The question “Is the AI industry profitable?” has two correct answers, and they point in opposite directions. At the chip-design and leading-edge-fabrication layers, AI is already one of the most profitable industries in commercial history. At the layers the market calls “AI”, frontier model labs, GPU-rental builders, and most applications built on someone else’s model, it is among the most loss-making activities ever financed.

Both statements are true. The investment question sits in the distance between them.

The reason is mechanical. Across the AI stack, the cost of intelligence is falling rapidly. But a falling cost only becomes profit somewhere. Whether that decline lands as margin, as a lower price to the customer, or as a transfer to a supplier depends on one question: who owns the bottleneck between the falling cost and the price the customer will pay?

Where a firm owns that bottleneck, it keeps the cost decline as margin. Where it owns none, competition forces the decline through. Walk the AI value chain from chips to applications, and the pattern is already visible. Profit sits where a pass-through is blocked. It evaporates where competition lets pass-through run free.

Start with the two companies that keep the money. NVIDIA reported fiscal-2026 revenue of US$215.9 billion, GAAP operating income of US$130.4 billion, and GAAP net income of US$120.1 billion, a net margin of nearly 56 per cent. TSMC earned 2025 net income of US$55.2 billion on revenue of US$122.4 billion, a 45.1 per cent net margin. Together, Nvidia and TSMC booked roughly US$175 billion of net profit in their latest fiscal years.

This is not a forecast. It is where AI profitability already exists.

Both companies sit behind gates that the rest of the stack must pass through. NVIDIA’s moat rests on CUDA, networking, scale, and the difficulty of coordinating around an alternative. TSMC’s moat is harder still: leading-edge fabrication is gated by physics, capital, yield learning, and process knowledge that takes years to reproduce. These are not normal suppliers. They are toll collectors.

The cloud layer is more ambiguous. AWS, Microsoft Azure, and Google Cloud are large, profitable businesses. AWS earned a 37.7 per cent operating margin in the first quarter of 2026, and Microsoft’s Intelligent Cloud has run margins in the low 40s. But hyperscaler free cash flow is being consumed by the AI build-out, and the cloud owners are trying to escape Nvidia’s toll through custom silicon. Amazon’s Trainium, Google’s TPUs, and Microsoft’s Maia are attempts to become bottleneck owners rather than resellers of someone else’s bottleneck.

Also Read: The AI economy is moving faster than our institutions

Where a cloud owner runs its own silicon, it can keep more margin. Where it buys Nvidia capacity, finances data centres, and rents compute to model labs, its economics compress. The cloud business is profitable, but AI infrastructure may not be unless demand arrives fast enough and custom silicon works well enough.

The neoclouds show what happens when revenue grows without a bottleneck. CoreWeave more than doubled first-quarter 2026 revenue to US$2.08 billion and reported a 56 per cent adjusted EBITDA margin. But adjusted operating margin was only 1.0 per cent, and GAAP net loss was US$740 million, with quarterly interest expense of US$536 million. Depreciation on GPUs and debt service consume the economics. CoreWeave buys Nvidia hardware at market prices, finances it with borrowing, and rents capacity into a competitive market. It owns no gate.

The frontier labs invert the popular intuition. OpenAI’s annualised revenue run-rate was roughly US$20 billion at the end of 2025. Anthropic reportedly reached around US$30 billion in April 2026 and about US$47 billion by late May. The growth is real, even if the figures are reported rather than audited. But revenue is not profit. OpenAI’s gross margin has been reported to be around one-third, constrained by inference costs, and internal projections reported publicly pointed to a multibillion-dollar loss in 2026.

Through the Abundance Economics lens, the labs are not toll collectors. They are in demand. A large share of their revenue flows upstream to chips and cloud and lands there as margin. The model itself is becoming less defensible because two forces push price down at once: open-weight models keep closing the capability gap, and the cost of fixed model performance keeps falling. In a layer with low switching costs and credible substitutes, falling input costs cannot be retained. Competition forces it through. That is why a lab can scale revenue explosively and still lose money.

The application layer needs more care. “AI apps” are not one category. Thin wrappers over frontier APIs own no gate and are likely to be crushed. Embedded workflow systems can be different because they control customer data, procurement position, operating processes, or a regulated context. Distribution-owned applications can also hold margin where they already own the user relationship.

