The cryptocurrency market currently exhibits profound signs of structural weakness as we navigate the middle of 2026. Bitcoin cycles have historically experienced massive drawdowns from their respective peaks. Previous bear markets routinely erased between 60 per cent and 80 per cent of the total market value. This specific cycle reached its absolute peak around the US$126,000 mark in October 2025.

Applying a standard 65 per cent drawdown to that peak places the potential bottom precisely in the US$44,100 range. We must look at the historical precedent to understand this trajectory. The 2017 peak experienced an 85 per cent decline. The 2021 peak suffered a 75 per cent correction. The data clearly points toward diminishing percentage drawdowns with each successive cycle. A 65 per cent drop fits perfectly within this established mathematical pattern and aligns with a much deeper correction than most retail participants currently anticipate.

I view Bitcoin fundamentally as a tech stock plus. The entire tech sector currently operates under the direct influence of the AI narrative. When the AI sector experiences a downturn, the entire tech complex follows suit. Consequently, Bitcoin will inevitably dip severely when the underlying tech leaders falter. We witnessed this exact correlation materialise in early June 2026 when AI memory chip stocks took a massive hit overnight. The sell-off began on June 5 and continued with extreme volatility tracking into the second week of June. This single session erased over US$1.3 trillion in market value from the semiconductor sector alone. The sheer scale of this capital destruction underscores the fragility of the current tech rally and its direct impact on digital asset pricing.

The initial trigger for this massive tech slump originated from Broadcom reporting its Q2 2026 earnings. The company revealed that its AI networking revenue missed analyst expectations. This disappointment occurred despite the revenue growing an impressive 143 per cent year over year.

Also Read: SeaX Ventures backs Onibi in strategic funding round to accelerate AI-powered RPG across Asia

The market reacted violently to this slight miss because investors had priced in absolute perfection. Major memory manufacturers subsequently experienced severe declines. SK Hynix dropped 7.5 per cent on June 10. Samsung Electronics fell 6.1 per cent on the exact same day. Micron Technology faced the most brutal punishment. The stock experienced extreme volatility and dropped roughly 17 per cent over just two sessions following the initial negative news. The Philadelphia Semiconductor Index suffered a major single-session drop in many years. The index fell about 10 per cent in a single day, with analysts citing extreme valuation sensitivity and crowded trades as the primary reasons for the violent correction.

Tech stocks continued their downward slide into June 10 and June 11. Asian chip stocks and various AI memory names fell sharply as fears of a massive tech bubble intensified. We must understand why memory stocks took the heaviest punishment during this sell-off. Despite the extraordinarily high demand for AI High Bandwidth Memory, deep concerns emerged regarding a broader memory chip crisis.

Industry reports highlighted significant inventory buildups for legacy memory products. Investors also engaged in aggressive profit-taking. After an annual rally that pushed many memory stocks to unprecedented heights, market participants simply took the opportunity to lock in their massive gains. The combination of oversupply fears in legacy products and extreme profit taking created a perfect storm for the memory sector. Market participants recognise that legacy memory products face severe margin compression. This realisation forces institutional funds to reduce their exposure to the entire semiconductor complex. The resulting cascade of sell orders accelerates the downward price momentum across all related technology assets.

Some analysts maintain that the underlying demand fundamentals for artificial intelligence remain entirely robust despite this catastrophic sell-off. They point to continued high levels of infrastructure spending by major hyperscalers as evidence that the long-term thesis remains intact. The market cares more about immediate capital flows than long-term promises.

Also Read: The new founder skill is knowing what not to build

We also face a massive shift in capital allocation as big AI initial public offerings approach the market. SpaceX leads this upcoming wave of massive tech listings. This impending influx of new supply guarantees significant capital rotation from existing technology and crypto assets into these new public market opportunities. The market simply lacks the liquidity to sustain current valuations while simultaneously funding these massive new public debuts. Venture capitalists and retail investors alike will redirect their capital toward these fresh opportunities. This rotation ensures that existing digital assets and mature technology stocks will face persistent selling pressure throughout the remainder of the year. The liquidity drain will fundamentally alter the risk appetite across the entire financial ecosystem.

This macro tech weakness directly explains the current on-chain reality for Bitcoin. For the initial time in this specific cycle, more Bitcoin sits at an unrealised loss than in profit. The network currently holds roughly 10.5 million coins underwater against just 9.8 million coins in the green. This underwater crossover represents a critical technical inflexion point. Bitcoin currently tests its 200-week moving average near the US$61,300 level.

Every time this specific underwater crossover appeared in the past, the price landed deep in a bear market near a major cycle low. The community completely disagrees on the interpretation of this data. Some participants desperately believe a bottom forms right here. Others recognise the historical pattern and prepare for significantly more pain ahead. I look at all these converging data points and see a very clear picture.

The evidence overwhelmingly points away from a simple bottom formation. The market structure indicates we have much more downside to explore before reaching a true generational buying opportunity. We must respect the historical data and prepare for a prolonged period of capital destruction.

—

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 US$1.3T wiped out: AI stock collapse signals Bitcoin’s next leg down? appeared first on e27.

Onibi, the game studio developing the AI-powered open-world multiplayer RPG Tomo: Endless Blue, has closed a strategic funding round led by SeaX Ventures and Pix Capital. The investment will be used to accelerate the title towards a full commercial launch, support studio expansion across Southeast Asia, and fund preparations for an upcoming Alpha release on Kickstarter.

The announcement follows a strong start to the studio’s crowdfunding campaign: Onibi exceeded US$100,000 on Kickstarter within 60 hours of launch, signalling early commercial momentum for the project.

Founded by Benjamin Devienne, Onibi is building what it describes as a new generation of open-world multiplayer games in which proprietary AI models generate unique villages, cultures, non-player characters, dialogue, quests, and stories for each individual player. Tomo: Endless Blue combines that AI-driven world generation with physics-based voxel systems, scalable multiplayer infrastructure, and creator tools.

