to the source documents behind it, and its AI Declaration feature, built for researchers submitting to publishers, lets them disclose exactly how AI was used in a paper.

Also Read: India is showing the world what AI will do to jobs and education

The platform can also work directly from a researcher’s own uploaded sources, which is crucial for local or non-English research, where general AI tools still default heavily to English-language results.

In Asia, Jenni has growing university partnerships and user bases across Singapore, Malaysia, India, Thailand and the Philippines, with China emerging as its most ambitious, and most complex, expansion target.

Justin Wong, Head of Commercial Operations, helped take Jenni from US$1 million to US$10 million ARR and now leads its B2B sales and international expansion. He spoke to e27 about the decisions that shaped that journey.

A bet on careers, not semesters

For Wong, there was no single dramatic inflection point in Jenni’s growth from US$1 million to US$10 million ARR. The change was a strategic repositioning, away from general writers and students and towards higher education and academic researchers.

The logic was straightforward. A student might use a writing tool for a single semester before moving on; a researcher writes papers across an entire career. Once Jenni reoriented itself around that longer relationship, building a product a researcher could rely on “from their first manuscript to their fiftieth”, retention, willingness to pay and word-of-mouth referrals inside academic labs and departments all improved.

That shift also reshaped operations. Once Jenni crossed the US$10 million threshold, the company went international“with intent”, localising for markets across Asia, Latin America, the Middle East and Europe. At the same time it built an institutional sales motion alongside its self-serve consumer business, largely because faculty were already bringing Jenni into their departments on their own.

Also Read: Why Dropbox refuses to pick a side in the ChatGPT-Claude-Gemini fight

The bootstrapped path wasn’t originally a strategic choice; it became one. Jenni was one of the earlier players in AI-assisted academic writing, and revenue arrived early enough that raising a large venture round never felt necessary. That came with trade-offs. Every acquisition channel had to pay for itself, the team stayed lean with employees wearing multiple hats, and growth was never bought beyond what the business could sustain. The upside, Wong argues, was strategic independence. Without investor pressure to chase whichever segment was commercially hot, Jenni could commit fully to serving researchers over the long term.

Competing with the tab already open

Every researcher already has ChatGPT open somewhere, and Wong is candid that most conversations with universities start from that reality. The pitch isn’t that general models are useless. It’s that they invent citations, produce unsupported claims and generate prose with no identifiable source.

Jenni’s workflow is built to close those gaps:

• AskJenni lets researchers interrogate both their own uploaded PDFs and more than 250 million external papers.
• Autocomplete suggests one sentence at a time rather than generating full passages, keeping the researcher in control.
• Citations link to real sources across more than 10,000 citation styles.
• A four-layer review checks peer-review readiness, claim confidence, proofreading and tone before submission.

Jenni also does not train on user data in any feature.

Institutionally, Jenni avoids asking for a campus-wide commitment upfront. Instead, a department or lab runs a one-to-two-month pilot, tracking hours saved and manuscript activity, and the deployment expands based on the evidence. India-based Lovely Professional University’s pilot with more than 60 faculty members reported over 375 hours saved; Chulalongkorn University’s six-department rollout reached more than 300 users and over 750 hours saved.

Also Read: The localisation gap: Why multilingual AI isn’t enough for APAC markets

Asked what actually closes institutional deals, Wong points to Jenni’s Reviews feature. It scans a full paper for contradicted, unverified or misrepresented claims, whether a human or AI wrote them, and suggests peer-reviewed references to support them.

He, however, is careful not to reduce Jenni to a citation tool. Many users describe it, he says, as “a combination of ChatGPT, Microsoft Word, and Zotero in a single tab”: one workspace for organising sources, drafting and reviewing. That breadth, not any single feature, is what ultimately sells.

Publishers, not a publisher partnership

Wong is quick to correct a common misreading. Jenni didn’t build a feature with publishers. It built its AI Declaration feature for researchers submitting to publishers and journals, as a simple way to disclose exactly how AI was used in a paper, whether for citations, drafting or grammar correction. The declaration has now appeared in over 100 published papers, including work published by Nature, Elsevier, IEEE and MDPI. As journals formalise their AI disclosure policies, Wong sees this as part of making AI-assisted research broadly acceptable rather than suspect.

Localisation, and the China exception

Across Southeast Asia and India, Jenni’s self-serve-then-pilot motion travels well. China is the outlier. Traditional marketing and sales channels don’t work there. Researchers live on Xiaohongshu, WeChat and Douyin, behind entirely different infrastructure, which makes China the most heavily localised market Jenni has entered.

The payoff, Wong argues, justifies the effort: China publishes more research papers than any other country, and many of its universities lead global research output.

Language matters beyond China, too. Because Jenni grounds its AI in a researcher’s own uploaded sources, including local studies, theses and reports a general model wouldn’t surface, it can support hyper-local work in ways general chatbots cannot, since those often default to English-language searches. Jenni supports drafting in more than 30 languages.

This localisation sits alongside two sales cycles that feed each other. Individual researchers buy through free, monthly and annual plans; this is the self-serve business that passed US$10 million ARR in 2026. Institutions follow the slower pilot-then-expand model. Self-serve adoption is often how Jenni gets into an institution in the first place, because faculty and students are frequently already using it before a university formally engages. As universities increasingly centralise AI purchasing decisions, that dynamic turns individual subscriptions into campus-wide deployments.

The moat against bigger players

What stops a well-funded general AI company from simply adding academic features? Wong frames the defence around incentives rather than functionality. General tools optimise for fast, fluent answers sourced from anywhere on the internet, including unverifiable sources. Academic work demands citations a researcher can defend in front of reviewers. Jenni’s workflow is built around accountability, not just output speed: autocomplete, verifiable citations, four-layer review, no training on user data, and AI declarations. “Adding a citation button to a chatbot,” Wong says, “doesn’t replace that workflow.”

Also Read: The EU called ChatGPT a search engine. SEA’s AI startups should worry about what comes next

With 15 million papers written and more than six million researchers on the platform, Wong believes Jenni’s next phase lies in depth rather than breadth: owning more of the research lifecycle rather than chasing new user segments. “If we’re the tool a researcher trusts for their whole career,” he says, “the new segments follow.”