Also Read: Give physical AI a soul: Why your voice AI still feels like a bot

Palantir is the clearest example. It is not just “an AI app.” It is an embedded data-and-workflow layer inside government and enterprise operations, and that position can behave like a bottleneck. By contrast, implementation consultants capture demand but earn consulting economics. Accenture may book billions in generative-AI work, but its overall operating margin remains around 15 per cent.

• The first capital-allocation implication is simple: do not price the AI stack as one trade. NVIDIA, TSMC, hyperscalers, neoclouds, model labs, workflow software, and thin apps have different economics because they sit in different places in the pass-through chain.
• The second implication is that revenue is the wrong metric at the model layer. A lab’s run-rate measures how much compute it is buying as well as how much value it is keeping. Treating model revenue like Nvidia revenue is a category error.
• The third implication is that the most durable profit pool may be less glamorous than the market assumes. NVIDIA’s moat is powerful but contestable: the cloud owners’ custom silicon is an attack from above. TSMC’s moat is harder to clone because it rests on physics, capital intensity, yield learning, and years of manufacturing execution. That does not make TSMC risk-free. It has Taiwan exposure, customer concentration, and cyclicality. But the moat itself is structurally harder to erode.
• The fourth implication concerns the capital now being committed. The major hyperscalers are reportedly tracking toward a combined 2026 AI capital spending approaching US$700 billion. That spending is ahead of demand. At the enterprise buyer level, an MIT study found that 95 per cent of organisations deploying generative AI had seen no measurable profit-and-loss impact. If the five per cent of successful deployments scale into the majority, the capital may be repaid. If not, much of the non-bottleneck stack remains a transfer mechanism feeding the gates.

The thesis can break in several ways. NVIDIA’s gate could erode faster than expected if custom silicon scales. A frontier lab could become a true toll collector if one model achieves a durable capability lead, or if regulation entrenches a small number of approved model owners. Enterprise demand could arrive faster than current evidence suggests. Or the binding constraint could migrate from chips to power, shifting the profit pool toward whoever controls dispatchable energy near data centres.

There is also a circularity risk. Some AI demand is financed by the same companies that benefit from it, through equity stakes, cloud commitments, reseller structures, and compute deals. That does not make Nvidia’s or TSMC’s profits fake. Their margins are real. But it does mean some of the revenue feeding those margins may be more fragile than organic demand would be.

The investor question is not whether AI is profitable. It plainly is, at the gates. The question is whether the demand behind those gates is durable enough, and arrives quickly enough, to repay everyone standing in line behind them.

—

The post Is the AI industry profitable? Yes, just not where you’re looking appeared first on e27.

MoneyHero Group wants you to focus on the bright spots. Revenue climbed 15 per cent. Its shiny Wealth vertical surged 53 per cent. Its AI transformation story is compelling. And its Adjusted EBITDA loss? Down a whopping 68 per cent year-over-year to US$1.1 million.

On the surface, the Singapore- and Hong Kong-based personal finance comparison platform appears to be a company turning a corner.

Also Read: MoneyHero’s ‘turnaround’ built on a shrinking user base and retreat from SEA

Dig past the press release language, and the picture is considerably more complicated.

The net loss nobody wants to talk about

For the three months ended 31 March 2026, MoneyHero posted a net loss of US$6.7 million, nearly three times the US$2.4 million loss it recorded in the same period last year. That is not a rounding error. That is a 175 per cent deterioration at the bottom line.

The company’s explanation? Non-cash accounting items: a US$1.1 million swing in the fair value of warrant liabilities and US$2.4 million in unrealised foreign exchange losses from regional currency weakness against the US dollar. Fair enough, these are real accounting adjustments. But even if you strip them out entirely, the residual loss still lands around US$3.2 million, worse than the prior year’s US$2.4 million net loss. The narrative that operational performance is “robust” requires significant suspension of disbelief.

The press release buries this detail in a single paragraph, quickly pivoting to the Adjusted EBITDA figure, a non-IFRS metric that requires stripping out no fewer than six categories of charges to reach that headline-friendly US$(1.1) million number.

The mystery of US$1.6M in legal fees

Perhaps the most glaring anomaly in the report is a line item that receives precisely zero words of explanation in the management commentary: US$1.596 million in “non-recurring legal and professional fees and other expenses” incurred during the quarter.

In the same period last year, this figure was US$0.

