Every enterprise software vendor now has an AI agent story. Oracle NetSuite’s version, unveiled at SuiteWorld in Las Vegas on October 26, comes with two twists: its agents are meant to know your industry, and its newest offering is designed to catch tech startups on day one and keep them until IPO.

The company announced SuiteAgents, a set of AI agents built to help organisations work through industry-specific business processes. They sit on NetSuite’s unified data model, and the company says they draw on its 28 years of accumulated industry knowledge to flag issues, recommend next steps and help carry out work inside existing workflows.

Also Read: Three key learnings from SuiteWorld 2025: Why context, not automation, is the real AI prize

“Business leaders manage a lot of moving parts, and the industry signals that matter can easily be missed when teams are focused on day-to-day operations,” said Evan Goldberg, founder and executive vice president of Oracle NetSuite. The agents, he added, are meant to help finance and operations teams “respond faster while retaining oversight, judgment, and accountability.”

The division of labour is fairly clear. Ask Oracle is the conversational layer, where users put questions to the system using industry-specific skills, knowledge and tools. SuiteAgents work in the background, handling ongoing monitoring and multi-step execution, all within NetSuite’s existing governance models.

Six industries, three shipping

NetSuite has lined up agents for six verticals. For software companies, the agent looks for conditions that can hurt revenue retention, margin, financial health and compliance across contract orchestration, revenue compliance, billing and forecasting. The product agent assesses inventory exposure and supply changes, and advises on what they mean for inventory efficiency, supply performance and margin.

The services agent turns statements of work into editable project plans, reviews staffing capacity, flags time-entry exceptions and assesses project health. In construction, the agent converts estimates into job budgets and prepares payment applications based on verified progress, approved changes and retainage terms, the portion of payment withheld until a job is finished.

The healthcare agent combines inventory, service-line, reimbursement and payer-mix data to identify supply and financial risks, while the nonprofit agent focuses on grant-related financial and compliance issues.

The catch is availability. Only the Software, Product and Construction agents are live in NetSuite Next today. The rest are slated for a future release, and the company’s own disclaimer notes that timing is subject to change and that the announcement is not a commitment to deliver any functionality. Buyers in healthcare, services and nonprofits should read the roadmap as intent, not inventory.

Why Goldberg thinks this time is different

At a press conference after his keynote, Goldberg pointed to NetSuite’s AI Connector Service, launched about a year ago, which he said now has around 13,000 companies using it. What NetSuite learnt from watching those customers, he said, has been fed directly into NetSuite Next and convinced the team that this is “a totally different generation” of business software.

Asked about the mixed results enterprises have had with chatbots, Goldberg conceded that a maturing technology will always produce uneven outcomes, and that AI is non-deterministic by nature. But he cited customer examples from his keynote of “companies saving hundreds of thousands of dollars and thousands of hours”.

Also Read: Who referees the AI agents? Enterprises aren’t sure, new report shows

Goldberg, however, declined to forecast how many of those 13,000 AI Connector users will move into NetSuite Next, saying general availability had only just launched. NetSuite is pursuing what it calls an “outside-in” approach, using AI tools outside NetSuite with its data, alongside an “inside-out” approach, working within the platform. Goldberg expects many users to mix the two: data-heavy NetSuite work inside, and tasks pulling from many other sources outside.

The trust problem

Trust came up repeatedly. Goldberg acknowledged that AI models “always answer very confidently,” which is not the same as being right. His answer is context: keeping most of a business’s data in one place and building the collective knowledge of NetSuite experts and its partner channel into the skills that guide the AI. He said he had shown in his keynote that the same question, run through SuiteSkills on three different AI models, produced the same answer.

That framing will resonate in a region where tech leaders are learning to trust AI agents, but not yet with critical systems.

On data, Goldberg was unequivocal: it belongs to the customer. NetSuite only uses high-level, aggregated usage information, such as which questions get asked repeatedly, to improve the service. For firms already anxious about data sovereignty, that is a necessary assurance, if not a complete one. The question of whether the user or the integrator bears responsibility when AI gets something wrong went largely unanswered.

Free, for now: Quickstart goes after startups

The more strategically interesting announcement may be Quickstart. Goldberg was clear that the target is tech startups, which he said are being created in greater numbers than ever because cloud infrastructure and AI have made building companies easier. NetSuite wants to take them “from their original idea to IPO.”

Quickstart offers accounting and subscription capabilities, plus some product features for startups with a physical element. When companies outgrow it, they can switch to NetSuite Next in one click. It runs largely in the background, so founders can get financial information wherever they work. As for the free tier, Goldberg was candid: “It’s not gonna be free forever.”

Also Read: How a 39-store retail chain in Laos moved from Excel to AI agents

Asked whether micro businesses, which a reporter said make up about 97 per cent of Australia’s 2.7 million small businesses, were in scope, Goldberg said NetSuite would start with tech startups and see how it goes. On whether NetSuite might bundle credit lines or working capital, he pointed to existing integrations with partners and said the company is still learning what founders want, and how to monetise it.

This is a notable shift for a product long pitched at the mid-market. Goldberg agreed that, between Quickstart and SuiteExpress, NetSuite expects to become a choice for far more companies than before.

SuiteExpress, and the wait outside the US

SuiteExpress is NetSuite’s answer to the problem every ERP salesperson knows: messy data, poorly mapped processes and weak governance. Goldberg described it as ingesting everything a company currently does, then using AI and NetSuite’s expertise to map it into something more rational, with companies moving from complicated environments to NetSuite in weeks. Anyone who has seen the hidden cost of cheap ERP implementations will want to see that claim tested.

For now, SuiteExpress is available only in the US. Goldberg said NetSuite is moving products outside the US “way faster than ever” and intends to deliver value to all markets “as close to simultaneously as possible,” but offered no firm date. Southeast Asian firms, already weighing a growing field of AI accounting ERP options, will have to wait.

The competitive logic is plain. NetSuite faces SAP and Microsoft Dynamics 365 at the larger end, and Xero and Intuit’s QuickBooks at the small-business end. QuickBooks is the very system, as one reporter noted, that growing companies have traditionally ripped out when they move to NetSuite. Quickstart is an attempt to get there first.

The zero-day close, and other promises

On NetSuite’s vision of an autonomous financial close, Goldberg was careful. He said he is not a CFO, but that it is hard to overestimate what AI may eventually do; the real question is how quickly. His practical point was human: people procrastinate and leave much of the close until the deadline looms. Agents can do that pre-work continuously, at any hour. Before anyone reaches a zero-day close, he said, finance teams would be “really happy with one day, two days rather than 20 days.”

The pace has changed too. NetSuite now ships monthly releases, something it never did before. An “Auto” setting already picks the AI model for a task, and Goldberg said recommending the right model or reasoning level is a direction NetSuite will pursue. Admins can roll new features out to early adopters first, avoiding a “big bang” launch. That kind of control is what regulated sectors such as banking demand before handing agents real work.

Also Read: Parallel Loop nets US$6.6M to sell enterprises the one thing AI models lack: context

Looking two to three years ahead, Goldberg expects more industry solutions. AI writes code fast, he noted, but working out with customers what actually works does not speed up as much. He is also watching time-series and other data-centric models that could improve forecasting. Seeing around the corner, he admitted, is a claim made for years.

That candour is useful. The plan is coherent, but half the agents have not shipped and SuiteExpress has not left the US. As Southeast Asian enterprises push for AI governance before scale, NetSuite will be judged on delivery, not keynotes.

—

(The author is in Las Vegas covering NetSuite SuiteWorld 2026 at the invitation of Oracle.)

The post Evan Goldberg on AI trust: Confident answers aren’t enough, context is everything appeared first on e27.

Bukalapak has kept its profit streak alive, though only just. The Indonesian tech firm reported positive adjusted EBITDA of US$55,088 for the third quarter of 2026, a swing from a loss of about US$991,000 a year earlier. Over nine months, adjusted EBITDA reached roughly US$606,000, against a US$2.9 million loss in the same period of 2025. Quarterly revenue held flat at about US$88 million, while nine-month revenue climbed 19 per cent to around US$308 million.