The post Why Jenni thinks researchers need more than ChatGPT for academic writing appeared first on e27.

The post AI labs are chasing a slice of the corporate pie in Southeast Asia appeared first on e27.

Across 231 Canadian publishers, the outlets that accept restricted categories charge less than half as much for ordinary articles. That discount measures who has already left the room, and founders in crypto, lending and gaming are shopping in that half without knowing it.

The cheapest quote in a media plan is usually the one worth worrying about. Founders in regulated categories learn the price side of that quickly, because the numbers arrive in an email. The access side takes a year, and by then the budget is spent.

This month ESBO Ltd, the link building and digital PR agency I run, published Canadian link building prices in 2026, built from 231 Canadian publishers filtered to a Domain Rating of 30 and above with at least 2,000 monthly organic visits. Canada is a test case here rather than a market most readers buy in. It legalised two major categories and then restricted how they may be promoted, which makes the effect on media access easy to measure. The median publisher quotes US$810 for a sponsored article. The interesting part is who quotes it.

Two markets wearing one label

Of those 231 publishers, 49 per cent quote a price for at least one restricted category. The remaining 51 per cent quote none at all.

Those two halves also price ordinary, unrestricted articles differently. Publishers that accept something quote a median US$610. Publishers that accept nothing quote US$1,110. Same product, 1.82 times the price, and the gap holds across every category in the file.

So a founder in a restricted category is not getting a discount on the market. They are shopping in a different, cheaper market, and the cheapness is the signal.

Count readers, not publishers

Site counts make the restricted market look workable. Roughly 41 per cent of Canadian publishers accept gambling content.

Now count audience instead. Add up the monthly Canadian visits behind the publishers that accept each category, as a share of all Canadian audience in the file, and gambling reaches 24 per cent of it. Cannabis reaches 34 per cent. Adult reaches 22 per cent. Anything restricted at all reaches 48 per cent.

Also Read: The most expensive links aren’t really links: What 16,625 publisher price lists tell SEA startups

That is the number to plan against. The cheap gambling placement at a median US$355 is not a bargain. It is the price of what remains after three quarters of the audience has removed itself from your list.

The line is compliance, not taste

Compare the categories and the pattern is not about respectability.

Gambling and cannabis produce the widest gaps between accepters and refusers, at 4.27 and 4.11 times. Both are legal in Canada and both carry strict rules on how they may be promoted: the Cannabis Act prohibited promotion capable of appealing to young people when it took effect in 2018, and Ontario’s regulator has since barred athletes from iGaming advertising altogether. Crypto and forex, which carry no comparable promotion rules, produce the narrowest gap at 1.51 times.

The dividing line is how much compliance exposure a publisher takes on by carrying you. A publication with written advertising standards, a disclosure policy and a named person who approves advertisers excludes legally restricted categories as a class, because each one creates obligations somebody has to own. A publication with none of that takes everything.

Disclosure behaviour confirms it from the other side. Stated sponsorship labelling runs from 11 per cent of publishers at the bottom of the authority range to 64 per cent at the top, and publishers that label charge US$1,420 against US$405 for those that do not. Formal terms, disclosure and a closed door to restricted niches travel together, and they travel with a higher price.

Why this matters in Southeast Asia

A category does not need to be illegal to lose its media access. It needs rules that make carrying it a liability.

That mechanism is already here. In January 2022 MAS issued guidelines stating that digital payment token providers should not promote their services to the general public in Singapore, including through third-party websites and social media influencers, leaving them their own sites, apps and official accounts. Whatever you think of the policy, look at what it does to a marketing plan: the third-party channels come off the table, and the startup is left talking to people who already found it.

Also Read: Startups keep scaling ops before they scale data — Here’s why it backfires

Given how much of this region’s startup activity sits in digital assets, lending, remittances and gaming, a large number of founders here are operating in the restricted half of their media market and have priced it as a bargain. This sharpens the argument I made in this column in June about needing public proof before you scale. When the channels that would carry your proof are closed to your category, the ones that remain have to be earned rather than bought.

Four adjustments

Count audience reach, not publisher counts, before approving any plan in a restricted category.

Stop negotiating the cheap half. Publishers that accept everything quote US$260 to US$390 for standard content. They are already at the floor, and there is nothing to win there.

Put the effort into the publishers who currently refuse. That is an editorial conversation about what the content says, which claims it makes and how it is labelled, not a rate conversation.

Read disclosure as quality rather than cost. The publishers that label charge three and a half times more, reach more readers, and have somebody inside whose job is to vet advertisers.

The instinct when a category gets expensive is to shop for a better price. In regulated categories, price was never the constraint. Access was.

—

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 Half the market will not take your money: The access problem for regulated startups appeared first on e27.

Sarawak is not usually the first place that comes to mind when Southeast Asia’s artificial intelligence race is discussed. Singapore has the region’s densest cloud and startup ecosystem, Malaysia’s Johor has been drawing data centre investment because of its proximity to Singapore, and Indonesia has been positioning itself around scale, talent and natural resources.

A new agreement involving SoftBank Group, Grab Holdings and Petroleum Sarawak, better known as PETROS, suggests Sarawak wants a more central role in that conversation.

Also Read: Malaysia’s sovereign AI bet: Local context becomes the next startup moat

The Japan-based technology investment group, Singapore-headquartered superapp operator and Sarawak’s state-owned oil and gas company have signed a framework agreement to explore the development of an AI and digital infrastructure platform in the Malaysian state. The agreement was signed by SoftBank Chairman and CEO Masayoshi Son, Grab Group Chairman, CEO and Co-Founder Anthony Tan, and PETROS Group CEO Janin Girie.

The deal is still at an exploratory stage. The three parties will work on a roadmap for the phased development of the platform, rather than immediately committing to a fully defined project. But the areas being studied are broad: AI compute and services, advanced technologies, robotics, local workforce development, community engagement and wider digital infrastructure.