The studio draws its development team from some of the most commercially successful titles in the industry, including Fortnite, League of Legends, Baldur’s Gate 3, Fall Guys, Grand Theft Auto, and World of Warcraft. The team’s collective experience spans multiplayer systems, live-service games, world-building, player engagement, and scalable game-technology infrastructure.

Also Read: The weavers of Bengal, my mother, and what to tell tomorrow’s graduates

Beyond its game-play ambitions, Onibi is positioning Tomo: Endless Blue as the foundation for a user-generated content platform where players can build their own RPG experiences with AI-assisted tools. The studio’s long-term goal is to allow players to move from a simple prompt to playable content — creating villages, stories, quests, and shared worlds — while reducing the technical barriers that typically separate players from game creation.

“Tomo: Endless Blue is built around infinite replayability: a world that keeps surprising players long after their first adventure,” said Benjamin Devienne, Co-founder and Chief Executive of Onibi. “By combining proprietary AI models, procedural generation, multiplayer systems, and UGC tools, we want every island, village, quest, and player-created experience to feel different. The backing of SeaX Ventures and Pix Capital helps us push that technology further, grow the Kickstarter community, and accelerate our strategy in Asia.”

SeaX Ventures, which led the round, cited Onibi’s combination of production credentials and its approach to AI-native game development as central to its investment thesis. Dr. Kid Parchariyanon, Founder and Managing Partner of SeaX Ventures, said the studio represented a rare convergence of world-class execution experience and original thinking about how games are made.

“AI-native game development is one of the most consequential frontiers in the broader deep-tech wave, and the team’s track record — across some of the most successful multiplayer franchises ever shipped — is exactly the kind of foundation this category requires,” Dr. Parchariyanon said. “We are delighted to back Onibi as it brings Tomo: Endless Blue to a global community of players and creators, with Asia at the centre of that strategy.”

—

Image Credit: Onibi

The post SeaX Ventures backs Onibi in strategic funding round to accelerate AI-powered RPG across Asia appeared first on e27.

Singapore-based IT automation startup ShiftControl has launched a next-generation IT operations platform designed to run natively on Google Workspace, the company announced at Echelon Singapore, the region’s flagship technology conference.

The platform is aimed at small and mid-sized businesses that rely on Google Workspace as their core operating environment but lack the dedicated IT resources to manage employee access across an expanding portfolio of cloud applications.

ShiftControl serves as an orchestration layer on top of Google Workspace, providing administrators with a unified view of every employee, their team assignments, and the applications they can access and actively use. From that dashboard, admins can automatically push or revoke access rights across all connected SaaS tools.

The company says the problem is acute for growing businesses. A typical 60-person company may run more than 40 separate SaaS subscriptions, frequently tracked by hand in spreadsheets and managed by a founder or operations lead rather than a trained administrator. Onboarding, offboarding and periodic access reviews are often manual, creating security exposure when departing employees retain access and wasted spend on licences no one uses.

The platform applies AI to surface unused licences, flag accounts that retain access after an employee leaves, and recommend access changes as roles shift. ShiftControl says businesses most commonly run into these problems as headcount approaches 40 to 50 people or the point at which informal, manual processes tend to break down.

Also Read: The weavers of Bengal, my mother, and what to tell tomorrow’s graduates

Co-founder Dan Gericke said legacy IT tools were not designed for how modern small- and mid-sized businesses operate. “The typical IT stack was designed for a different kind of company. It’s expensive, requires a large IT team to run and maintain, and it’s operating on old technology,” Gericke said.

“For a growing number of small and mid-sized businesses, those tools don’t work for them. They run on Google Workspace, they don’t have a full-blown IT team, and they expect their tools to be AI-native by default. We rebuilt our product for them.”

ShiftControl was founded by two former ExpressVPN executives who said they experienced firsthand the difficulties of scaling a modern tech business with the IT tooling currently available. The company’s stated mission is to make IT operations simple enough for any business to run without specialist staff.

Customers already using the platform include cybersecurity firm Blackpanda, mobility company GetGo and philanthropic organisation The Majurity Trust.

The company, founded in Singapore, now serves customers across London, Hong Kong and North America.

The launch positions ShiftControl within a competitive but growing market for identity and access management tools tailored to smaller organisations. The platform integrates with modern HRIS platforms to align employee lifecycle events with access changes, reducing operational overhead and security risk, a capability increasingly sought by lean operations teams managing IT responsibilities without formal training.

The post ShiftControl launches AI-native IT operations platform built for Google Workspace appeared first on e27.

One of the most misleading moments in AI deployment is when the model sounds exactly as it should.

It uses careful language. It gives balanced caveats. It avoids prohibited phrasing. It appears measured, compliant, and responsible. The tone feels safe enough for internal rollout and polished enough for senior stakeholders to relax. At that point, many organisations conclude that the safety question is largely under control.

That is often where the real danger begins.

A model can produce the right answer in form while arriving there through the wrong internal logic. It can sound cautious without being grounded. It can refuse in the right places for superficial pattern reasons rather than because the system is reliably distinguishing safe from unsafe use. It can generate a persuasive explanation that resembles judgment without containing much of it. From the outside, the output looks safe. In practice, the organisation may be mistaking behavioural polish for actual control.

This is the illusion of safety. It appears when institutions start reading surface alignment as structural alignment. That distinction matters more than most current deployment models admit.

Safety is not the same as acceptable language

A great deal of current AI governance still treats safety as an output problem. If the model does not produce certain kinds of harmful content, if it uses appropriate tone, if it adds the right warnings, if it avoids obvious policy breaches, then the system begins to look governable.

That view is too shallow.

Safety is not only about what the model says. It is about whether the model’s behaviour remains dependable when context becomes messy, incentives become conflicting, or users push into edge cases that were never cleanly anticipated. A model that says the right thing because it has learned the stylistic shape of acceptable answers is very different from a system that behaves reliably because the organisation has designed the surrounding operating conditions well.