The company categorises this as a non-recurring item and strips it out of Adjusted EBITDA. But US$1.6 million in legal costs is not a footnote; it is 9.7 per cent of the quarter’s total revenue. What litigation, regulatory matter, or professional engagement generated this bill? The report does not say.

We have reached out to the company for details and we will update this piece with the details as and when we hear from them.

This is likely a significant contributor to the 60 per cent spike in general, administrative and other operating expenses, which ballooned from US$2.19 million to US$3.51 million year-over-year, another figure conspicuously absent from the management commentary.

Revenue grew, but gross margins compressed

MoneyHero’s revenue of US$16.5 million is real and commendable. Double-digit growth across its core verticals — Credit Cards up 10 per cent, Personal Loans and Mortgages up 13 per cent, Wealth up 53 per cent, Insurance up 12 per cent — tells a story of genuine commercial momentum, particularly in Hong Kong, which surged 33 per cent to US$8.5 million and now accounts for 51.3 per cent of total revenue.

Also Read: MoneyHero swings to profit, but only on cost cuts and FX gains

But here is what the report does not highlight: the cost of revenue grew at 23.6 per cent, significantly faster than the 15 per cent revenue increase, rising from US$6.4 million to US$7.9 million. Gross margins are quietly compressing.

The company instead draws attention to the combined decline in technology costs, employee benefits, and advertising and marketing expenses, which came down 13 per cent year-over-year to US$8.5 million. This is a legitimate operational achievement, but the framing deliberately excludes the cost of revenue, which is by far the largest single cost line. It is a selective presentation designed to emphasise efficiency while obscuring margin erosion.

The “strategic retreat” in Southeast Asia

MoneyHero’s two smaller markets (the Philippines and Taiwan) posted revenue declines of 17 per cent and 12 per cent, respectively. The company frames these as deliberate strategic decisions: optimising margins, cutting low-quality volume, and building structural leverage. Perhaps. But consider this: the Philippines is home to 6.9 million of MoneyHero’s 9.8 million registered members (70 per cent of its entire user base), yet it contributed just US$1.47 million, or 8.9 per cent, of total revenue in the quarter. A market representing seven-in-ten of the group’s members is generating less than a tenth of its revenue. That is not a margin quality story. That is a monetisation failure, and calling it “strategic” is cold comfort for investors watching Southeast Asia’s largest member base sit largely idle.

Meanwhile, the platform’s overall traffic footprint shrank dramatically: monthly unique users fell 31 per cent year-over-year from 5.7 million to 3.9 million, and total sessions dropped 29 per cent from 17.5 million to 12.4 million. Clicks fell 33 per cent. The company is converting a smaller, higher-intent audience more efficiently (that part is true and defensible), but the scale of audience attrition is a genuine long-term risk that the report effectively sidelines.

Cash burn and the runway question

MoneyHero ended the quarter with US$27.984 million in cash, down from US$31.185 million at the end of 2025. That is a US$3.2 million cash burn in a single quarter. The company describes its balance sheet as “healthy” and highlights its debt-free status, which is accurate. But net current assets also declined, from US$37.5 million to US$32.8 million over the same period.

At the current burn rate, the company has roughly eight to nine quarters of runway, about two years. That is not a crisis, but it is not the picture of financial comfort the press release implies. The clock is ticking toward the “sustainable Adjusted EBITDA profitability” the company keeps promising.

The Adjusted EBITDA problem

“Our Adjusted EBITDA loss narrowed sharply by 68 per cent year-over-year,” said Danny Leung, Interim Chief Executive Officer and Chief Financial Officer, in the company’s earnings statement.

That figure is technically accurate. But to get from a US$6.7 million net loss to a US$1.1 million Adjusted EBITDA loss, the company removes US$5.4 million worth of charges — unrealised FX losses, warrant fair value changes, share-based payments, legal fees, depreciation, and interest. The adjustments are five times larger than the resulting metric. When a non-IFRS measure requires stripping out more than 80 per cent of the underlying loss to produce a headline number, investors should treat it as a directional indicator at best, not a proxy for cash profitability.

What is genuinely promising

None of this is to say MoneyHero is without real momentum. Its approval rate expansion, from 36 per cent to 48 per cent, is a meaningful operational achievement, particularly as total approved applications held flat at 156,000 despite a significant reduction in total applications. The Wealth vertical’s 53 per cent growth and the broader shift toward higher-margin products are structurally sound strategies. Hong Kong’s dominance is real. The AI-driven cost reduction story, though early, has tangible evidence in the declining technology and headcount costs.