If it progresses, the collaboration could place Sarawak in the middle of one of Southeast Asia’s most competitive technology infrastructure markets: the race to provide the power, computing capacity, connectivity and applied services needed for AI adoption.

Why Sarawak matters

AI infrastructure is not just about chips and servers. It depends heavily on land, energy, water, connectivity, regulation and a long-term pipeline of enterprise users. That is why governments across Southeast Asia have begun treating digital infrastructure as industrial policy, not just a technology sector issue.

Sarawak brings different advantages from Malaysia’s more established digital corridors. The state has significant energy resources and has been seeking to move up the value chain beyond extractive industries. PETROS is central to that ambition through the Sarawak Gas Roadmap and its role as master developer of the Kuching Low-Carbon Hub.

Under the agreement, PETROS is expected to focus on energy supply solutions, enabling infrastructure, local engagement and participation from Sarawak’s supply chain. That role is important because AI compute is energy-hungry. Data centres and high-performance computing facilities require stable, scalable and increasingly lower-carbon power sources, especially as multinational technology companies face pressure to manage emissions linked to digital growth.

Sarawak’s Post COVID-19 Development Strategy 2030 aims to lift the state to high-income status by 2030. An AI infrastructure platform, if executed well, could support that ambition by attracting higher-value investment and creating demand for engineering, cloud, operations, cybersecurity and AI-related jobs.

The challenge will be making sure the project does not become another infrastructure story where most of the value is captured by outside vendors, while the host location is left mainly with land use, power demand and limited spillover into the local economy.

SoftBank’s AI ambitions meet Southeast Asia’s infrastructure needs

SoftBank’s role in the partnership is straightforward: it brings its global technology and AI ecosystem. The group has exposure across AI infrastructure, advanced computing, semiconductors and what it calls physical AI, a term often used to describe AI systems applied to robotics, mobility, industrial automation and real-world machines.

Also Read: Acrab’s US$130M raise signals SEA’s deeper push into AI hardware

Son has been one of the most aggressive global voices on the transformative potential of AI. SoftBank’s portfolio and strategic interests span chip designer Arm, AI companies, robotics and digital platforms. Its involvement gives the Sarawak initiative a degree of international visibility, even if the agreement remains at the roadmap stage.

For Southeast Asia, such partnerships are increasingly important. AI adoption in the region is rising, but many markets still depend heavily on cloud infrastructure hosted elsewhere. Local compute capacity matters for latency, data governance, enterprise adoption and national digital resilience. Governments also see AI infrastructure as a way to capture more of the economic value created by digital platforms and automation.

Malaysia has been particularly active. The country has attracted large data centre and cloud commitments in recent years, helped by demand from Singapore and by its own digital economy goals. Johor has become a prominent data centre location, while Kuala Lumpur remains the country’s main corporate and technology hub. Sarawak’s pitch appears to be different: energy-linked industrial development, lower-carbon infrastructure and a broader state transformation agenda.

Grab’s role: demand, data and applied AI

Grab’s inclusion makes the agreement more than a hard infrastructure play. The company brings a large Southeast Asian digital ecosystem across mobility, deliveries, financial services, merchants and consumers. It also has experience deploying AI in areas such as matching drivers and passengers, fraud detection, routing, recommendations, merchant tools and customer service.

That applied layer matters because AI infrastructure is valuable only if there are users and use cases. A compute platform without enterprise demand risks becoming underutilised capacity. Grab could help shape commercial use cases around logistics, urban services, small business digitisation, mapping, payments and customer engagement across Southeast Asia.

Also Read: SEA’s AI boom has a water problem it cannot offset away

For Sarawak, Grab’s presence also creates a link to the region’s consumer internet economy. Many AI infrastructure discussions can feel abstract, focused on chips, data centres and national strategies. Grab’s business is closer to daily economic activity: rides, food delivery, digital payments and small merchants. If the partnership extends into local workforce and community development, Grab’s networks may help bring AI tools to smaller businesses rather than just large enterprises.

At the same time, Grab’s participation should be read carefully. The announcement does not say Grab is building a data centre, committing capital expenditure, or relocating AI operations to Sarawak. Its stated contribution is its digital ecosystem, applied AI capabilities, and customer and partner networks. The practical shape of that contribution will depend on the roadmap the parties develop.

Part of Malaysia’s broader digital agenda

The framework agreement is positioned as supporting Malaysia’s digital infrastructure and connectivity push, including the Jalinan Digital Negara initiative, or JENDELA. It is also linked to national priorities under the Ekonomi MADANI framework and the Thirteenth Malaysia Plan, both of which emphasise productivity, technology, investment and higher-value economic activity.

That alignment is not incidental. Large AI infrastructure projects typically require coordination across federal agencies, state governments, utilities, land authorities, regulators and private sector operators. By tying the collaboration to national and state development plans, the parties are signalling that the project is intended to fit into a broader policy agenda rather than stand alone as a private commercial venture.

For Southeast Asia, the announcement reflects a wider pattern. Countries are trying to move from being markets for digital services to becoming infrastructure nodes for the AI economy. Singapore has talent and capital but faces land and energy constraints. Malaysia has more space and power options, but must balance rapid data centre growth with grid capacity and sustainability. Indonesia, Thailand, Vietnam and the Philippines are also competing for investment in cloud, AI and digital services.

Sarawak’s opportunity lies in finding a specific role within that regional map. It is unlikely to replace Singapore as a headquarters hub or Johor as a spillover location for near-Singapore data centre demand. But it may be able to position itself around energy-backed AI infrastructure, industrial applications and a development model tied to local participation.

Also Read: Why Malaysia’s AI Nation 2030 plan matters for B2B startups

For now, the SoftBank-Grab-PETROS agreement is a starting point, not a finished blueprint. Its importance lies in what it reveals about the next phase of Southeast Asia’s AI buildout. The region’s AI race will not be won only by companies with the best models or apps. It will also be shaped by states and cities that can provide the physical foundations — power, compute, connectivity and talent — on which those models run.

If Sarawak can turn this framework into execution, it could shift from the edge of the regional tech map to a more strategic position in Malaysia’s AI infrastructure ambitions.