The problem is that these two states can look very similar at the output layer.

The wrong reason can still produce the right answer

Large language models do not need stable, principled internal reasoning in order to produce text that appears careful, intelligent, or safe. They can arrive at a good-looking answer by patterning against the language of caution, policy, balance, or refusal. That does not mean the behaviour will remain reliable when the context shifts. It only means the model has learned what a safe response usually sounds like.

Also Read: Red team with red flags: What happens when your LLMs outsmart your safety nets

This matters because organisations tend to judge safety through visible behaviour rather than through causal confidence. If the system regularly produces sensible-sounding outputs, the institution starts treating it as though it is operating on sound judgment. But the output may be the product of linguistic mimicry rather than robust behavioural control.

That gap becomes especially serious in business settings where plausible language is enough to move decisions forward. The model does not need to be correct in a deep sense. It only needs to be convincing enough, measured enough, and internally acceptable enough to reduce challenge.

Once that happens, the organisation is no longer being protected by safety. It is being comforted by style.

The most dangerous model is often the one that knows how to sound governable

There is a reason this problem matters so much in enterprise deployment.

Institutions are not merely asking whether a model is helpful. They are asking whether it can be trusted inside workflows that carry financial, legal, operational, reputational, or customer consequences. In that environment, the model that sounds responsible can become more influential than the model that is merely capable.

This is where an especially subtle failure mode appears.

A model begins to produce the language of governance. It sounds audit-friendly. It sounds risk-aware. It sounds balanced, cautious, and institutionally literate. It includes the sorts of statements compliance teams like seeing and executives find reassuring. But underneath that surface, it may still be working from weak signals, shallow correlations, or brittle pattern recognition that does not survive pressure.

The organisation then makes a serious mistake. It begins to trust not just the output, but the tone of the output as evidence of safety maturity.

That is not control. It is aesthetic reassurance.

Saying the right thing can still mean understanding the wrong thing

When an LLM says the right thing for the wrong reasons, the problem is not simply that the answer might fail later. The problem is that the organisation has very little clarity on what the model is actually tracking when it behaves well. Is it recognising a real safety boundary? Is it following a pattern that resembles safe language? Is it responding to token cues that happen to correlate with good outputs in training? Is it generating a plausible refusal while still leaving the dangerous intent intact in another form?

These are different conditions, and they matter enormously once the system is placed inside real institutions.

A company cannot build serious governance around mere output resemblance. It needs some confidence that the system’s behaviour is stable across reformulation, sequence effects, contextual pressure, and adjacent use cases. If that confidence does not exist, then what looks like safe behaviour may only be a temporary correlation.

Also Read: Psychological safety and the art of purging

The sharper failure is not misinformation — it is misplaced confidence

There is a tendency to describe LLM risk mainly in terms of false content. Hallucinations, fabricated claims, wrong facts, misleading advice. Those matters, but for many organisations, the more serious issue is confidence distortion.

A model that sounds careful can alter the organisation’s confidence in a decision even when the underlying reasoning is weak. It can make incomplete work appear complete. It can make fragile analysis feel balanced. It can give users permission to move faster than they should because the language carries the emotional weight of judgment. In that setting, the real failure is not merely that the model was wrong. It is that the model changed the threshold at which humans felt comfortable proceeding.

This is why polished caution can be more dangerous than obvious overreach.

If the model speaks recklessly, people stay alert. If it speaks in the calm tone of institutional competence, people often become less demanding at exactly the point where scrutiny matters most.

The result is a form of decision inflation. Language that resembles responsibility starts being mistaken for responsibility itself.

LLM safety becomes harder once the institution starts reading tone as evidence

This is especially visible in sectors like banking, cybersecurity, legal operations, enterprise support, compliance, and internal decision support.

In these environments, the model’s tone matters because tone affects whether people feel an output is ready for action. A measured answer can reduce resistance, accelerate circulation, and lower the instinct to seek a second view. That would be fine if the tone reliably tracked genuine robustness. Often it does not. That is the illusion of safety in institutional form.

The system begins to pass because it has learned the language of responsible conduct, while the people around it stop demanding proof that the conduct is truly responsible under stress.

—

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 illusion of safety: What happens when LLMs say the right things for wrong reasons appeared first on e27.

Early-stage teams don’t lose to better-funded competitors. They lose to compounding drag. And right now, AI is introducing a new kind: the illusion of speed without systems.

Most conversations about AI in software development still fixate on accuracy: Is the model good enough? Is it hallucinating? Can it replace engineers?

But in practice, AI fails for a simpler reason: we ask it to do too much at once.

In a recent build, I discovered that the difference between chaotic, bug-prone output and clean, production-ready code wasn’t a better prompt. It was a better collaboration design.

When “working code” keeps breaking

The project started with a clear goal: building a better API client.

I described the vision to an AI coding assistant. It delivered quickly:

• A clean three-panel layout
• Resizable sections
• Theme system
• Modals and popups
• Everything wired together in one file

At first glance, it worked.

Then the loop began: Fix one bug → two new bugs appear Add a feature → some previous feature disappears

The larger the system became, the less reliable or more regressive the output tended to.

At this point, many would be thinking: maybe this model just isn’t as good. Or maybe I needed a better prompt, refine it, structure it, turn R.I.C.E into super-R.I.C.E, COAST into super-COAST… or reach for yet another prompting framework.

The constraint most people miss

Instead of restarting or rewriting prompts, I asked differently, like a partner would: “You seem to make more mistakes as the system grows. How can I help?”

AI was surprised by my question, and its answer also surprised me, and reframed everything: “I don’t get tired. But I do get crowded. Think of me as a desk, put too many papers on it, and things start falling off.”

This is the reality most teams overlook: AI doesn’t run out of memory. It runs out of attention.

Push it past a certain line of tightly coupled logic, and state tracking fractures. What looks like inconsistency or hallucination is actually just attention dilution.