Also Read: Decoding MoneyHero’s Q1: The profit push amid shrinking revenues

The question is whether the company can translate these genuine operational improvements into actual IFRS profitability and do so before its cash reserves force a capital raise or a more dramatic restructuring.

For now, MoneyHero is a company with a strong narrative, a compelling direction, and some numbers it would rather you did not look at too closely.

The post MoneyHero’s winning quarter has a US$6.7M problem appeared first on e27.

Southeast Asia’s semiconductor story is no longer limited to assembly, testing and outsourced manufacturing. This list points to a region, led largely by Singapore and Malaysia, that is building more of the stack itself: custom ASICs, silicon IP, chiplet packaging, photonics, test equipment and fab services.

Some of these startups are tackling narrow but essential problems, such as radiation-hardened chips, RF test components and FPGA design software. Others are pushing local industry further upstream into design and advanced packaging.

Also Read: The quiet layer keeping the chip boom alive

Taken together, they suggest a semiconductor ecosystem that is becoming more specialised, more technical and less reliant on being the backend of someone else’s supply chain.

GreatAsic Technology

FusionAP

Silicon Box

Zero-Error Systems (ZES)

SkyeChip

Global TechSolutions (GTS)

Swift Bridge Technologies

Infinecs Systems

MaiStorage

Oppstar

LightSpeed Photonics

Core Semiconductor

Cloptech

Plunify

Divergent Technologies

The post 15 Southeast Asian semiconductor startups moving beyond assembly appeared first on e27.

Three months ago, I reviewed a case that looked like routine onboarding fraud — until none of the patterns I expected to find were there.

The application was for a mid-sized supplier with a decent credit profile, clean documents, and a sensible business model. The verification photos checked out. The voice call with the principal sounded normal. The contract was signed. Two weeks later, the bank account had gone dark, and the customer who had introduced them no longer recognised the name.

The application was synthetic. The photos were generated. The voice on the call was cloned. The business model existed only in the pitch deck.

I have spent fifteen years inside Indonesian risk functions — banking, insurance, sharia microfinance — and I have lectured on fraud detection in two of those years. The patterns I learned to look for, the patterns I taught others to look for, are not the patterns showing up in the casework now. The playbook I trusted for a decade has stopped working — faster than most risk teams in ASEAN are willing to admit.

What changed

Three patterns are new enough that they deserve to be named in the open.

Synthetic identity at scale. Until about eighteen months ago, identity fraud was bottlenecked by the cost of fabrication. A reasonable fake ID, a plausible address, a working phone, a consistent social presence — each piece required real effort. Generative AI has collapsed that cost curve. A single attacker can now generate hundreds of internally-consistent identities in an afternoon, each passing every check designed before 2024.

Voice and video impersonation. The “CEO email scam” of 2018 has evolved. The 2026 version is a thirty-second voice call from a number resembling your CEO’s, with the CEO’s actual voice asking for an urgent wire transfer. The voice is generated from three minutes of public conference recordings. The verification protocols banks trained employees on five years ago do not catch this attack.

Slow-burn synthetic onboarding. The most expensive new pattern is the patient one. An attacker creates a synthetic business identity, lets it operate for six to twelve months building a transaction history, applies for credit on the back of that history, draws down the credit, and disappears. The fraud is only visible in aggregate — after the loss is locked in.

Also Read: The AI economy is moving faster than our institutions

Why the old playbook fails

Most fraud playbooks across the region were built on three assumptions that no longer hold.

Fabrication is expensive. Identity verification, document checks, and onboarding interviews all assumed the cost of producing convincing fake material was high enough to deter scale. That assumption is gone. The marginal cost of one more fake identity is indistinguishable from zero.

Human verification is the gold standard. The voice call, the video interview, the in-person meeting — these were the fallbacks when automated checks were ambiguous. Each is now itself vulnerable to generated content.

Fraud is an event. The traditional playbook treats fraud as a moment — a fake invoice, a suspicious transaction, a flagged login. The 2026 pattern is increasingly a campaign — a multi-month sequence of legitimate-looking actions designed to build trust before the loss. By the time the loss arrives, the institution has already paid its onboarding cost on the relationship.