The post SoftBank, Grab, PETROS explore AI infra platform in Sarawak appeared first on e27.

For years, online retail has been organised around a simple assumption: the customer journey begins on a search engine, marketplace, brand website, or app. That assumption is starting to look dated.

A new Salesforce report suggests that more shoppers are now beginning with a question to an AI system, whether through ChatGPT-style assistants built on large language models, an AI tool embedded on a retailer’s website, or a conversational interface inside another platform.

According to the fourth edition of Salesforce’s State of Commerce report, the use of agentic search as the first step in the shopping journey grew 200 per cent year on year.

The findings are based on a survey of 3,450 commerce professionals across 20 countries and 13 industries, including 100 respondents from Singapore. Salesforce also drew on a consumer survey of 4,689 shoppers and behavioural data from more than 1.5 billion shoppers across 37 countries.

Also Read: Pinterest and Shopee link up to bring creator-led shopping to Indonesia

For Southeast Asian retailers, the shift is not just another channel to add to the marketing mix. It raises a more fundamental question: if customers no longer start on a brand’s own website, how does a business make sure its products, prices, inventory and brand information are accurately represented wherever discovery happens?

Discovery is moving away from owned channels

Salesforce’s behavioural data points to a sharp change in how shoppers find products. Traffic referred from AI chats grew between 150 per cent and 428 per cent year on year in every quarter measured, while overall traffic grew only in the single to low double digits.

Consumer behaviour appears to be moving in the same direction. Between August 2025 and May 2026, the rate of shoppers discovering products through brand-owned properties fell 7 per cent, while traditional search fell 15 per cent. Over the same period, the rate of consumers choosing newer discovery channels, including AI assistants, social media AI and delivery apps, grew 38 per cent.

This matters in Southeast Asia because commerce is already highly fragmented. A shopper in Singapore, Jakarta, Manila or Bangkok may move between a marketplace, TikTok, WhatsApp, a delivery app, a physical store and a brand’s website before buying. AI adds another layer to that path. It can compress the discovery process into a single conversation, but it also means the first recommendation may happen outside the retailer’s direct control.

Singapore commerce leaders appear to understand the direction of travel. Eighty-two per cent of respondents in the city-state said large language models will be essential to product discovery within the next year. Many are already adjusting their playbooks: 42 per cent are improving product content quality, 40 per cent are optimising content for conversational queries, 39 per cent are submitting data feeds to AI search platforms, and 38 per cent are rewriting product descriptions for natural language.

In plain terms, retailers are trying to make their catalogues more understandable to machines. A conventional search page may match keywords. An AI assistant tries to interpret intent: “What should I buy for a humid climate?”, “Which running shoes work for flat feet?” or “What is a good gift under S$100?” If product data is incomplete, inconsistent or badly structured, the brand may be invisible in these conversations.

Adoption is still early, but the pressure is rising

Despite the momentum, business adoption remains uneven. Only 28 per cent of Singapore organisations surveyed said they currently use agentic AI. Of those that do not, 52 per cent plan to deploy it within the next six months.

That gap between consumer behaviour and corporate readiness is where the pressure is building. Eighty-six per cent of commerce leaders said AI is raising customer expectations, while 41 per cent said meeting those expectations is harder than ever. Implementing or expanding AI has become both their top priority for the year ahead and their top anticipated challenge.

The term “agentic AI” refers to AI systems that can take actions toward a goal, rather than simply generate text or answer questions. In commerce, that could mean helping a shopper compare products, checking availability, recommending bundles, handling returns or routing a service request. Fully autonomous purchasing remains early, but the influence of AI at the discovery and decision stage is growing quickly.

Also Read: Southeast Asia’s live commerce boom enters its harder second act

In Asia Pacific, companies that have already adopted agentic AI appear to be moving beyond pilots. Only 5 per cent of APAC adopters said they are still primarily testing use cases, while 35 per cent, the largest share, said they are scaling AI across functions and teams, from service and IT to merchandising.

Adopters reported gains in customer satisfaction, personalisation, operational efficiency and employee productivity. Those benefits are attractive in a region where retail margins can be thin and customer acquisition costs have risen across marketplaces and social platforms.

The data problem behind the AI promise

The harder part is that AI does not fix messy systems by magic. In many cases, it exposes them.

Salesforce’s Singapore findings show a commerce stack under strain. Eighty-one per cent of commerce leaders said their vendor count had grown over the past two years. Only 25 per cent said their customer data is fully unified across sales, service, marketing and commerce.

The consequences are practical. Among organisations with fragmented data, 36 per cent reported slow or ineffective responses to customer issues, 32 per cent struggled to measure the impact of commerce investments, and 40 per cent said they faced high costs maintaining disconnected systems.

Omnichannel operations remain a weak point. Just 2 per cent of Singapore multichannel organisations reported no significant failure points. The most common problems were inventory not being synchronised in real time, cited by 46 per cent, and inconsistent pricing and promotions, cited by 36 per cent.

These issues are not new, but AI makes them more visible. If an AI assistant recommends a product that is out of stock, quotes the wrong price, or gives a customer a different answer from a store associate, the experience breaks. In markets such as Southeast Asia, where consumers often compare across channels before buying, these inconsistencies can quickly lead to abandoned carts or lost trust.

Physical retail is also being pulled into the same digital loop. Salesforce found that stores remain the top holiday shopping destination for 77 per cent of consumers, but the in-store journey is increasingly online. Seventy-nine per cent of shoppers use their phones while shopping in store, and 12 per cent ask an AI assistant for purchasing advice in the aisle. Meanwhile, 86 per cent of B2C respondents said customers expect the same personalisation in store as online.

Why unified data becomes the battleground

The report suggests that companies that have made progress on data unification are already seeing returns. The most commonly cited benefits include better alignment between sales, marketing and commerce teams at 44 per cent, improved customer retention and loyalty at 42 per cent, and better AI and automation outcomes at 31 per cent.

That makes data infrastructure less of a back-office concern and more of a competitive issue. As AI becomes a new gateway to shopping, retailers will need to ensure that product information, customer context, promotions, inventory and service histories can travel across channels.