What looked like a model flaw was really a system constraint.

Also Read: AI as an audience: Welcome to the citation economy

From monoliths to components

Once that constraint became clear, the workflow changed immediately. Instead of asking, “Build the whole application,” the approach shifted to:

• “Build the workspace switcher.” → Test it in isolation
• “Now add the context menu.” → Test again
• “Now integrate.”

The results weren’t marginal. They were immediate and measurable:

• Verification time: Cut from five to 10 minutes to ~30 seconds
• Iteration scope: Reduced from 800+ lines to 100-200
• Bug rate: Swapped compounding errors for predictable, isolated fixes
• Confidence: Shifted from declining to stable across cycles

This wasn’t just a coding tweak. It was a step change in reliability.

A simple pattern that scales

What emerged was a repeatable workflow—a way to build with AI that aligns with its strengths:

The component-first AI development pattern

• Define the contract: What the component does, inputs, outputs
• Build in isolation: Keep scope small and focused
• Verify immediately: Short feedback loops
• Fix locally: Avoid debugging inside a large system
• Repeat per component: Maintain consistency
• Compose at the end: Integrate only after validation

This pattern keeps AI within its working range—while giving humans tighter control over quality.

Why this matters for startups

For early-stage teams, this isn’t just a coding technique—it’s an operating model.

Teams that treat AI like a monolithic generator (“build the whole feature”) will encounter compounding bugs, fragile systems, slower iteration over time, and an increase in salient technical debt due to the propensity to outsource deep thinking to machines.

Teams that design workflows around AI constraints unlock faster cycles, lower QA overhead, and the ability to ship with junior+AI teams without sacrificing reliability and hollowing out critical long-term competencies.

In lean environments, that translates directly to burn rate efficiency, hiring leverage, and faster time-to-market.

The advantage isn’t better prompts or bigger models. It’s good old systems thinking.

Also Read: The future is full of humans working with humans, AI systems and other technologies

Rethinking the human–AI relationship

This experience also changes how we should think about roles.

AI is not a perfect executor—work with it for some time, and you can see its mistakes.
In any case, it’s not a replacement for engineering judgment.
It’s probably closer to a high-speed collaborator with its own quirks and constraints.

That shifts the responsibilities:

• The human defines structure, scope, and validation
• The AI executes quickly within that structure

The human becomes the orchestrator—maintaining coherence as complexity grows.

Philosophy meets Programming

Once you realise this, the model of “DeepSeek-ing” changes.

You can’t specify everything up front. Some meaning is always lost in translation.

Systems break when you try to out-constrain them. Because building with AI isn’t purely mechanical. It’s closer to a dance.

And most failures come from treating it like a machine. The real skill is finding the balance—the middle way between structure and emergence.

The real shift

The biggest insight isn’t technical. It’s a mindset shift.

Most teams try to push AI harder. The better approach is to design smarter around its limits.

The winners won’t be the best prompters. They would be the better system thinkers.

The post AI doesn’t fail because it’s wrong — It fails because you overload it appeared first on e27.

(L-R) GreatAsic founder and CEO Ong Chin Hu and co-founder and CTO Michael Liew Woon Chin

Malaysia’s GreatAsic Technology, a specialised chip design company focused on delivering high-performance, custom silicon solutions for “tomorrow’s technologies”, has closed a US$6.9 million pre‑Series A round, led by Vertex Ventures Southeast Asia & India and joined by Ehsan Kapital and Gobi Partners.

The raise signals a concrete step in the country’s ambition to move beyond semiconductor assembly and testing towards front‑end chip design.

Also Read: Why smart money is choosing semiconductors over Bitcoin: What can be done?

The round will bankroll engineering hires, operational expansion, and accelerated development of GreatAsic’s planned silicon projects targeting data‑centre, automotive, and edge AI markets.

A domestic pivot to design

For decades, Southeast Asia’s semiconductor story has been dominated by manufacturing: wafer fabrication, assembly, test and packaging. Malaysia, in particular, has been a critical node for assembly and testing—the traditional “Made in Malaysia” role within global supply chains.

GreatAsic’s fundraise comes amid the Malaysian government’s National Semiconductor Strategy, which explicitly aims to cultivate design capabilities so local firms can produce not just assembled chips but also chips designed on home soil.

“Malaysia has world‑class engineering talent and a once‑in‑a‑generation opportunity to design, not just build, the chips that power the AI era,” said Ong Chin Hu, GreatAsic’s founder and chief executive. “This funding lets us hire engineers, expand operations and accelerate our planned ASIC projects, and build a Malaysian semiconductor design ecosystem that endures.”

Access to Arm IP

A notable technical milestone for GreatAsic is its access to Arm Holdings’s semiconductor intellectual property. The startup is among the first Malaysian design houses to secure Arm Flexible Access (AFA) and Arm Neoverse Compute Subsystems (CSS) tokens, an important enabler for teams building modern SoCs for cloud and edge compute. The company says the AFA agreement is already formalised with the Malaysian Investment Development Authority (MIDA).

Also Read: Building the ASEAN AI archipelago: How Southeast Asia can secure its place in the global AI value chain

Arm’s IP ecosystems lower the barrier for fledgling design teams by providing pre‑validated cores and subsystem building blocks, reducing integration time and risk. For a regional startup, such access shortens the path from architecture to tape‑out and production. It increases the odds of competing for Asia‑Pacific customers that need customised silicon for AI inference, low‑latency automotive compute or specialised edge devices.

Founders with pedigree

GreatAsic’s leadership brings decades of silicon credentials. Founder Ong previously held senior roles at Intel, Marvell, and StarFive; co‑founder and CTO Michael Liew Woon Chin has held senior technical positions at Broadcom, Intel, Marvell, and StarFive. The team highlights multiple full‑mask tape‑outs and high‑volume production experience, turning what might otherwise be a speculative startup into a group with a demonstrable delivery track record.