Gaming now does the heavy lifting. Its marketplaces, Itemku and Lapakgaming, generated about US$77 million in quarterly revenue, and nine-month gaming revenue rose 27 per cent on the back of international expansion. That push has a price: selling and marketing spend jumped 45 per cent year on year to about US$3.9 million in the quarter. Mitra Bukalapak, the MSME unit, saw nine-month revenue fall 21 per cent as it moved to a third-party model, though its margins improved. BMoney, the investment arm, grew quarterly revenue 34 per cent.

With about US$925 million in cash and liquid investments, Bukalapak has room to manoeuvre. The harder question is whether a company once pitched as Indonesia’s e-commerce champion can build a durable business on gaming top-ups rather than the marketplace that made its name.

REGIONAL

Son reportedly picks Malaysia for US$50B AI and data centre push: Prime Minister Anwar Ibrahim said SoftBank’s founder chose Sarawak and Peninsular Malaysia for AI and data centre projects. SoftBank has not confirmed the plan, which would test a grid already strained by Johor’s data centre boom.

Ailytics raises US$7.4M to turn idle CCTV into safety AI: Tin Men Capital led the Series A, joined by SG Growth Capital and Japan’s Safie. The Singapore startup grew to 400-plus deployments in 15 countries, though its 73% injury-reduction figure remains customer-reported.

Singapore commits US$112M to cut industrial emissions: EDB and AEDO will fund carbon capture, utilisation and storage research, with up to US$78M for company-led projects under the new DRIVE programme. Industry accounts for nearly two-thirds of Singapore’s emissions.

SGX hunts for new CEO as Loh Boon Chye plans 2027 exit: Loh’s contract runs to mid-July 2027 and no transition date is set; chairman Koh Boon Hwee will help pick a successor. SGX posted record revenue of US$1.17B and US$595M adjusted net profit in FY2026.

Singapore launches US$59M fund to decarbonise power generation: The EMA-AEDO initiative covers hydrogen, solar and biomass, with targets such as hydrogen extraction costs below US$1 per kg. Startups can co-design projects with public researchers, and the state funds up to 75% of qualifying costs.

Korea’s LS Eco Energy puts US$49M into Vietnam ultracapacitor line: The LS Group unit will build medium-sized ultracapacitor cells at LS Cable & System Vietnam by June 2027, funded partly by a US$26M rights offering. The filing names no customers.

Singapore breaks ground on US$49M low-carbon testbed: Amogy, IHI and local startup Metha8 are the Jurong Island facility’s first users. Amogy will pilot ammonia-to-hydrogen cracking at one tonne a day, while IHI tests catalysts that turn CO2 into aviation fuel.

Danantara CIO defends Indonesian corporates amid MSCI scrutiny: Pandu Sjahrir said 134 of 146 large companies listed in Indonesia turned a profit, with first-half earnings up 15%. He urged investors to separate market-governance failings from corporate fundamentals.

Igloo targets five-fold growth for its insurtech software arm: The Singapore firm will use acquisitions to take Igloo Tech Solutions into the Middle East. Its AI claims tool handled 50,000 cases in Q1 with 94% accuracy, and group EBITDA breakeven is targeted for end-2026.

Carro bets Chinese EV influx will feed its used-car business: The Singapore marketplace now sells Dongfeng, Zeekr and Wuling cars. It also rejected 2008 comparisons over its US$352M car-loan securitisation, saying it holds the junior tranche and takes the first loss.

AnyMind opens in Los Angeles to bridge Asian brands and the US: The Singapore-founded, Tokyo-listed firm has helped 15 Asian brands grow in the US over seven months. It will lead with AnyLive, its live-commerce platform, and agentic AI tools.

Data centres could swallow most new power in Johor and Batam: BMI says Johor’s data centres would absorb more than 90% of Peninsular Malaysia’s new generation by 2031. In Batam, DayOne’s campus alone has contracted 450MW, half the island’s installed capacity.

Singapore and Malaysia renew energy pact on cross-border power: EMA and Malaysia’s Energy Commission renewed their 2020 cooperation pact at Singapore International Energy Week, covering electricity trading, renewable energy certificates, solar, storage and interconnection capacity.

INTERVIEWS AND FEATURES

Wealthy founders hedge with gold, not supercars, HSBC finds: Some 61% of entrepreneurs surveyed plan to add gold over 12 months, and half will raise private equity allocations. The barbell strategy keeps personal wealth defensive and confines risk to their own companies.

Thirty Southeast Asian climate startups building beyond the pledges: Seventeen of the 30 are based in Singapore while Vietnam has none, and about half have yet to disclose a round above US$1M. VFlowTech’s US$20.5M raise leads the list.

INTERNATIONAL

Anthropic’s early backers sit on billions ahead of IPO: With Anthropic privately valued at US$965B, Spark Capital’s US$75M stake has grown more than ninetyfold to US$7B, and Iconiq’s US$6B cheque is worth US$17B. The listing could value the lab at up to US$2T.

Instacart explores a sale after talks with DoorDash and Uber: The US$13B grocery delivery firm has hired an investment bank to weigh its options, The Information reported. Uber reportedly walked away, and Bloomberg says talks are early. Shares rose 9.2% on the news.

Zero Industries raises US$3M to beam laser power to satellites: Maezawa Fund led the pre-seed round for the Japan-linked, San Francisco-based startup, which targets a 2030 orbital demonstration. Efficiency is the hurdle: an NTT-Mitsubishi test delivered 152W from a 1kW laser.

Unacademy founders raising US$73.5M fund for deep-tech startups: Roman Saini will run Longshot Capital full-time. The fund has about US$31.5M in commitments, targets a January 2027 first close, and will back seed-stage AI, defence and robotics bets. upGrad acquired Unacademy at a US$206M valuation weeks ago.

Tata Communications to spend US$450M on fibre for India’s AI corridors: The three-phase rollout adds about 10,000 km of fibre, starting with the Mumbai-Pune-Hyderabad-Chennai corridor before reaching Vizag, Bhubaneswar, Jamnagar and GIFT City. It links data centre clusters with cloud firms, global capability centres and banks.

CYBERSECURITY

Decades-old PHP flaws force site owners to patch or go dark: Human researchers in OpenAI and Wiz’s Hack the Planet initiative found the bugs, but attackers raced to exploit them around disclosure. PHP underpins WordPress and Drupal, leaving under-resourced sites most exposed.

Japan’s privacy law trips up founders using GDPR playbooks: APPI leans harder on consent for cross-border transfers but offers a cloud exception GDPR lacks. Japanese buyers expect vendors to arrive with APPI-level contract terms, and those that do shorten their sales cycles.

SEMICONDUCTOR

GlobalFoundries tests an AI-run control room at its Singapore fabs: Built with Siemens, the centre merges five fab control rooms into one and deploys sensor-laden robots. GF gives no downtime, energy or cost targets, and centralisation hands attackers a single high-value target.

Vietnam’s first chip fab likely to miss 2027 target, no ASML deal yet: Military-run Viettel broke ground in January, yet satellite images show no buildings. Sources say months of talks over a DUV lithography machine remain unresolved, and Dutch export-control exposure is unclear. Viettel insists the US$1.3B project is on schedule.

AI

OpenAI cuts frontier model training off from the live internet: The lab will train on cached and simulated web copies after a model exploited a misconfiguration in September to reach the internet. Anthropic began imposing similar limits two weeks earlier.

Apollo economist finds no AI payoff outside Big Tech: Torsten Slok’s look at the S&P 500 minus seven tech giants shows flat profit margins and revenue per worker, with no evidence of productivity gains. He argues business value hinges on judgement, which AI lacks.

inDrive to test Pony.ai robotaxis in Kazakhstan from 2027: Astana trials will start with safety operators, using Pony.ai’s Gen-7 vehicles and Nvidia’s Hyperion platform. Nvidia says the platform could lower robotaxi development costs, vital in emerging markets where rides are already cheap.

Seven in 10 APAC travellers open to AI handling work on holiday: Booking.com’s Travel Predictions 2027 survey of 28,000-plus travellers across 32 markets found 47% would let AI manage emails and admin on holiday. Booking.com and rival Expedia are both adding AI planning tools.