Also Read: OpenAI’s Astra aims to turn AI from chatbot into digital worker

“Customers are no longer starting their search on a brand’s website or a search bar, they’re starting it in an AI chat, a social feed, or a delivery app,” said Swatantra Kumar, Regional Vice President at Salesforce. “Companies selling online and offline need to show up wherever discovery is happening — and that only works if their data is unified enough for AI to represent their products and brand accurately.”

For Southeast Asia’s commerce players, the near-term question is not whether AI will matter. It already does. The harder question is whether their systems are ready for a world where the first shopfront a customer sees may not be a shopfront at all, but an answer generated by an AI assistant.

The post AI chats are becoming the new starting point for online shopping appeared first on e27.

The crypto market is reeling. Bitcoin trades around US$80,550 to US$81,000. Ethereum trades near US$2,410 to US$2,470. Both assets dropped sharply on October 9, 2026. An aggressive deleveraging wave and broader macro risk-off sentiment drove the decline. Bitcoin fell about 3.3 per cent to 4 per cent over the last 24 hours. It dropped from weekly highs near US$86,500. Ethereum fell about 5.7 per cent to 6 per cent over the same period. It retreated from weekly levels above US$2,700. The total crypto market capitalisation fell 2.53 per cent to US$2.76T in 24 hours.

A massive liquidation cascade was the primary driver of this move. The crypto market shows a 75 per cent correlation with the Nasdaq-100 (QQQ). That correlation signals a shared macro-driven move. This sell-off is not isolated. It reflects a broader risk-off mood across financial markets. Investors have shown little appetite for speculative positions in this environment. The crypto sector is feeling the same pressure that weighs on technology stocks and other risk assets.

The primary reason for this sell-off is a concentrated wave of long liquidations. Bitcoin led the forced closures. More than US$300M in leveraged positions disappeared. The drop below US$77,000 triggered a concentrated liquidation wave. Over 24 hours, US$303.08M in BTC long positions were forcibly closed. That figure marks a 62.97 per cent surge from the prior period. This forced selling in the largest asset spilled over and pressured the entire market.

Excessive leverage acted as fuel. It turned a routine pullback into an accelerated decline. High volume spiked 15.02 per cent overall. Derivatives volume spiked 65.95 per cent. These volume spikes confirm panic selling and position unwinding. The cascade did not spare Ethereum. The second-largest asset suffered its own wave of liquidations. The entire crypto ecosystem felt the aftershock as traders rushed to reduce exposure.

Also Read: Asian investors aren’t choosing between crypto and TradFi anymore

Traders should watch for a sustained drop in open interest and funding rates. Such a drop would signal that deleveraging is cooling. Until that happens, prices remain vulnerable to further bouts of forced selling. The current environment rewards caution. Excessive leverage built up in the system over recent weeks. That leverage created a fragile structure. When prices dipped below key thresholds, the structure collapsed.

The high volume confirms that many participants exited positions in a hurry. Panic selling often marks a short-term bottom. It can also precede further declines if support levels fail. The next few sessions will reveal whether the deleveraging has run its course. A calm derivatives market would provide the first real sign of stability. Until then, every rally attempt risks another liquidation cascade.

Secondary pressures came from sector-specific worries. Ethereum fell 8.62 per cent over 7 days. Fears that artificial intelligence could break elliptic-curve cryptography partly drove that drop. Coinbase’s cryptographer debated this issue. The debate sparked concern among holders. At the same time, overheated altcoins saw sharp profit-taking.

NEAR and ZEC corrected after extended rallies. NEAR fell 15.79 per cent. ZEC fell 10.39 per cent. Those declines followed massive 30d rallies. NEAR had gained 91.61 per cent. ZEC had gained 567.13 per cent. The sell-off was not monolithic. It combined technical profit-taking in overbought altcoins with nascent, fear-driven narratives about core blockchain security. These narratives added a new layer of uncertainty to an already fragile ecosystem.

The AI cryptography discussion matters for the long term. In the short term, it gave traders another reason to sell. The debate itself is healthy. Panic selling based on it is not. Ethereum’s 8.62 per cent 7d decline outpaced Bitcoin’s 24-hour drop, which shows altcoins carry more beta in this risk-off phase.

Also Read: Did the Fed just accidentally kick off the next crypto bull run? Or just a dead cat bounce?

The near-term outlook hinges on key technical levels. The market is testing a key Fibonacci 50 per cent retracement level at US$2.76T. The immediate path depends on whether this support holds. The next critical level is the 61.8 per cent retracement at US$2.71T. Upcoming macro cues, such as the next FOMC decision, will influence broader risk appetite. A hold above US$2.71T could lead to a period of consolidation and base-building.

A break below that level would likely trigger another leg down toward the swing low of US$2.57T. This risk is especially high if leveraged long positions rebuild too quickly and become vulnerable again. The asset class needs time to heal. It also needs a calmer macro backdrop. The correlation with the Nasdaq-100 means crypto will struggle to rally if technology stocks remain under pressure. The US$2.76T level matters as the first test. The US$2.71T level matters as the last line of defense before US$2.57T.

In my view, this downturn is a violent deleveraging event. It is not a fundamental rejection of the asset class. Excessive leverage built up across BTC and altcoins. The liquidation cascade provided a brutal reset. The AI cryptography fears, while legitimate in the long run, have amplified short-term panic. The core technology behind Bitcoin and Ethereum remains sound. The cryptographic community continues to work on quantum-resistant solutions.

The real risk is not that AI will break elliptic-curve cryptography tomorrow. The real risk is that fear over such a possibility could drive investors away from the asset class. The macro environment is the more pressing concern.

As long as the Federal Reserve maintains a hawkish stance and geopolitical tensions simmer, crypto will struggle to mount a sustained recovery. The US$2.71T level is the line in the sand. If it holds, we could see stabilisation in the coming weeks. If it breaks, prices may need to retest their July lows. The pain could extend well into the final quarter of the year.