Vertex Ventures’s Chan Yip Pang framed the investment as a bet on both the team and Malaysia’s broader strategic shift. “We have known this team for several years… For decades the country has assembled and tested the world’s semiconductors; designing them is a far harder and more valuable undertaking, and few teams in the region are equipped to take it on,” Chan said.

Ecosystem stacking: parks, funds, and public backing

The pre‑Series A also features Ehsan Kapital, a semiconductor‑focused fund backed by Selangor’s state actors and ecosystem operator SIDEC, which runs the Malaysia Semiconductor IC Design Park. SIDEC’s chief executive, Yong Kai Ping, characterised GreatAsic’s raise as proof that the Park’s blend of infrastructure, EDA (electronic design automation) tools and public‑private support is beginning to yield competitive local design houses.

Also Read: FusionAP’s US$2M raise signals Malaysia’s push up the semiconductor value chain

Public investments in physical infrastructure and training, coupled with venture capital focused on semiconductors, are typical of successful design hubs. South Korea and Taiwan show that deep local talent combined with concentrated industry support can move countries up the value chain.

For Southeast Asia, however, the path is more fragmented: talent pools are dispersed across Malaysia, Singapore, Vietnam and Indonesia, while fabs remain largely located elsewhere in Asia. That makes momentum from firms like GreatAsic significant: they not only create jobs but prove the viability of larger local systems—EDA workflows, IP licensing, packaging and test partnerships—needed to scale chip design.

Why this matters for Southeast Asia

Southeast Asia’s AI and cloud markets are growing rapidly, and many enterprises in the region are exploring custom silicon to reduce costs and latency, or to gain control over AI performance. Custom ASICs and AI SoCs can offer energy efficiency and price advantages over off‑the‑shelf accelerators, if a local design house can reach production at scale.

GreatAsic’s focus on data centre, Edge AI, and automotive aligns with regional demand. Automotive electronics are growing in markets such as Thailand and Indonesia; edge AI devices, such as retail cameras, smart factories, and logistics sensors, are spreading across Singapore, Malaysia, and Vietnam. A localised supply of custom silicon could shorten lead times, reduce integration complexity, and stimulate new product designs tailored to Southeast Asian conditions.

Risks and the long road ahead

Designing chips is capital- and time-intensive. Moving from design to tape‑out to production, and then to customers, involves technical risk, long sales cycles, and partnerships with fabs, foundries, and packaging providers. While GreatAsic has industry‑seasoned founders, the company will still need to demonstrate successful silicon and commercial traction to validate the thesis that Malaysia can sustain multiple design houses competing regionally.

The US$6.9 million raise is meaningful for early engineering growth and initial silicon runs. Still, broader scaling, especially if GreatAsic aims for high‑volume data‑centre chips, will require further capital, foundry access, and ecosystem partners. For now, the funding represents a timely proof point: regional venture capital is willing to back semiconductor design efforts within Southeast Asia, and public‑private initiatives are creating environments conducive to that investment.

Also Read: The quiet layer keeping the chip boom alive

GreatAsic’s raise is not a guarantee of a new regional centre for chip design, but it is an indicator: Southeast Asia’s semiconductor narrative is evolving beyond assembly lines, as capital, talent and policy begin to converge around the higher‑value work of designing silicon.

The post Malaysia’s GreatAsic raises US$6.9m to pivot nation from chip assembly to indigenous design appeared first on e27.

SUMMYS, a Kuala Lumpur-based venture builder, has concluded an unusual pitch contest that deliberately combined sustainability, cultural sensitivity, and commercial rigour to channel Southeast Asian startups toward Japan’s most pressing social problems.

The winners — spanning robotics, agritech, medical AI, and circular materials — now have routes into Japan’s heavyweight startup circuit through a partnership with IVS, one of the country’s largest tech conferences.

A low-energy stage, high-stakes outcomes

The ‘Jungle Forge Award — Social Impact Edition’ eschewed the typical conference trappings. Pitches were held outdoors without electricity: no spotlights, projectors, or microphones. The setting was designed both as a statement and an experiment to demonstrate that persuasive entrepreneurship need not rely on high-energy, high-emissions production.

Also Read: From pilot to production: Where robotics actually breaks

That choice resonated with the selection criteria. Judges assessed teams not only on technical feasibility and business models, but on their grasp of Japan-specific social issues and their ability to communicate in ways that resonate with Japanese audiences. In the run-up to the event, SUMMYS CEO Mariel Asami Fukase even led a Japanese-language lesson for participants; several founders used Japanese phrases in their pitches.

“It’s about respect as much as relevance,” said one judge. “If you want to operate in Japan, you must understand the problem and speak to the people who live it.”

Robotics takes the grand prize: tackling a regional labour crisis

Robopreneur, a Malaysian robotics firm, took the Grand Prize. The company offers service robots and “Physical AI” solutions for sectors where Japan and much of Asia face chronic labour shortages: hospitals, security, cleaning, and tourism.

Qarbotech won both the Silver Award and the Green Award for an agritech system that boosts yields and farmer incomes without adding to labour burdens. The judges cited on-farm validation and clear economics, making it a promising match for Japan’s shrinking agricultural workforce and its policy focus on food self-sufficiency.

Also Read: Qarbotech named winner of inaugural EQT Impact Challenge

Global Cerah secured the Open Innovation Award for a circular model that converts organic waste into protein and fertiliser. Judges pointed to its scalability and potential to strengthen Japan’s food system resilience, a key policy issue amid climate-driven supply-chain disruptions across the Asia Pacific.

Pixelence, an AI company that improves brain MRI diagnostics without contrast agents, won the AI Award. Japan’s population is among the oldest in the world, and early detection of dementia and other neurodegenerative disorders is a growing clinical and social priority. Judges argued that Pixelence’s technology could reduce costs and clinician workload while improving diagnostic reach.