THOUGHT LEADERSHIP

How AI agents found US$570K in lost sales at a Laos retailer: QUEST EDTECH’s CEO rebuilt LDC’s Excel reporting with AI agents for about US$100 a month. The hardest fixes were human, such as agreeing what counts as out-of-stock, since four definitions gave rates from 0.6% to 77.3%.

Japan needs execution, and Southeast Asian startups can supply it: A former FPT Software analyst argues Vietnam–Japan ties are shifting from outsourcing to co-creation, citing FPT’s AI tie-up with Sumitomo and SBI. Pilots must still turn into paying contracts.

Indonesia’s farms are ageing as young people head for cities: Farmer numbers fell 7.5% in a decade to 29.3 million, and only 2.6 million younger farmers use digital tools. The author says the Petani Millennial programme lacks transparency and urges certainty on land and credit.

Two part-timers and an AI orchestrator behind a Dubai summit: SR21’s Alex Lin explains how AI derived the “Breaking Point” theme, briefed 48 senior leaders and fact-checked speakers in real time. Logistics, he argues, was its least significant contribution.

The post Ecosystem Roundup: Bukalapak stays in the black as gaming carries the load appeared first on e27.

Thirteen years after venture capitalist Aileen Lee coined the term, the unicorn has stopped being mythical. There are now 1,879 privately held companies valued at US$1 billion or more, according to a new analysis by BestBrokers, which drew on data from Crunchbase, PitchBook and TechCrunch.

The headline number hides a narrower story. Capital in 2026 is crowding into a few sectors (AI, robotics, defence) and a few countries, chiefly the US and China. Southeast Asia, for all the talk of being the next growth frontier, shows up in the report mostly as a footnote.

A caveat before the numbers: BestBrokers is a broker-comparison site, not a research house, and private valuations are soft by nature. Many reflect a single priced round or a secondary sale. Read them as a gauge of investor appetite, not audited worth.

Also Read: One-person company: Why Southeast Asia’s next unicorn may look nothing like Grab

Here are 10 numbers that define this year’s unicorn class.

1. 1,879 unicorns, and more than half are American

The US is home to 971 of the world’s billion-dollar private companies. China follows with 330, the UK with 85 and India with 77. Latin America, often lumped with Southeast Asia as an emerging venture market, counts 38 across six countries, led by Brazil’s QuintoAndar at US$5.1 billion.

2. US$965 billion: Anthropic is now the world’s most valuable unicorn

The crown changed hands after SpaceX’s IPO, which raised more than US$86 billion. Anthropic now sits at the top, with OpenAI close behind at US$852 billion after a US$122 billion round led by Amazon, Nvidia and SoftBank. ByteDance is third at US$550 billion.

Together, those three are worth roughly US$2.37 trillion. Then comes a cliff: Databricks at US$190 billion, Stripe at US$159 billion and Revolut at US$115 billion. For Southeast Asian AI startups building on top of a few foundation models, that concentration is both a supply chain and a dependency, and regulators are starting to take notice.

3. 233 new unicorns in 2026, and 139 of them are in the US

That is close to 60 per cent of the year’s new billion-dollar startups. China added 34, the UK 14, Germany eight, India six and Canada five. The report names no Southeast Asian startup among the 2026 arrivals.

4. One in four new unicorns is an AI company

Of the 233 startups that crossed the line this year, 56 (24 per cent) are in AI. They range from frontier labs such as DeepSeek to sellers of coding agents (Factory), customer-service agents (Wonderful) and video generators (PixVerse).

AI firms also hold seven of the 13 most valuable new unicorns. Hark, founded by Figure AI CEO Brett Adcock, hit US$6 billion after a Series A of more than US$700 million. World-model developer General Intuition is at US$6.2 billion, and the UK’s Ineffable Intelligence reached US$5.1 billion on what the report calls Europe’s largest-ever seed round. Closer to home, governments are trying to engineer the next wave, with Thailand roping in OpenAI for a public-private accelerator.

5. 29 robotics unicorns, and China minted 17 of them

Robotics is the second-busiest sector, accounting for 12.4 per cent of new unicorns as humanoids and embodied AI attract serious money. XPENG Robotics leads China’s cohort at US$6.3 billion, followed by AI2 Robotics and X Square Robot at US$3 billion each. The US has added seven, led by Mind Robotics at US$3.4 billion.

Southeast Asia has a foothold, not a fleet. Singapore’s dConstruct raised a US$125 million Series A earlier this year, but the city-state’s lead in the region is less reassuring than it looks. And as one founder told e27, data, not hardware, is the real bottleneck in humanoids.

6. US$91 billion: DeepSeek is the year’s breakout

The Hangzhou-based lab, founded in 2023 by Liang Wenfeng, reportedly raised around US$15 billion in only its second funding round. Its valuation is more than 14 times that of the next most valuable new unicorn.

DeepSeek earned its spot by making low-cost R1 and V3 models that rattled Western rivals while using far less compute. The size of the gap also says something about the scarcity of credible non-US frontier labs, which lets investors pile into the few that exist.

7. 14 defence unicorns, as geopolitics meets the cap table

Germany’s STARK leads at US$4 billion, helped by a EUR300 million (about US$350 million) contract to supply autonomous strike drones to the German armed forces. Nine of the 14 are American, led by Mach Industries, which more than doubled its valuation to US$3.7 billion in three months after a US$600 million Series C extension.

Also Read: Fintech, DeFi and applied AI define Southeast Asia’s new venture discipline

France’s Harmattan AI became the country’s first defence unicorn after a US$200 million round led by Dassault Aviation. Autonomous weapons are now a venture asset class; the debate over how they should be governed has not kept pace with the funding.

8. 21 new fintech unicorns, as Southeast Asia’s fintech tightens its belt

Fintech made up 9 per cent of new unicorns, with Corgi Insurance leading at US$4 billion. Investors favoured payments, lending, financial infrastructure and AI-assisted financial services. Healthtech was close behind with 17, led by US device maker MiRus at US$4.4 billion after a US$1.5 billion investment from Boston Scientific.

The regional mood is cooler. Fintech funding in Singapore dropped to US$499 million as dealmaking turned selective, and the sector’s next phase is about scaling on the balance sheet rather than on user numbers.

9. 27 unicorns put Singapore eighth in the world

Singapore is the only Southeast Asian country the report breaks out. Its unicorns are led by data centre operator DayOne, which has filed for a US listing. The ranking shows the city-state punching above its weight. It also shows how much of the region’s unicorn story still runs through one small island.

10. Six new Indian unicorns, including a homegrown LLM maker

India has 77 unicorns in total, topped by Walmart-backed PhonePe at US$14.5 billion. Its 2026 class is small but telling: fintech lender KreditBee and AI startup Sarvam AI lead it at US$1.5 billion each. Sarvam’s arrival suggests investors will back a local-language model maker outside the US and China, at a far smaller price tag.

What it means for Southeast Asia

The 2026 unicorn class is built on capital intensity: compute clusters, humanoid hardware, chip fabs and defence contracts. Those are bets that need patient money in very large cheques, and Southeast Asia’s venture market, still recovering from its correction, is not built to write many of them.

Also Read: Echelon Philippines 2025 – The Unicroach vs Unicorn approach: Chasing profitability over growth at all costs

That does not leave the region empty-handed. It leaves it with a choice between owning a layer of the AI and robotics stack or renting one from the companies at the top of this list. The super-app era taught the region how to build distribution; this cycle is asking whether it can build anything that others need to buy.

For now, the unicorns are multiplying. Most of them just aren’t grazing here.

The post Anthropic, DeepSeek and drones: 10 takeaways from the 2026 unicorn boom appeared first on e27.

Everyone is worried about the wrong AI failure.

For two years, the loudest AI safety conversation has centred on hallucination: the model that invents a fact, a court case, a source that never existed. That risk is real and well documented.

But a quieter problem is spreading underneath it, and it may end up mattering more. AI answers can be completely accurate and still be impossible to check. Increasingly, that isn’t an edge case. It is the default experience.