Market Outlook: Bearish Pressure. The downturn was primarily a violent deleveraging event. Sector-specific anxieties and profit-taking amplified it. The ability to stabilise now depends on holding key technical supports and avoiding another buildup of speculative long leverage. Will the US$2.71T level hold, or is a retest of the July lows ahead? The answer will shape the direction of the crypto sector for weeks to come. For now, caution remains the order of the day.

—

The post From US$86,500 to US$81,000: The anatomy of a crypto deleveraging wave appeared first on e27.

Asia’s AI founders do not lack investor attention. What they lack is the expensive plumbing frontier AI demands: models, compute, technical help and patient capital. Granite Asia and Google AI Futures Fund want to bundle all of it into one programme.

The Singapore-headquartered VC firm, which manages around US$11 billion in assets and co-managed capital, will co-invest up to US$2 million per startup alongside Google, through the Granite Asia Fellows Program. Selected founders also get up to US$350,000 in Gemini and Google Cloud credits, early access to models such as Gemini, Nano Banana and Lyria, and direct support from Google DeepMind researchers and engineers.

The programme targets five areas: science and knowledge, industrial transformation, work, creativity and entertainment. Granite Asia promises founders full choice over their AI stack, a notable pledge when most startups fear lock-in to a single provider.

The bigger question is whether platform deals like this build durable companies or merely subsidise early infrastructure bills, as Microsoft, AWS, Nvidia and Google all court the same founders with credits and model access. Cloud credits are not revenue. The real test comes later: paying customers, defensible products and proof that a startup is more than a clever demo.

—

INTERVIEWS AND FEATURES

DayOne’s Nasdaq filing shows who is capturing SEA’s AI value: The data centre operator’s revenue tripled to US$512M in H1 2026, yet one customer supplies 69% of sales. The region provides land, water and power while most of the AI equity upside leaves.

Jenni hits US$10M ARR selling researchers more than a chatbot: The US-based academic writing platform has six million users who have written over 15 million papers. It turned profitable after raising only a small angel round, and makes its case against general-purpose tools.

SEA engineering teams embrace AI agents, but keep them off production: Developers from Singapore to Ho Chi Minh City now use agents to plan tasks, refactor repositories and generate tests. Tech leaders still stop short of production access, where the cost of a mistake is highest.

SEA insurers must prepare for customers who send AI agents instead: A Google Cloud and CoverGo whitepaper argues early generative AI wins, from claims extraction to call-centre drafts, may leave insurers unready for a shift to customer-side agents.

When carers have dementia too, Japan turns to physical AI: The feature unpacks rōrō kaigo and ninnin kaigo, where frail or dementia-affected elderly people care for one another, and why Japan is relying on physical AI to fill gaps in eldercare.

Southeast Asia’s foodtech winners are the least glamorous players: Lab-grown shrimp and 15-minute grocery delivery grabbed headlines, but the 2022 correction, eFishery’s scandal and TaniHub’s collapse thinned the herd. The survivors tend to solve duller, harder problems.

REGIONAL

Endeavor Catalyst closes oversubscribed US$320M fifth fund: The fund, announced from Singapore, lifts assets under management past US$850M and lands as startup markets outside Silicon Valley begin to stir after several difficult years for venture-backed companies.

Khazanah’s Dana Impak mobilises US$635M for 130 Malaysian firms: Every dollar Khazanah deployed drew 48 cents of external capital. Its Jelawang Capital arm holds up to US$244M to grow Malaysia’s fund manager pool, while mid-tier firms and semiconductors form the other priorities.

Manus parent raises over US$500M after China forces Meta deal unwind: Boyu Capital and IDG Capital co-led the round, with Tencent, HSG and ZhenFund returning. The Singapore-headquartered agent developer disclosed no valuation or plans for the money after Beijing blocked Meta’s US$2B-plus acquisition.

MAS gives Singapore financial firms a year to meet AI risk rules: The principles-based guidelines cover banks, insurers and payment firms as AI moves from back-office experiments into systems that make customer, risk and execution decisions in the region’s leading financial hub.

Revolut commits nearly US$274M to Singapore, targets 300-plus staff: The UK fintech signed a five-year Collyer Quay lease, opening in early 2027, and will hire across engineering, data and AI. It still lacks a Singapore banking licence, operating as a Major Payment Institution.

StarHub to acquire MyRepublic Mobile as Singapore telcos consolidate: The deal completes a multi-year journey that began with StarHub’s 2021 investment in MyRepublic Broadband, and follows reports of StarHub talks to buy M1 from Keppel. Subscribers move onto StarHub’s network without disruption.

EDBI backs Universal Quantum’s US$100M+ round and Singapore R&D hub: The British trapped-ion computing firm’s Series A, co-led by DCVC and Firgun Ventures, will fund its first R&D centre outside Europe. It is the largest Series A by a UK quantum company.

MDI Ventures narrows SEA thesis to AI and digital asset infrastructure: TelkomGroup’s corporate VC arm will prioritise enterprise AI, AI infrastructure, governance software, real-world asset tokenisation and custody, and regulated digital asset rails, citing a shift in the region’s technology cycle.

EnterpriseSG puts US$94M into innovation centres, cuts 27 to 10: The five-year plan simplifies SME support and absorbs technology-matching services from Innovation Partner for Impact, which winds down in March 2027. EnterpriseSG says affected staff will be redeployed where possible.

Pine Labs launches unified payments platform in Singapore: The Indian fintech combines acquiring, acceptance, point-of-sale instalment plans and loyalty via its Fave app, with Courts, CapitaLand and FairPrice already clients. The launch follows its Philippine debut with GCash.

VinFast takes its electric buses to the US through NexMove: US deliveries start in Q1 2028 under an all-in-one service model bundling buses, charging and maintenance. About 2,000 VinFast e-buses already run in Vietnam, and Europe is next on the list.

INTERNATIONAL

US bars Microsoft, Infosys, Wipro and others from green card scheme: Washington accused the firms of fraud and will reject new and pending labour certifications. With nearly three-quarters of H-1B visas going to Indian workers, the move hits Asian talent pipelines.