The Next Generation Award went to Midwest Composites for an inventive approach that upcycles discarded tea leaves into composite materials suitable for automotive, EV and aerospace applications. The contest featured child judges; a nine-year-old’s enthusiastic reaction — “It was so cool that tea leaves that would be thrown away can become a new material” — was a reminder of the storytelling power founders can gain by tying sustainability to everyday products.

Prizes included more than trophies. The top two startups secured exhibition space at IVS, direct access to international investors and corporate partners, and introductions aimed at fundraising and business development in Japan.

SUMMYS framed the physical trophy — a watch — as symbolic: a reminder that building impact takes time, and that the organiser intends to walk the journey with participating startups.

The Japan-Southeast Asia bridge

For Southeast Asian founders, Japan represents both a market and a source of corporate partnerships, manufacturing expertise and patient capital. But entering Japan requires more than a scalable product; it demands cultural fluency, local validation, and integration with incumbent systems. The Jungle Forge Award’s emphasis on communication, including basic Japanese language ability, is a pragmatic nod to those realities.

The event’s livestream and hybrid voting, which included venture capitalists, corporate innovation leaders, child judges, and the public, reflected another lesson: narratives that combine technical rigour with social resonance travel better. Startups that can demonstrate applicability in Japan while highlighting regional scalability stand a stronger chance of turning pilot projects into commercial ties that flow in both directions.

What’s next

SUMMYS has signalled continued work to foster open innovation between Japan and Southeast Asia, positioning the award as part of a longer-term collaboration with IVS and Japanese corporates. If successful, the model could become a regular pipeline for Southeast Asian founders seeking not just funding but operational corridors into one of Asia’s most advanced and socially challenged markets.

Also Read: How Japan can empower a new wave of SEA startup innovation

For investors and corporates watching from both regions, the competition offered a tidy proof point: pragmatic, culturally aware startups from Southeast Asia can present viable, low-carbon solutions to Japan’s demographic and environmental pressures, and, crucially, those solutions may loop back to benefit the region as well.

The post From rice fields to hospitals: Winners of SUMMYS’s Social Impact Award aim at Japan’s ageing and labour shortages appeared first on e27.

Today, Bitcoin trades at US$61,789.80, reflecting a 1.36 per cent decline over the past 24 hours. This drop mirrors a broader 1.17 per cent contraction in the total crypto market capitalisation. Mainstream commentators attribute this movement entirely to a sudden risk reaction.

My independent analysis reveals a more complex convergence of geopolitical shocks and institutional liquidity drains. The immediate catalyst for this sell-off is escalating tensions in the Middle East. President Donald Trump announced a military response after Iran shot down an Apache helicopter. This geopolitical shock instantly triggered a flight from risk assets across global markets.

Bitcoin behaved precisely as a correlated risk asset in this environment, dropping to an intraday low near US$60,892 before buyers stepped in. We see this exact same behaviour in traditional equities. The S&P 500 briefly dipped 2.2 per cent on the news before recovering into the close. Major benchmarks finished mixed, and the Dow Jones Industrial Average managed only a marginal gain. This tight correlation between cryptocurrency and traditional tech-heavy indices confirms that institutional algorithms currently treat digital assets as an extension of the broader risk complex.

Structural weaknesses in institutional demand continue to suppress price action beyond the geopolitical headline. U.S.-listed Bitcoin exchange-traded funds extended their outflow streak, underscoring a persistent lack of buy-side conviction. Analysts at Wintermute correctly point out that this environment reflects weak institutional inflows rather than outright panic.

This specific dynamic makes establishing a durable bottom incredibly difficult. Concurrently, the market experienced a severe leverage flush. Traders lost over US$112 million in Bitcoin long positions within a single day. This forced liquidation accelerated the downward momentum and punished overextended speculators.

Also Read: From rice fields to hospitals: Winners of SUMMYS’s Social Impact Award aim at Japan’s ageing and labour shortages

I have always viewed highly leveraged crypto trading as a form of gambling with slightly better odds than a casino. The liquidations simply represent the house collecting its due. The removal of this excess leverage clears the order book and sets the stage for potentially less volatile price discovery in the coming sessions. We must also contextualise this crypto sell-off within the broader global macroeconomic environment to fully grasp the implications. Technology stocks face their own headwinds. The 3.6 per cent drop in Apple shares following the final World Wide Developer Conference keynote from CEO Tim Cook highlights these pressures. Shares had already fallen close to two per cent on Monday due to poor market reception of the Siri artificial intelligence update.

The market now turns its attention entirely to the macroeconomic data driving central bank policy. The government will release the May United States Consumer Price Index report on June 10. This print serves as the primary directional catalyst for the near term. Consensus expects headline inflation to rise to 4.2 per cent.

This expectation follows an April inflation reading of 3.8 per cent year-on-year. That April figure marked the highest level since 2023. A massive 17.9 per cent jump in energy costs largely drove that previous spike. If the May data prints cooler than expected, we could see a relief rally pushing Bitcoin toward the US$64,000 resistance level. Conversely, a hot inflation reading will reinforce hawkish monetary policy and likely force a retest of the critical US$60,000 support zone.

From a technical perspective, the current market structure demands careful observation from all active market participants. Bitcoin currently trades below key moving averages and maintains a bearish short-term trend. The Relative Strength Index on the 14-day timeframe sits at 23.89. This deeply oversold condition suggests that a technical bounce remains highly probable. A cooler inflation print could fuel a rally targeting the US$64,000 level, which aligns perfectly with the 78.6 per cent Fibonacci retracement level. If buyers fail to defend the US$60,000 support, the price will likely cascade toward the next major liquidity zone around US$55,000. Traders must watch the US$64,000-US$66,000 supply zone closely. A decisive reclaim of those levels would provide the first technical confirmation of strengthening momentum.