The click is disappearing

Zero-click searches, where a person gets an answer and never visits a website, reached 68 per cent in early 2026, up from 60 per cent in 2024, according to data reported by DesignRush. Global publisher traffic fell 33 per cent year-on-year, small publishers lost 60 per cent of their referral traffic, and Business Insider lost more than half its search traffic and cut 21 per cent of its staff.

This isn’t really a story about journalism nostalgia. It’s a story about incentives. The people who once had a reason to produce timestamped, checkable, primary information — the reviews, price histories, safety recalls, and on-the-ground reporting that both humans and AI models quietly depend on — are losing the traffic that funded doing it properly.

The citation is disappearing faster than the click

Even when an AI system answers using real, current information, it rarely tells you where that information came from. Similarweb data reported by Search Engine Journal found that only 6.8 per cent of ChatGPT answers included an external link as of May 2026, up from roughly 1 per cent a year earlier. That still means 93 out of 100 responses send the user nowhere to verify anything.

Meanwhile, the machines are reading more of the web than ever. Digiday reports that automated traffic now accounts for 57.4 per cent of all web requests, more than actual humans, and AI agent traffic grew 45 per cent quarter-on-quarter in the second quarter of 2026 alone. Yet 97 per cent of the files publishers built specifically so AI systems could read and credit them received zero requests.

Put plainly: the machines are consuming the open web faster than ever, crediting almost none of it, and the people who built that web are being paid for it even less than before.

Also Read: Ecosystem Roundup: Taiwan blocks Grab’s US$600M foodpanda deal over Uber’s stake

This stopped being just a search problem. It’s now a money problem

The same pattern is showing up at the exact moment it matters most: when people decide what to buy. Digital Applied reports that half of US shoppers say they have made a purchase after researching it inside an AI tool, even when the transaction itself happened elsewhere. But in-chat checkout — the one step in that journey that leaves an auditable, timestamped record — converts roughly three times worse than a normal click-through to a retailer’s own site. OpenAI has already partly retreated from it, and Walmart’s Daniel Danker described in-chat checkout as “a very temporary moment in time.”

So the part of the shopping journey moving fastest into AI is discovery and reasoning, the part with no receipt. The part that still leaves a paper trail is the part being pushed back to ordinary websites. That should concern anyone who thinks about consumer protection, not only people who worry about publishing economics.

Why I keep returning to this

Before I built anything in software, I worked as a paramedic. The habit that job drilled into me was never cleverness. It was sequence: check the evidence, understand the situation, then act. You don’t get to skip the first step because the second one feels urgent, and in an emergency, skipping it is how people get hurt.

A great deal of AI product design today runs in the opposite order. The action — the answer, the recommendation, the purchase — arrives instantly. The evidence behind it, if it exists at all, is buried, unlinked, or already gone by the time anyone would think to check it.

Also Read: How AI became my sparring partner

What “showing your work” should actually mean

None of this is an argument for slowing AI down or forcing every chatbot to behave like a search engine with extra steps. It is an argument for a narrower, buildable standard: any AI system that influences a real decision — what to buy, what to believe, what to do next — should be able to show, on request, what claim it made, when it made it, and what that claim was based on.

That is a design requirement, not a moral lecture. Timestamps and traceable sourcing are inexpensive to build in from day one and expensive to bolt on after the fact. Most teams currently treat them as optional, if they think about them at all.

Southeast Asia does not have the luxury of waiting for someone else to solve this first. The region is folding AI assistants into commerce, banking and everyday decisions at speed, often for consumers who have far fewer alternative ways to double-check a claim than someone with a research team or a market data terminal. If the underlying layer of verifiable information keeps thinning out while AI interfaces keep expanding on top of it, the gap between people who can check what they’re told and people who can’t will not close. It will widen quietly, behind a much friendlier-looking interface.

The fix is simple to describe, even if it takes real engineering discipline to build: keep the receipts, and put them where the person making the decision can actually see them.

An answer you cannot check is not intelligence. It is just a well-formatted rumour.

—

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 The most dangerous thing about AI isn’t hallucination. It’s that it stopped showing its work appeared first on e27.

In a convention centre full of AI agents, autonomous closes and pyrotechnic keynotes, James Chisham has the least glamorous job at Oracle NetSuite’s SuiteWorld in Las Vegas.

He is the man who makes the software obey the taxman.

As Vice President of Product Management and International Strategy, Chisham is responsible for making a US-built ERP behave correctly in Kuala Lumpur, Manila, Mexico City and Munich. In Southeast Asia, that means a patchwork of e-invoicing mandates, tax engines and data rules that change faster than most finance teams can update their spreadsheets.

Also Read: Evan Goldberg on AI trust: Confident answers aren’t enough, context is everything

In a conversation with e27, Chisham was candid about where NetSuite is ahead, where it is still catching up, and where the answers remain, in his own words, “a mix”.

One door, many mandates

Chisham’s starting premise is simple. E-invoicing is coming everywhere.

“There’s every likelihood that every market will eventually adopt e-invoicing, so we wanted to be ahead of that wave,” he said.

Southeast Asia is already proving him right. Malaysia has been phasing in MyInvois, its national e-invoicing system, while Singapore is rolling out InvoiceNow requirements for GST-registered businesses. Each mandate has its own formats, validations and deadlines.

NetSuite’s answer, according to Chisham, is what he calls an “extensible core”. Customers create and receive invoices in NetSuite Next exactly as they always have, while the system handles each jurisdiction’s requirements in the background.

“It could be a Malaysian mandate, a Singapore mandate, a Germany mandate, a France mandate. As far as the NetSuite Next experience is concerned, it works in the same way,” he said.

He argued that this is the company’s edge over specialist e-invoicing providers, the firms for whom compliance is all they do. “Our advantage is that single point of entry, multiple mandate support,” he said. For a growing startup expanding from Singapore into Malaysia and the Philippines, one vendor instead of three has obvious appeal. Whether one vendor can stay as current as three specialists is the question buyers should press on.

Demand first, Vietnam next

NetSuite’s localisation footprint in the region follows its customers, not a map. Chisham said the company has been “driven by where we have our demand, where we have a big customer base”, naming markets such as Singapore and the Philippines.

That explains the gaps. NetSuite’s Southeast Asia localisation package officially covers Malaysia, the Philippines, Singapore and Thailand. That leaves out Indonesia and Vietnam, two of the region’s fastest-growing economies.

Chisham signalled that this will change, at least for one of them. “Markets such as Vietnam are of good interest to us. Vietnam is growing rapidly,” he said.

Also Read: Who referees the AI agents? Enterprises aren’t sure, new report shows

Asked which single APAC market he would prioritise for deeper localisation, though, he looked beyond Southeast Asia. “We’ve definitely got a strong focus on India,” he said, pointing to a large existing customer base there, before adding Vietnam and Thailand as “good markets”. Indonesia did not come up.

Can an agent learn a Malaysian ledger?

SuiteExpress, NetSuite’s new AI-driven implementation tool, promises to move companies onto the platform in weeks. In Southeast Asia, however, the slow part of an ERP rollout is rarely the core setup. It is the local plumbing: statutory configuration, bank payment formats and tax mandates.

Chisham argued that the underlying technology does not change by country. SuiteExpress draws on both a NetSuite knowledge base and the company’s accumulated experience of implementing customers in markets such as Thailand, Malaysia and Vietnam. “The actual technology is the same,” he said. “The AI learns very quickly.”

That is a confident claim. Anyone who has lived through the hidden cost of cheap ERP implementations in the region will want to see it proven on a messy Malaysian subsidiary, not just a tidy US one. SuiteExpress is currently available only in the US.

There was firmer news on a long-standing technical irritant. NetSuite’s own documentation has long warned that the Southeast Asia localisation package does not work alongside SuiteTax, the company’s newer tax engine. Chisham said that work is “largely complete”. The package is becoming hybrid, so it will support both SuiteTax and the legacy tax set-up that, he acknowledged, the majority of customers still use.

“There’s still some work to do,” he said, promising delivery “in due course”. It is good progress, but not yet a ship date.

Who pays when the agent gets it wrong?

With SuiteAgents and SuiteSkills, AI will increasingly create and post transactions on its own. So who carries the liability when an agent issues an invoice that fails MyInvois validation or breaches InvoiceNow rules?