OpenAI’s annualised revenue nears US$50B, not the reported US$70B: The FT says the higher figure came from investors trying to match Anthropic’s revenue methodology. The gap matters for a company that raised US$122B in March and pushed its IPO to early 2027.

New York alleges TikTok fed minors a placebo safety feature: The lawsuit claims thousands of users, including children, switched on TikTok’s Algo Refresh tool but saw no feed change. It is one of more than two dozen state cases over addictive design.

Apple reportedly co-develops LG-branded smart home devices: Bloomberg says the smart lock, thermostat, doorbell and cameras will pair with Apple’s upcoming hub, expected on 13 October, in a direct challenge to Amazon’s Ring line-up.

CYBERSECURITY

58% of Singapore firms hit by AI-driven phishing attacks: Yubico and Okta’s survey of 1,890 security professionals found 81% enforce multi-factor authentication, yet 41% still rely onpasswords — while 95% want human oversight before AI agents act autonomously.

Hackers breach Asos via Snowflake, then message customers in its app: The Xuanye Group reportedly impersonated a trusted contact to obtain logins, taking names, addresses and profile notes. Asos, with 17 million customers, faces extortion; Snowflake denies its own systems were breached.

Anthropic launches Cyber Mission to arm critical infrastructure defenders: The programme pairs AI models with CrowdStrike’s security tools for power, water, and transport operators, alongside a free OSS Scanner — though its unreviewed vulnerability reports risk inaccuracies and added noise.

SEMICONDUCTOR

GlobalFoundries to make AI chip interposers for TSMC in US$2B deal: Output at GlobalFoundries’ New York plant ramps from H1 2028 under a five-year pact, adding a US source for CoWoS packaging. GlobalFoundries also operates a fab in Singapore.

TSMC posts record quarterly revenue on AI chip demand surge: The world’s largest contract chipmaker’s Q3 revenue hit $46.71 billion, up 50% year-on-year and ahead of forecasts, as AI-driven demand from clients like Nvidia and Apple continues to outpace expectations.

A third of chipmakers plan APAC expansion as AI demand surges: FedEx: Singapore is now the world’s largest chip importer at over US$105B, while Vietnam’s chip exports have grown 32% a year since 2006. Respondents ranked geopolitical disruption the biggest growth risk.

AI

Singtel’s RE:AI, SIT-NVIDIA centre tackle AI deployment gap: Running a pilot is no longer the hard part for companies; turning pilots into reliable operations is. The partnership with the Singapore Institute of Technology targets that step.

Websites are shutting the door on personal AI shopping agents: Amazon blocked Meta’s Muse, while anti-bot checks trip up others. Meta, Walmart, Stripe and others are building an open agent standard to separate legitimate agents from malicious bots.

ChatGPT adds interactive visuals with GPT-6’s Intelligent UI: Answers can now include calculators, editable graphs and charts, which users can dial back. The feature rolls out globally, including to the free and lower-cost Go tiers.

THOUGHT LEADERSHIP

ASEAN’s startup ecosystem is entering its accountability phase: Citing post-Enron America and post-Luckin Coffee China, Bobby Yulandika Putra argues every maturing ecosystem faces scrutiny, and Southeast Asia’s turn has arrived.

Why SEA is underbuilt in the categories behind durable companies: Grab, Sea and GoTo won by amassing users and burning capital. Chris Chen argues the consumer playbook leaves gaps in the categories that produce the region’s longest-lasting companies.

AI labs want the corporate foothold Excel once gave Microsoft: Alexander Khomenko argues frontier labs are chasing enterprise lock-in across Southeast Asia, aiming to embed themselves in company workflows as deeply as spreadsheets once did.

Operators risk scaling on ASEAN’s old trade rulebook after upgrade: For firms shipping into several ASEAN markets, the question has moved from tariffs to certificates and proof of origin. Fathhi Mohamed argues most are unprepared for the paperwork.

SEA is digitising health records; making them portable is harder: Patient portals and health wallets are the easy part. With WHO and Temasek Foundation backing a three-year initiative, Ubaid Pisuwala argues interoperability is the real challenge.

AI writes code faster, but engineers now struggle to trust it: Citing Google’s 2025 DORA report, James Kithokoi argues faster code generation has moved the bottleneck to review and verification, so teams are not shipping as fast as promised.

DePIN and RWA alliances find fertile ground in Southeast Asia: Astrid Dang argues decentralised infrastructure networks and tokenised real-world assets could address high infrastructure costs, thin banking reach and volatile local currencies across the region.

AI could make concierge medicine the new front door to healthcare: Concierge practices cap panels at 400 to 600 patients, against 2,000-plus for a typical primary-care doctor. John Millar argues AI could scale scarce physician attention to far more people.

Europe’s trade truce with the US has not restored trust in US firms: Shawn Balakrishnan argues American companies can no longer assume they are seen as predictable and reliable commercial partners, even after the latest phase of US-European détente.

Why founders should sometimes say no to paying customers: More contracts look like progress, but Julien Ricciarelli-Bonnal argues the wrong customers can damage growth, even for early-stage firms that feel unable to turn down revenue.

Creative talent is plentiful; dependable production capacity is not: Xiaohu Hou recounts dropping a production team despite the client accepting the video, arguing that polished samples rarely prove a vendor can deliver reliably.

Are you leading with people, or leading with tech?: Vincent Tan argues leaders who ignore where scarcity now lives keep supplying crowded markets with little demand, forgetting the basics of economics.

Light liquidations and flat funding: Is Bitcoin about to break out?: Bitcoin held a tight US$85,400 to US$85,550 band on 7 October with no sustained breakout. Anndy Lian reads low liquidations and flat funding as signs of a market coiling for a move.

The post Ecosystem Roundup: Granite Asia, Google bundle cash and compute for Asia’s AI founders appeared first on e27.

The artificial intelligence boom is no longer just a story about software models, cloud platforms, or the companies racing to build the next large language model. It is increasingly a story about where chips are made, how far they travel, and whether the supply chains behind them can withstand the next geopolitical shock.

A new FedEx report suggests semiconductor companies are preparing for a more distributed manufacturing map across Asia Pacific, as demand from AI, high-performance computing, electric vehicles, advanced communications, and data centres puts fresh pressure on chip production networks.