Also Read: How to build an AI-ready workforce: The skills that matter in the age of agents

Global trade and corporate spending metrics provide further context for this market environment. China reported robust May exports, rising 19.4 per cent year on year, and imports jumped 27.4 per cent. This beat expectations and widened the trade surplus to US$103.22 billion. Meanwhile, Bank of America warns clients to take profits because seven of its 10 bear-market signposts have been triggered. They highlight that hyperscaler capital expenditure will soon hit 100 per cent of operating cash flow. This contrasts starkly with the 40 per cent ratio from 2023.

These megacorporations will soon spend every dollar they generate on AI infrastructure. Investor demand in other sectors shows an immense appetite for new tech ventures. SpaceX’s initial public offering demand now reportedly approaches 4 times oversubscribed levels. Commodity markets also reflect this complex web of geopolitical and economic pressures across the globe today. Oil retreated after the US Energy Secretary noted that traffic in the Strait of Hormuz is increasing. This observation eased the supply premium created by tensions with Iran.

The confluence of geopolitical stress and institutional selling has driven Bitcoin lower. A sustained reversal requires either diplomatic de-escalation or a positive macroeconomic surprise from the inflation data. I will continue to monitor these structural shifts independently and look past the mainstream narratives. Identifying the true drivers of value in this evolving financial landscape demands rigorous analysis and a forward-looking perspective. The market is at a critical inflexion point, with macroeconomic data set to dictate the next major price move.

Based on what I see and referencing the historical cycle structures, US$44,XXX represents a high-probability macro floor, but it is the deep end, not the baseline, of the expected bottoming range.

—

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 US$61,789 breakdown: Why geopolitics just overrode every technical indicator appeared first on e27.

In our previous explorations of psychological safety, we’ve seen that psychological safety is the number one predictor of team performance, that it is not about being “nice” but about pairing high standards with high safety to create the Learning Zone. We’ve explored how Edgar Schein’s deep cultural diagnosis and Amy C. Edmondson’s practical interventions work together to build organisations that are both successful and truly human.

This piece addresses something critical: the founder’s own role in creating or destroying psychological safety. Because here’s the uncomfortable truth: you can implement every framework, read every book, and hire every consultant, but if you, the founder, are the problem, none of it will work.

The founder’s paradox

Let’s take a look at you, founder. You are a creature of beautiful contradictions. You possess a vision so clear that it is almost like a mirage, a bias for action so strong that it makes the laws of physics nervous, and standards so high they give astronauts vertigo.

These are your superpowers. They are the very reason your company exists.

And, if you are not careful, they can turn out to be the very things that will poison it from the inside out.

This is the founder’s paradox: the traits that make you exceptional at starting something are often the same traits that make you terrible at leading it. You are the sun that both warms and burns. This intensity gives life, but it can also scorch the delicate ecosystem of psychological safety required for your team to thrive. Before you can fix your team, you must first look in the mirror. Here are a few archetypes of safety-destroying behaviour. See if you can recognise yourself in any of these.

Also Read: AI startups are hiring around answers they haven’t earned yet

The archetypes of accidental tyranny

Every founder is a unique blend of strengths, but these strengths, when overused, manifest as these distinct archetypes.

• The visionary

Your vision is your gift, you see the future with breathtaking clarity.

The problem?

You’re so in love with your vision that you can’t tolerate anything that deviates from it. When a team member raises a concern or a dissenting opinion, you do not hear a valuable stress test, you hear a threat to the dream.

Maybe your internal monologue sounds like this: They just don’t get it. If they saw what I see, they’d agree.

What your team experiences: Your conviction feels like a brick wall. They learn that bringing you anything other than enthusiastic agreement, yes-man style, is a career-limiting move. The echo chamber you often complain about is built by you. Brick by brick, with every dismissed counter-argument.

• The perfectionist

Your standards are legendary. You demand excellence in everything, from the product UI to the font choice. You are fond of making such decisions and sending them in a voice memo. This is why your product is beautiful. It is also why your team is terrified.

Maybe your internal monologue sounds like this: It’s not quite right. We can do better. This small flaw will ruin everything.

What your team experiences: They feel like they are constantly walking on eggshells.

The fear of not meeting your impossibly high standards leads them to hide mistakes, avoid risks, and present only perfectly polished work. The messy, half-formed ideas where true innovation lives? They die in your Slack channel, strangled by the fear of your critique.

• The urgency addict

You move at the speed of light. You are a whirlwind of action, a testament to the power of: done is better than perfect.

You are addicted to the adrenaline of momentum.

Maybe your internal monologue sounds like this: Why is this taking so long? We need to move faster! We’re losing our window!

What your team experiences: Your pace feels like a perpetual fire drill. There is no time for questions, no space for reflection, no room for the “stupid question” that might have saved the project. They learn to just nod and run, even if they are running in the wrong direction. Your need for speed has trampled their need for clarity.

• The solver

You are a brilliant problem-solver. You see a problem and your brain instinctively jumps to a solution. This is how you’ve survived this long. But your compulsion to solve is robbing your team of a chance to learn.

Maybe your internal monologue sounds like this: It’s just faster if I do it myself. I already know the answer.

What your team experiences: They feel disempowered.

Why bother wrestling with a hard problem when they know you’ll just swoop in and fix it? They stop taking ownership. They become executors of your solutions, not owners of their domains. You’ve created a team of brilliant hands, but you’ve stunted the growth of their brains.

The emotional weather you create

Beyond these archetypes, there is a fundamental truth to be told: as a founder, you’re not just a person in the company, you’re the weather. Your mood sets the atmospheric pressure for the entire organisation. Be it a funding call that didn’t go down well or a frustrating bug issue that leads to sleepless nights, you bring that energy into the office, and it becomes everyone’s reality.

The higher you climb in authority, the more amplified your every word and action becomes. A casual, frustrated comment from you (“This dashboard is useless”) can feel like a public condemnation to the person who built it. Your furrowed eyebrows in a meeting can silence an entire room.

Also Read: Why startups need mobile apps to thrive in today’s competitive market

In the high power-distance cultures, especially common in Asia, this effect is magnified tenfold.