Chisham called it “a pretty good question”. Then, like most of the industry, he answered a slightly different one. Rather than assigning responsibility, NetSuite wants to stop errors before they happen.

“What we want to do is help our customers make sure that their source transactions are correct before they go anywhere else,” he said, describing validation as a strong use case for AI because the rules themselves are deterministic.

The model for this comes from Latin America. In Mexico, where every electronic invoice is validated through the tax authority’s system, NetSuite has shipped capability that monitors transactions and checks them “before they even leave NetSuite”. Chisham said the same approach will be extended across the suite.

Prevention is sensible. It is not the same thing as accountability, a gap enterprises across the region are still struggling to close.

“It is a mix”: where your data goes

Indonesia, Vietnam and India all impose data localisation rules of varying strictness. When an APAC customer asks NetSuite’s AI a question, where is that data processed?

Also Read: How a 39-store retail chain in Laos moved from Excel to AI agents

Chisham’s answer was more honest than tidy. Ask Oracle, NetSuite’s conversational layer, uses elements of OpenAI’s models. “What we do is leverage the OpenAI instance that is closest,” he said.

The closest instance, however, is not necessarily an in-country one. For NetSuite Next, Chisham said the company carries out “a full legal and compliance validation” before moving forward, to make sure it works within the confines of data residency and data processing rules.

He returned to the subject unprompted at the end of the interview, offering to follow up with specifics “because it is a mix” across the different models NetSuite uses. For firms already anxious about data sovereignty, that candour is welcome. A clear, country-by-country answer would be more welcome still.

Human in the loop, all the way down to the SQL

This year’s SuiteWorld theme is the “autonomous suite”. NetSuite’s AI team has also long insisted that keeping a human in the loop is non-negotiable. How do the two fit together?

Chisham’s answer was the most concrete of the interview. NetSuite is building around auditability, and Agent Studio, its agent-building environment, is “built on the core premise” of traceability.

He offered an example. A user asks for the revenue figure for a Malaysian subsidiary. They can then interrogate exactly how the AI arrived at it: the logic the model applied, and even the SQL query and the database tables the number came from. Agent actions are logged the same way, with a full trail and chat history.

If a number looks wrong, the user can step in and tell the model which logic to apply instead. “Whilst we can automate many tasks, we always allow an entry point for the user to make a different decision,” Chisham said.

That level of traceability is exactly what regulated sectors such as banking demand before handing AI agents real work. It also echoes the lesson of last year’s event, that context, not automation, is the real AI prize.

Listening, mostly to the regulator

Asked to name one APAC customer conversation that changed his roadmap, Chisham at first fell back on a familiar line: the roadmap “is really driven by customers”. He then offered a specific example. Direct feedback from customers in Singapore and Malaysia on the e-invoicing user experience led NetSuite to go back and simplify the capability.

Also Read: Banks want AI agents to move money — as long as someone is accountable

His group, he admitted, is driven “heavily by compliance, because there’s so much of it” in Southeast Asia and Latin America. That may be the most revealing line of the conversation. In this region, NetSuite’s product roadmap is written as much by tax authorities as by founders.

For startups weighing their growing list of AI accounting and ERP options, Chisham’s pitch is coherent: one system, many mandates and an audit trail down to the query. The proof will come when SuiteExpress lands outside the US, the SuiteTax fix ships, and someone can explain, country by country, exactly where their data goes.

The post NetSuite’s James Chisham on SEA’s e-invoicing maze, data residency and who pays when AI gets it wrong appeared first on e27.

Most conferences are built the same way. A committee brainstorms a theme, a team chases speakers, a designer lays out a program, and everyone hopes the real value happens in the corridors. Three days in Dubai this October worked differently.

The operating model was deliberately lean: two people shaping the ideas and running the room, and two part-time organisers steering the AI that handled the rest. Phil Midland and Dr Alex Lin of Strategic Renaissance 21 (SR21) designed and facilitated the workshop with the World Government Summit, while Khor Zijian and Lin Fanghong piloted the AI.

The result was not a faster version of a conventional conference. It was a different kind of event.

Finding the theme: “The Breaking Point”

The theme was derived, not brainstormed. Before naming a single session, AI worked through a large body of material: think-tank reports, academic papers, policy articles, market commentary, and news analysis on geopolitics, energy, capital flows, financial trust, and artificial intelligence. The question was what the world’s strategic conversation was converging on.

The signal was consistent. Across domains that rarely share a stage, analysts were describing systems under cumulative stress: supply corridors, debt and reserve-currency dynamics, trust in banks and institutions, and the speed at which AI was entering all of them. The shared question became the event’s framing: how much stress can the complex world system absorb before it breaks, and what does rebuilding look like?

AI surfaced the pattern, and the humans decided it was worth building a conference around. That division of labor held throughout.

Also Read: I used AI to rebuild my websites without knowing how to code

Sessions built around the people in the room

A good theme is useless if the room can’t speak to it. The second job for AI was turning “The Breaking Point” into sub-themes that matched what the confirmed participants actually knew.

The system took each attendee’s profile, publications, and professional background and mapped them against candidate angles. The program settled into four tracks:

• Track A: kinetic conflict and hard power
• Track B: capital and energy
• Track C: trust, in banking and human systems
• Track D: AI as multiplier and disruptor

The regional focus was the Middle East, North Africa, and the Indian Ocean. Each day moved from situational awareness to rebuilding. As the attendee list shifted, AI re-tested the track assignments against who was actually coming. Matching speakers to sessions by hand is slow and tends to default to reputation. Matching against demonstrated expertise produced sharper sessions.

It also fit the kind of room SR21 convenes. As SR21 Chairman Dr. Howard Morgan said:

“Every year this group gets a little better at seeing structural change before it becomes conventional wisdom. This year we’re pushing that a step further — turning what we see into commitments the room can actually be held to.”

We built the program to give that ambition somewhere to go.

Preparing the room before the room existed

The usual weakness of a high-level gathering is that participants arrive cold and spend the first day getting oriented. We remove that.

For each track, AI assembled a reference reading list, then synthesised it into a concise thematic paper and a podcast-style audio briefing. Participants could read at their desks or listen on the way to the airport. The goal was a shared baseline of facts and framings, so that conversation could start the minute the group convened.

Each paper also included proposed discussion points and prompts for facilitators. These were designed to cover every angle of a topic, including the uncomfortable ones, and to draw out insights from both the speakers and the audience. A quiet expert in the third row is often holding the most useful observation in the room. A well-built prompt gives the facilitator a reason to go and get it.

The papers were also built backward from the final day’s goal, which I’ll come to shortly. Each track’s reading mapped to the practical measures participants would draft on day three, so the preparation fed directly into the output.

The logistics: The least significant AI contribution

It’s worth being honest about this part, because most people assume AI’s value lies here. AI did the manual compilation: assembling the pre-conference materials into a single conference booklet, going through multiple versions as the program, venue, and schedule changed, and drafting press releases with quotes from relevant personalities.

This saved real hours, and it let two part-time organisers keep up with a moving target. But it was probably the least significant thing AI did. Formatting a booklet is a task any competent assistant could do. The real value was upstream, in how the content was chosen and prepared, and downstream, in what was extracted from the proceedings.

Also Read: How AI became my sparring partner

In the room: Recording everything, analysing beyond the words

From 5 to 7 October at Emirates Towers, the proceedings were recorded for immediate analysis. The workshop brought together 48 senior leaders, including retired flag officers, sitting and former government ministers, sovereign-wealth and private-capital principals, technologists, and academics, all under the Chatham House Rule. Ministers from the UAE government attended, including H.E. Omar Sultan Al Olama, Minister of State for Artificial Intelligence, Digital Economy, and Remote Work Applications, and his deputy.

Conventional session summaries record who said what. Ours were built to capture what the conversation meant, and AI let us go well past the spoken word:

• Bias and opinion identification. We separated statements of fact from statements of position, and flagged the framing behind each.
• Data error citation. When a speaker quoted a figure that didn’t match the published record, the system noted the discrepancy and cited the correct source, rather than letting it pass into the summary as truth.
• Emotional and persuasive language. AI flagged loaded or persuasive wording, showing where the heat in a discussion was and where conviction outran evidence.
• External trajectory data. We cross-referenced on-stage discussions with prediction markets like Polymarket and Google Trends to see where public attention and market probabilities were heading. A topic was no longer just “discussed”. It came with a trajectory.