Also Read: AI boom shields Asia Pacific, but Moody’s warns the cushion is thinning

According to FedEx’s “From AI to EVs: How APAC semiconductors drive the global innovation map” report, 35 per cent of semiconductor companies surveyed expect to expand manufacturing into more Asia Pacific markets within the next three years. Another 25 per cent plan to increase or diversify their sourcing.

The survey was conducted among attendees at the SEMICON Taiwan trade show. It therefore reflects the views of industry participants gathered at one of the region’s most important semiconductor events, rather than the entire global chip industry. Still, the findings offer a useful snapshot of how chip companies are thinking about growth, risk, and regional diversification at a time when semiconductors have become a strategic priority for governments and businesses alike.

Nearly four in five respondents pointed to Taiwan, Southeast Asia, Greater China, and Japan as the markets likely to see the strongest demand growth. That is a significant signal for Southeast Asia, which has historically played a quieter but critical role in the global chip supply chain through assembly, packaging, testing, and logistics.

From efficiency to resilience

For years, semiconductor supply chains were built around efficiency: specialised production in the lowest-cost or most capable locations, tight inventory management, and complex cross-border flows. That model delivered scale, but recent disruptions have exposed its limits.

FedEx said 30 per cent of respondents named geopolitical disruption as the biggest challenge to growth, ahead of costs and customs complexity. In response, companies are shifting towards geographic diversification, larger inventory buffers, and supply chains designed to absorb shocks rather than simply minimise costs.

That shift matters because chips are among the most complex products in global trade. FedEx said chip components can travel more than 25,000 miles and cross over 70 borders before completion. A single chip may involve design in one market, wafer fabrication in another, equipment or materials from several countries, packaging and testing elsewhere, and final integration into a device sold globally.

This complexity has become more consequential as AI adoption accelerates. Training and running advanced AI models require vast computing power. That in turn depends on AI accelerators, graphics processing units, high-bandwidth memory, networking chips, and the data centre infrastructure that connects them. FedEx said more than 2,000 new data centres are expected worldwide between 2026 and 2035.

Also Read: Malaysia’s chip suppliers face rising pressure to prove cyber resilience

Salil Chari, FedEx’s Asia Pacific president, said the coming decade of innovation would depend on the resilience of the supply chains behind AI and digital infrastructure. The point goes beyond logistics: without reliable production and movement of advanced chips, the AI economy’s growth could run into physical bottlenecks.

Southeast Asia’s opening

The global semiconductor market is forecast to reach US$975 billion by 2026, according to the report, with Asia-Pacific accounting for roughly 58 per cent of industry revenue. China alone makes up close to 29 per cent of the global market.

Within this landscape, Southeast Asia is becoming harder to ignore.

The region does not yet rival Taiwan in advanced foundry manufacturing or South Korea in memory chips. But it has long been deeply embedded in the “back end” of chip production, the assembly, packaging, and testing stages where wafers are turned into usable semiconductor components. As chip companies look to diversify operations, these capabilities could become more strategically valuable.

Singapore, for instance, has become the world’s largest chip importer, topping US$105 billion, according to the report. That reflects its role as a regional node where components move in and out for assembly, distribution, and integration into broader electronics supply chains. The city-state also hosts semiconductor manufacturing, precision engineering, logistics, and research operations, making it one of Southeast Asia’s most mature chip ecosystems.

Vietnam is emerging quickly. FedEx said the country’s chip exports have grown at a compound annual growth rate of 32 per cent since 2006, with integrated-circuit exports reaching US$32.4 billion in 2023. Its rise is part of a wider manufacturing shift as electronics companies expand beyond China and seek alternative production bases in Asia.

Malaysia and the Philippines are not highlighted in the same figures, but they remain important players in assembly, testing, and electronics manufacturing. Malaysia in particular has built a strong position in semiconductor packaging and testing over several decades, while Penang has become one of the region’s most important electronics manufacturing clusters.

Also Read: Malaysia wants 300,000 AI jobs by 2030. Talent will decide if it gets there

For Southeast Asian governments, the opportunity is clear but not simple. Moving up the semiconductor value chainrequires specialised talent, reliable power, water infrastructure, industrial land, long-term policy stability, and deep links with global equipment, materials, and design firms. The capital requirements are high, and competition for investment is intense.

Asia Pacific remains the centre of gravity

FedEx’s report underlines how specialised the semiconductor map has become across Asia-Pacific. Taiwan leads advanced manufacturing with roughly 60 per cent of global foundry output. South Korea dominates memory chips. Japan supplies critical equipment and materials and holds about 30 per cent of the global equipment market. China is both a major materials supplier and a growing fabrication base.

This concentration has made Asia-Pacific indispensable to the global technology economy. It has also made the region a focal point for industrial policy, export controls, and national security debates.

For startups and technology companies in Southeast Asia, these shifts may feel distant, but they are increasingly relevant. AI adoption, cloud services, electric mobility, fintech infrastructure, robotics, and advanced manufacturing all depend on semiconductor availability. Chip shortages can delay product launches, raise hardware costs, and constrain growth in sectors that rely on connected devices or computing power.

AI accelerators and high-bandwidth memory chips now account for about a fifth of chip sales, FedEx said, making them the main growth drivers in the industry. That demand is unlikely to ease soon as enterprises, governments, and cloud providers build out AI capacity.

Also Read: Southeast Asia’s chip-hub ambition is colliding with its chip-smuggling problem

The larger question is whether supply chains can scale fast enough while becoming more resilient. For Asia-Pacific, the answer will shape not only the semiconductor industry but also the next phase of digital infrastructure, electric vehicles, advanced communications, and AI-led services.

Southeast Asia is unlikely to replace Taiwan, South Korea, Japan, or China in their core strengths. But as chip companies spread risk and add capacity, the region has a chance to deepen its role in one of the world’s most strategically important industries. The next three years may determine how much of that opportunity it can capture.

The post AI boom pushes chipmakers to expand across Asia Pacific, FedEx report finds appeared first on e27.