Your team is culturally primed to defer to you, to read your signals and to adapt their behaviour to please you. If your emotional state is volatile, they will retreat into the safest possible position: silence.

This is why emotional regulation is not a soft skill for a founder. It is a core competency. You must learn that you are like a thermostat, not a thermometer: you set the temperature, you don’t just reflect it.

The vulnerability mandate: your most powerful tool

So how do you counteract your own safety-destroying strengths? The simple answer: you have to lead with vulnerability.

You must be the first to admit you were wrong.

You must be the first to say, “I don’t know.” You must be the first to thank someone for bringing you bad news.

Vulnerability is not weakness. It’s the ultimate signal of strength. It tells your team that the goal is not to be right; the goal is to get it right. It shows that your ego is secondary to the truth and the success of the collective mission.

Three practical ideas for self-correction

• The “shut up and listen” challenge

For the next week, go into every meeting with the explicit goal of being the last to speak. When you do speak, only ask questions. This simple constraint will force you to listen and create space for others.

Also Read: RIE2030’s hidden flaw: The one capability Singapore’s startups are missing

• The “bad news reward”

The next time someone brings you bad news early, stop what you are doing and publicly thank them. Say the words: “Thank you for bringing this to me. This is exactly what we need to be doing.” You are rewarding the behaviour you want to see. Do this more than three times, and you will be on your way to changing your culture.

• The “I screwed up” ritual

Start your weekly team meeting by sharing one thing you got wrong that week. It can be small. “I was too dismissive of Jessie’s idea in our last meeting, and after thinking about it, I realised she was right.”

By modelling fallibility, you give your team permission to be human.

Remember that the journey to building a psychologically safe organisation is not an external one. It does not begin with surveys, workshops, or off-sites.

It begins with the quiet, difficult and essential work of looking in the mirror and acknowledging that the biggest threat to your company’s culture might be you.

But here’s the good news: this also means that the power to fix it lies in your hands.

And that is the most powerful position a founder can be in.

—

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 startup founder’s paradox: How your strengths are killing psychological safety appeared first on e27.

Singapore-headquartered early-stage venture capital firm and accelerator, Accelerating Asia, has announced the five startups selected for its thirteenth cohort, chosen from a record 724 applications spanning 20 countries, with an acceptance rate of under one per cent, the most competitive in the firm’s history.

All five companies were already generating revenue or active usage at the time of selection, an unusual benchmark for an early-stage intake. The cohort spans consumer-goods distribution, last-mile logistics, customer-success software, e-commerce, and retail loyalty, across markets including Singapore, the UAE, Bangladesh, the United States, and Hong Kong.

Also Read: AI will replace inertia before it replaces people

Despite AI featuring in nearly every application this cycle, Accelerating Asia says the term’s ubiquity worked against candidates who leaned on it as a selling point. “When nearly every founder says they use AI, the phrase stops being a signal,” said Amra Naidoo, General Partner and co-founder. “The question that actually decided this cohort was whether the AI was the product or the leverage.”

The five companies selected are:

Driftly AI (UAE/US)

An AI-powered operating suite for consumer goods distribution, founded by Sheheryar Iqbal, former Head of Supply Chain at Airlift, where he launched over 80 warehouses and worked with 450-plus manufacturers. Embedded with flagship customer Gourmet for more than eight months, the company has doubled its clients’ sales footprint, saved over 20 per cent in margins, and is now on a path to US$1 million in ARR. Its warm enterprise pipeline includes Coca-Cola, Pepsi, and Nestlé.

DIGIBOX (Bangladesh/Singapore)

Bangladesh’s first shared last-mile logistics infrastructure, operating a network of IoT delivery lockers across 55 sites, with close to one million deliveries and 123,000 end users. The company designs and manufactures its own lockers in-house and handles roughly one per cent of all Daraz orders in the country. It cuts delivery costs by up to 40 per cent and failed deliveries by up to 80 per cent. Co-founded by Rezwanul Haque Jami, a two-time founder with prior exits.

Meza AI (US)

Positioning itself as “cursor for customer success”, Meza AI is an AI-native platform helping SaaS companies reduce churn and unlock upsell from existing customers. It has around US$230,000 in ARR, 10 paying customers, and a 100 per cent pilot-to-paid conversion rate, monitoring more than 1,500 accounts. Founded by repeat operators Abhishek and Priya Yadav, who previously scaled a consumer platform to over two million users.

Govaly (Bangladesh/Singapore)

The largest fashion and beauty e-commerce marketplace in Bangladesh, with over 100,000 users, 70,000 orders, and 1,000-plus verified sellers in its first year. GMV has grown 27x in 18 months with zero paid acquisition. The platform operates without a warehouse, shipping directly from verified sellers, and delivers roughly 80 per cent faster than the industry average. Founded by Himel Faraz and Jeion Ahmed, with a 45-person team.

Also Read: The next frontier for tech startups? The US$590B beauty industry

meed (Hong Kong)

A consumer-first retail loyalty platform that works with a single QR scan, natively integrated with Apple and Google Wallet, requiring no app or login. It has attracted more than 700 organic merchant sign-ups from 85 countries with no paid acquisition, zero churn among paying merchants, and an organic LTV-to-CAC ratio of approximately 140-to-1. Founded by Phil Ingram, a 28-year product veteran, with paying traction concentrated in UK salons.

Accelerating Asia co-founder and General Partner Craig Bristol Dixon framed the cohort’s significance for investors: “Five operators already earning, across five markets that most funds never travel to, and we are in early on every one. Early entry in overlooked markets is where the structural advantage lives.”

The five companies will spend the next 100 days in the Accelerating Asia programme, culminating in a Demo Day. Applications for Cohort 14 will open soon. The announcement coincides with the final close of the firm’s second fund.

The post From Bangladesh lockers to Hong Kong loyalty: meet Accelerating Asia’s most global cohort yet appeared first on e27.