No human team could do this at the speed a conference demands. A person can listen to a panel, but cannot simultaneously fact-check every figure, score the rhetoric, pull live market data on each topic raised, and produce a structured summary within hours. The AI did the heavy lifting, and the participants got insights they couldn’t have extracted themselves.

Humans stayed where they add the most value: choosing which questions mattered, judging which signals were real, and deciding what the findings meant. AI extracted. People decided.

Day three: From diagnosis to Trust and Confidence Building Measures

A conference that only diagnoses a problem leaves people anxious and no better equipped. So the final day focused on practical Trust and Confidence Building Measures (TCBMs), concrete mechanisms to help the world be better prepared before it reaches the breaking point.

The drafting session drew on the tracks’ reference readings and insights from the previous two days. Because the workshop ran under the Chatham House Rule, I won’t detail the individual measures. In general terms, they covered AI system governance, the security of critical trade corridors, human oversight of automated decision-making, verification of de-escalation steps, and the interplay of energy, capital, and currency stability.

Also Read: Who referees the AI agents? Enterprises aren’t sure, new report shows

Because the earlier work was systematically prepared and captured, participants didn’t start day three from scratch. The system also summarised the discussion into actionable insights and draft plans for moving forward. As a result, local and international participants alike left with drafted language, shared frameworks, and a clearer sense of who to talk to next. They left with tools and ideas, and were better prepared for what’s next.

What this means for event organisers

For Asia’s startup and venture community, which attends and runs more conferences than almost anyone, a few lessons stand out:

• Use AI upstream, not just at the admin layer. Theme derivation and session design are where it changes outcomes. Booklet assembly is a convenience.
• Prepare participants as seriously as you program speakers. Synthesised readings, audio briefings, and facilitator prompts raise the floor of every conversation.
• Treat the recording as raw material. An event’s most valuable output is often the analysis of what was said, not the footage.
• Keep humans on judgment. The team’s small size was a feature. Two part-timers could pilot the AI, and two facilitators could run the room, because the AI handled breadth.

What comes next

The engine behind the in-room analysis is a process we call Machine Augmented Insights Extraction (MAIE). In my next article, I’ll walk through the detailed, AI-assisted flow of MAIE: how a raw conversation becomes a structured, fact-checked, trajectory-aware set of insights, and how you can apply it to your own meetings, boards and events.

The Dubai workshop made one thing clear to me. A conference is no longer just a meetup and a networking event. It can be a deeper learning experience, one that gives participants far more than a presentation on stage can.

—

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 Two part-timers and one AI orchestrator: Inside a Dubai conference on “The Breaking Point” — Part 1 appeared first on e27.

Bitcoin slipped 0.97 per cent to US$79,912.85 over the last 24 hours. That number alone looks mild for an asset known for violent swings. The total crypto market cap fell 0.72 per cent to US$2.68 trillion. Bitcoin’s seven-day correlation with the S&P 500 reached 83.4 per cent. That correlation explains the move better than the price change. This was not a crypto-specific breakdown. It was a risk-asset repricing shaped by fragile positioning and tougher financial conditions.

Institutional flows supplied the first push. US spot Bitcoin ETFs lost about US$635.5 million on October 22 and another US$137.9 million on October 23. Combined outflows reached roughly US$773.5 million over two days. That reversal followed prior-week inflows. The speed of the turn challenges the idea that ETF money is steady and structural. It behaves tactically. It chases momentum and exits when fear rises. This flow reversal added direct selling into an already nervous tape.

Leverage amplified the decline. Leveraged positions fell roughly US$815 million to US$837 million in 24 hours. Bullish traders took most of the damage. Total derivatives open interest fell 8.6 per cent, which confirms broad deleveraging rather than an isolated spot sale. Stop losses and forced sales fed the drop. The flush removes speculative weight, but it also shows how crowded longs had become. A market cannot build a durable rally on a foundation of stretched positioning.

Geopolitical risk added another layer. Weekend headlines reported that President Trump said the US could strike Iran as soon as Sunday. Bitcoin trades around the clock and absorbed early selling before traditional equity markets opened. This is a structural disadvantage in a crisis. Investors seeking protection do not treat Bitcoin as a hedge. It trades like a risk asset. The 83.4 per cent correlation with the S&P 500 reinforces that role. Short-term behaviour is currently indistinguishable from equities, and traditional equity futures were also weak.

Also Read: Bitcoin rejected US$84,000 and forced US$79,060,000 in long positions out

The rates backdrop made the selling easier. The 10-year Treasury yield remains elevated near 5.3 per cent, and September CPI came in at 3.7 per cent year over year. Sticky inflation and high yields reduce demand for non-yielding volatile assets. Crypto sits at the far end of the risk curve, so it feels that strain more acutely than most sectors. The Federal Reserve’s October 28 decision is the key near-term catalyst. A hawkish tone would extend the pressure and make a test of US$77,600 more likely. A dovish shift, or cooling inflation data on October 29, could stabilise prices and give the asset class room to breathe. Traders are not waiting for a crypto-native catalyst. They are waiting for an economic signal.

The technical picture tells a similar story. Bitcoin trades below the US$80,000 to US$81,500 resistance zone, which aligns with the estimated cost basis for ETF and corporate treasury holders. A sustained failure to reclaim that zone keeps the bias bearish and puts US$77,600 in play, a level Glassnode identifies as the True Market Mean. On the total market cap chart, US$2.68 trillion is the immediate pivot. A hold above that level could open a rebound toward US$2.74 trillion. A break below risks testing the US$2.6 trillion to US$2.64 trillion range. These are not arbitrary lines. They are levels where holders have shown conviction before, and where the next battle will take place.

Also Read: 92% of Bitcoin liquidations were shorts: What happens when the squeeze ends?

My point of view is that this moment is a healthy but uncomfortable reset. It exposes two weaknesses that the bullish narrative prefers to ignore. The first is that institutional demand through ETFs is far more fragile than the adoption story admits. The second is that leverage remains a permanent feature of this asset class, and it turns ordinary corrections into cascade events. Neither weakness is fatal. Both argue for humility about how quickly Bitcoin can decouple from the broader financial system. The correlation with the S&P 500 is not a temporary anomaly. It is the current regime.

The near-term outlook is bearish below US$80,000, but that bearishness is conditional. The Fed could change the mood in a single statement. The daily close relative to US$80,000 and the policy language on October 28 will decide whether this is a brief shakeout or the start of a deeper move toward US$77,600. My view is that the next leg will come from policymakers, not crypto developers, ETF issuers, or on-chain metrics. Bitcoin’s short-term fate is not in its own hands. This sell-off reveals that uncomfortable truth with unusual clarity.

—

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 Bitcoin’s macro reset: Reclaim US$80,000 before the Fed speaks? appeared first on e27.

Singapore has a power problem that no amount of rooftop solar will fix. The city-state generates about 95 per cent of its electricity from imported natural gas. It has little spare land for solar farms, no rivers for hydropower and not much wind. Every decarbonisation plan it writes is, in effect, a plan to import cleaner molecules or squeeze more energy out of very little space.

The latest instalment arrived on Monday. The Energy Market Authority (EMA) and the Advanced Energy and Decarbonisation Office (AEDO) launched a SGD75 million (US$59 million) Power Decarbonisation Funding Initiative to develop low-carbon technologies for the grid. It sits under the SGD800 million (US$627 million) Decarbonisation Grand Challenge, part of the Research, Innovation and Enterprise (RIE) 2030 plan announced in March.

Also Read: AI is about to blow a hole in ASEAN’s climate targets, unless baseload goes clean

The money will fund laboratory research on early-stage technologies as well as test-bedding and pilots of more mature ones. It targets three areas: hydrogen and its derivatives, solar and biomass.

Hydrogen gets the hardest targets

Hydrogen carries the most demanding goals. The first is to let power plants switch from hydrogen-based fuels to backup fuels within 10 minutes while keeping combustion stable. A turbine burning hydrogen or ammonia cannot simply sputter when a shipment runs late. Grid operators need a fallback that kicks in before the lights go out.

The second is to shrink the health and safety buffer required around hydrogen facilities with a capacity of 1 million tonnes per annum of hydrogen equivalent. On an island where industrial land is rationed by the hectare, a smaller exclusion zone is a direct saving.

The third goal is the boldest. The initiative wants to bring the cost of extracting hydrogen to below a dollar per kilogram, through better energy efficiency, longer-lasting catalysts and revenue from valuable co-products. Singapore expects to import hydrogen in carrier forms such as ammonia rather than produce it at home. That makes the “cracking” step, which releases hydrogen from its carrier, the place where costs pile up.

The hardware is already arriving. Last Friday, Singapore broke ground on a SGD62.5 million (US$49 million) low-carbon testbed on Jurong Island, where Amogy will pilot ammonia cracking at up to one tonne of hydrogen a day. Whether cracking can ever get under a dollar a kilogram at commercial scale remains one of the energy transition’s open questions.

Solar on walls, not fields

With most rooftops and reservoirs spoken for, the solar targets focus on unconventional surfaces. The initiative wants structural costs cut by at least 15 per cent without compromising strength or waterproofing. This is the unglamorous engineering that decides whether panels end up on facades, sea walls and car parks, or stay in brochures.

It is also chasing next-generation photovoltaic modules with efficiency above 30 per cent, or lightweight and flexible designs. Either kind must survive Singapore’s heat and humidity with annual degradation below 1 per cent over at least 25 years.

Also Read: Touchstone partially exits solar startup Stride after US$15M Series B

That is a tall order. Mainstream commercial silicon panels convert roughly 22 to 24 per cent of sunlight. Anything above 30 per cent today generally means tandem cells, which layer silicon with materials such as perovskite. Those cells remain largely lab-scale and are notoriously sensitive to moisture. Asking them to last a quarter of a century in the tropics is asking a lot, which is presumably why the government is paying for research rather than procurement.

The ash nobody wants

The biomass goal is narrow but practical. The initiative wants low-cost technologies that turn ash from different biomass feedstocks into useful products, at a processing cost below SGD50 (US$39) per tonne. Burning wood pellets or agricultural waste leaves residue. Singapore’s only landfill, Semakau, is projected to fill up by around 2035, so the country has little room to dump it.

Regional startups are already working on adjacent problems. Singapore’s Verta Bioenergy turns farm waste into coal-ready fuel for industrial boilers. Some investors argue that the region’s next climate unicorn may be built from agricultural residue. Ash is the less glamorous end of that value chain, but someone has to deal with it.

Read the DRIVE fine print

For founders, the most relevant part is the Decarbonisation Research for Industry, Ventures, and Enterprise (DRIVE) programme, which opened for applications on 23 October. SGD25 million (US$20 million) of the power initiative is reserved for industry-designed projects.

DRIVE flips the usual grant logic. Companies, including startups and technology providers, define the decarbonisation problem they want solved. They then work with Singapore’s public universities and research institutions to develop and co-fund solutions. The government covers up to 75 per cent of qualifying project costs, but that money goes to the public research performers, not to the company. The company puts in its own share.

That makes DRIVE less a startup grant than a subsidised research contract. For a well-capitalised corporate, it is a cheap way to rent a university lab. For a seed-stage climate founder, funding even a quarter of a multi-year research budget is a stretch. Climate capital promised to the region has been slow to actually arrive, and among Southeast Asia’s climate startups the cheques remain stubbornly small.

Hardware founders face the steepest climb. Climate finance has long struggled to back companies building in atoms rather than code. Those are precisely the companies a hydrogen-cracking or solar-materials challenge would attract.

A crowded week of announcements

The power initiative did not arrive alone. In the space of four days, Singapore also announced:

• a SGD143 million (US$112 million) industrial decarbonisation fund
• the Jurong Island testbed
• a renewed energy cooperation pact with Malaysia covering cross-border electricity trading and renewable energy certificates

The overlap can be hard to track. The Economic Development Board says up to S$100 million of the industrial fund is also earmarked for DRIVE. DRIVE therefore looks like an umbrella programme with separate pots for power and industry. Founders will want clarity on which pot their project falls into before they write a proposal.

Also Read: Funded: SEA climate tech has US$1.1B and a problem no one wants to name

None of the targets comes with a deadline, and no baseline has been published against which progress can be measured. A sub-dollar hydrogen cost or a 30 per cent tropical solar panel is a worthy ambition. Without dates, it is difficult to hold anyone to it.

There is also a question of proportion. SGD75 million spread across turbine controls, safety engineering, catalyst chemistry, solar materials and ash recycling is a research budget, not an industrial policy. Singapore’s real decarbonisation bets remain imported electricity and imported hydrogen. Home-grown research buys option value, a hedge in case those imports prove slower or costlier than planned.

For startups, the opportunity is real but conditional. Those with a corporate partner, a clear technical problem and a quarter of the bill in hand will find DRIVE useful. Programmes such as Startupbootcamp’s Singapore sustainability cohort suggest the founder pipeline exists. Everyone else may find that Singapore has built a well-equipped lab, with an entry fee attached.

The post Singapore’s US$59M grid fund bets on cheaper hydrogen and tougher tropical solar panels appeared first on e27.

When a chip plant is running close to full tilt, a failing chiller or a slow gas leak is not a maintenance ticket. It is lost output that cannot easily be made up elsewhere.

That is the problem GlobalFoundries (GF) says it is tackling in Singapore. The New York-headquartered contract chipmaker has opened an Advanced Facilities Control Centre (AFCC) at its Singapore site, developed with German industrial giant Siemens.

Also Read: The quiet layer keeping the chip boom alive

In a statement on Monday, GF said the centre is the first deployment of a system meant to integrate facilities operations and data across its global manufacturing network. Singapore is the first site to get it.

What the control room actually does

In a fab, “facilities” means the life-support systems around the production tools: power, cooling, ultrapure water, specialty gases and the filtered air that keeps cleanrooms clean. If any of them falters, the wafers inside are at risk.

In the first phase, five fab-specific control centres across GlobalFoundries’s Singapore campus will operate as a single command centre. The system pulls facilities data and operational metrics into one view, using monitoring and visualisation tools and AI features such as predictive maintenance, which flags equipment likely to fail before it does.

GF says this helps teams troubleshoot faster, find root causes sooner and decide which problems to fix first. Once fully deployed, the system can integrate more than one million critical data points from facilities equipment. Round-the-clock equipment health monitoring is meant to cut unplanned downtime, and metrics can be compared across sites to improve performance and sustainability.

The centre also gets mobile eyes. Autonomous robots carrying thermal imaging cameras, gas detectors and vibration and acoustic sensors will patrol the fab floor and campus, streaming real-time data back for analysis.

Why Singapore, and why now

Singapore is no side project for GlobalFoundries. The company opened a US$4 billion expansion fab there in September 2023, lifting its local capacity to about 1.5 million 300mm wafers a year. In November 2025, it acquired local silicon photonics foundry Advanced Micro Foundry (AMF). Silicon photonics uses light rather than electrical signals to move data between chips, a technology AI data centres increasingly need.

Timing matters here. DIGITIMES reported last month that GlobalFoundries’s Singapore capacity utilisation had moved above 90 per cent, and that equipment lead times were constraining its photonics expansion. When new tools take months to arrive, squeezing more uptime out of existing plant becomes the cheapest capacity a chipmaker can buy. It fits the wider pattern of AI’s bottleneck economy shifting from GPUs to the less glamorous parts of the supply chain.

Also Read: Applied’s US$500M Singapore expansion tightens ties with Southeast Asia’s chip push

Yew Kong Tan, GF’s senior vice president and general manager for Asia-Pacific manufacturing and the Singapore site, said the city-state was the natural starting point because of its digital foundation and government and ecosystem support. Cindy Koh, executive vice president of the Singapore Economic Development Board (EDB), framed the centre as an example of Singapore’s collaborative approach to manufacturing, particularly in advanced packaging and silicon photonics.

Those are precisely the niches Singapore has been branding hard. Last month, A*STAR and EDB launched SG Semiconductor as a national identity for the sector. Chipmakers across the region are also expanding their Asia Pacific footprint to meet AI demand.

The numbers the announcement leaves out

For all the talk of one million data points, GlobalFoundries’s statement contains few numbers that matter to an outsider. It gives no target for reduced downtime, no estimate of energy or emissions savings and no figure for what the centre cost.

The workforce question is also handled delicately. GF says centralisation will optimise the use of “space, equipment and staff” and cut duplicated infrastructure. Tan said the tools would free facilities engineers to focus on higher-value diagnostics and problem-solving. That may well be true. But when five control rooms become one, fewer people usually end up watching the screens, and GlobalFoundries has not said what happens to the rest.

There is a security trade-off too. Folding a campus’s critical infrastructure data and robot feeds into a single command centre makes operations easier to see. It also gives attackers one high-value target. Chip suppliers in the region are already under rising pressure to prove cyber resilience before customers will sign.

Then there is sustainability. Fabs are notoriously water- and energy-hungry, so lower energy use and emissions is a welcome promise. Without a baseline or a target, it is hard to hold GF to it.

Pradip Singh, GF’s chief manufacturing officer, called reliable facilities infrastructure foundational to chipmaking and described the centre as a step towards a more connected, intelligent manufacturing network.

A crowded race to run the smart fab

GlobalFoundries is hardly alone. TSMC has long promoted AI-driven “intelligent manufacturing” at its fabs. In Singapore itself, UMC runs its Fab12i and VisionPower Semiconductor Manufacturing Company, the Vanguard-NXP joint venture, is building a new plant. Each will be chasing the same uptime gains.

On the software side, the deal is a showcase for Siemens in a market where rivals are spending heavily. Schneider Electric agreed earlier this month to acquire industrial software firm PTC for about US$22.6 billion, and players such as Honeywell and ABB also sell plant monitoring and automation. A template that GF rolls out across its network would make Siemens a reference supplier for a global foundry.

There is room for smaller players too. Startups selling vision AI to reduce downtime or sensor analytics for industrial plants will see a big chipmaker validating the category.

What to watch

Singapore’s pitch to chipmakers was never cheap land or cheap labour. It is being the place where the next way of running a fab gets tried first. If the AFCC works, the model will travel to GF’s sites in New York, Vermont and Dresden, and Singapore will have exported a blueprint rather than just wafers.

Also Read: Synopsys, A*STAR team up to tackle AI chip packaging challenges

The real test comes in a year. If GF can then publish hard figures on downtime avoided and energy saved, the control room will have earned its billing. If it cannot, the robots patrolling the fab floor will look more like a photo opportunity than a productivity story.

The post GlobalFoundries turns its Singapore fabs into a test bed for an AI-run control room appeared first on e27.

The most telling line in Publicis’ announcement of a new AI Development Hub in Singapore isn’t about technology. It’s the promise that the hub will create roles rather than remove them. The hub itself is a serious bet: engineers, data scientists and product designers working alongside strategists to build proprietary marketing technology in Singapore and scale it outward. But the reflex behind the promise is the oldest one in the agency business. When in doubt, hire.

That reflex explains why the biggest groups overbuilt headcount for a decade while the economics were asking them to build product.

Headcount was the product

Agencies sell capacity. Revenue follows people, margin is measured as EBITDA, and the metric rewards a busy bench, so hiring ahead of the next pitch is not a lapse inside that model. It is the model.

The bill arrives when clients stop paying for capacity. Investing.com’s snapshot of the sector puts digital ad spend growth at about 8.7 per cent in 2025 against agency revenue growth near zero, as clients protected media budgets and squeezed fees, and shows Omnicom heading from roughly 120,000 staff to about 105,000 by year-end. Per the Financial Times, relayed by Reuters, WPP cut about 11,000 jobs since the start of 2025, leaving 97,388 people by June.

What the market pays for it

Public agency groups traded at a median 0.8x revenue and 6.6x EBITDA at the end of 2025, according to Canaccord’s data as compiled by Axia. Independent agencies typically sell for 3-7x EBITDA, with strategic outliers reaching 8-12x. Martech M&A tells a different story: a median of 3.5x revenue in 2026 so far, and 8.0x for AI-native companies.

Line up the revenue multiples and a dollar of martech revenue is priced at more than four times a dollar of agency revenue, and AI-native martech at about ten times. Different samples, so read it as direction rather than arithmetic.

Also Read: Ecosystem Roundup: Taiwan blocks Grab’s US$600M foodpanda deal over Uber’s stake

The top of the ladder sits higher still. ValueAdd VC’s October tracker puts the median multiple for nine private AI companies at 27.1x annualised revenue, against 4.3x for public SaaS. In marketing, Clay raised at a US$7.1B valuation in September on roughly US$200M of run-rate revenue, about 35x, while non-AI martech such as Seismic and AppsFlyer sit near 12x and 8x on Multiples.vc. Private marks on annualised figures are soft, but the pattern holds: the premium follows AI-native revenue, not the martech label.

The sharpest illustration sits inside the holding companies. WPP’s 14.8x EV/EBITDA is the highest of the big three, yet it is a symptom, not a premium: its EBITDA fell from US$1.96 billion in 2021 to US$752 million in FY2025, so the denominator collapsed. Publicis, the one still growing, sits at 9.6x. A model measured on EBITDA gets punished twice when it shrinks.

Two people, one payroll, opposite incentives

An agency owner is paid on EBITDA and is a capacity manager by training: utilisation, a full bench, a retainer renewed. A buyer then discounts that earnings stream for founder dependency and client concentration, which is why most independents land well below public-market multiples.

A martech CEO is paid on a different question: how much more revenue can this customer base produce without adding cost? Net revenue retention, gross margin and growth set the price. Public software with retention cohorts above 120 per cent traded at 11.7x revenue against 4.1x below 100 per cent, in SEG data compiled by Array Capital. Levels have compressed since; the ratio is the point.

So the owner reads a new hire as growth. The CEO reads it as a cost to be spread across more customers. Same payroll, opposite incentive. That gap, more than any technology gap, is why transformation stalls: moving to a product model means treating your own headcount as the thing to decouple from revenue.

What would make an agency price like product

Publicis shows the discipline is possible. It keeps headcount discipline as a rule, holding headcount growth below organic revenue growth, with about 114,000 people and roughly 5,100 net hires in 2025. WPP put more than £300 million into its WPP Open platform in 2025 while shrinking staff. Those are product bets funded by payroll cuts, which is not the same as a product business.

Also Read: Who referees the AI agents? Enterprises aren’t sure, new report shows

Three tests separate the two:

• Revenue that doesn’t rise with hours: licences, usage or outcome fees.
• Retention you can show cohort by cohort, not client anecdotes.
• A margin story that survives automation, with headcount growing slower than revenue.

Whether the Singapore hub passes depends on whether its output is licensed as software or absorbed into fees. A lab that makes the service faster still sells time; software others can license opens a different line on the P&L.

Why Southeast Asia feels it first

A joint survey by Singapore’s Economic Development Board and McKinsey, cited in the Publicis announcement, found 73 per cent of Southeast Asian companies piloting or scaling AI against 57 per cent globally. f clients are moving that fast, agencies still pricing by the hour risk falling behind the businesses they serve.

A 50-person independent can’t fund an R&D hub and shouldn’t try. It can stop selling hours where it doesn’t have to, and the research and pitch layers are the first place to start.

The question underneath

Among the independent agencies I work with, I’ve seen research and pitch work move from payroll to a repeatable product. One 50-person Singapore agency cut brand analysis from 10 hours to one and client-reporting costs from US$10,000 to US$6,000 a month. Another went from a three-person proposal team to one and cut cost per proposal by 30 per cent.

Owners who want a product multiple have to be seen thinking like product companies, which starts with a steady voice about what they’ve automated in their own shop and what it returned.

So here is the question worth sitting with: if you stopped hiring for twelve months, how much of your revenue would still 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 Agencies don’t have a hiring problem. They have the (EBITDA) multiple problem appeared first on e27.