What is starting to work

Three responses are emerging.

Cross-channel correlation. Risk teams that connect onboarding, transaction monitoring, and customer service data into a single view are catching slow-burn fraud earlier. The signal is rarely visible inside one channel. It is almost always visible across three.

Liveness and behavioural verification. Identity checks that include real-time, randomised prompts — actions an attacker cannot pre-render — are catching synthetic identities at the door. Deployment across the region is uneven, but the institutions doing it well are seeing the difference in their loss numbers.

Internal red-teaming. The teams catching the most generated content are the ones running their own attacks against their own defences. That detection muscle is the closest thing to a real defence we have.

Also Read: AI governance in banking operations and decisioning

What needs to happen

The next eighteen months will be the most expensive in ASEAN fraud history for the institutions that have not retired the old playbook. Three moves would meaningfully shorten the gap.

Retire the verification protocols built for pre-2024 fabrication costs. They were designed for a world that no longer exists.

Invest in cross-domain risk talent before the loss events force it. The people who can sit between fraud, identity, and data engineering are not being trained anywhere at scale.

Treat fraud as a campaign, not an event. Build the systems and the reviews to detect patterns across months, not transactions across minutes.

The macro stakes

ASEAN’s financial system has digitised rapidly over the last five years. The fraud surface has digitised faster. The institutions that will absorb the next wave of losses are not the ones with the smallest fraud teams — they are the ones whose fraud teams are still working from the playbook that taught them to expect events instead of campaigns, individuals instead of synthetics, and effort-bottlenecked attacks instead of zero-marginal-cost ones.

The new playbook exists. The question is how quickly the institutions reading the old one will admit they are.

—

The post Synthetic identities now cost nothing to make, and ASEAN’s banks have not caught up appeared first on e27.

A few years ago, the logic of investing in AI seemed simpler: find the right model or product and bet on it. Today, models that take a year to build become obsolete in months. Betting on a specific AI product is like picking a favourite app at a moment when the operating system itself is changing.

If the bet isn’t on the product, then what? The real race is happening at the infrastructure level. Whoever controls the chips, the memory, the manufacturing equipment — controls the future of AI.

This is where the conversation begins about how and why Japan and Ukraine have ended up in a strategically important position but on opposite sides of the same stake. And why the partnership between these two countries is about structural logic and mutual reinforcement.

Japan’s quiet comeback

Japan is deliberately reclaiming its status as a global centre of semiconductor manufacturing with characteristic precision, backed by real results. According to the Brookings Institution, Japanese companies control 88 per cent of the global market for semiconductor coater/developers, 53 per cent of silicon wafers, and 50 per cent of photoresist — a step without which no chip can be manufactured. The government has committed over ¥10 trillion (US$65 billion) to AI and semiconductors by 2030. For private investors, this means the state has already absorbed a significant portion of the infrastructure risk.

At the same time, Japan is candid about its challenges. Over 70 per cent of Japanese organisations report a shortage of technical talent — 23 percentage points above the global average. Only around 30 per cent of Japanese companies report measurable results from digital transformation, compared to 80 per cent in the US and Germany.

Japan is building a powerful foundation. But infrastructure without engineering talent to deploy it has limits.

Pressure-tested innovation

Ukraine is the other side of this equation. Over the past two years, the number of Ukrainian AI specialists has grown by 17 per cent, reaching 6,100 people. The deepest expertise is concentrated in NLP and computer vision. The total IT workforce stands at approximately 300,000. In 2025, Ukrainian IT exports reached US$6.66 billion — making IT the second-largest export sector. About 20 per cent of Fortune 500 companies have dedicated development teams in Ukraine. This is not a niche market. It is a scale, proven in practice.

Also Read: Deeptech’s secret: Ignore the market, master the engineering, and let opportunity find you

In parallel, a transformation has taken place that rarely gets noticed from the outside. Ukraine has become one of the leaders in digital governance. In the 2024 UN E-Government Survey, the country ranked 30th out of 193 states on the E-Government Development Index, and first in the world on the E-Participation Index.

An entire ecosystem has emerged around digital self-governance. Diia serves over 24 million users across 240 services. Diia.AI became the world’s first national AI assistant for public services. Diia.City, a legal framework that has attracted over 4,000 technology companies, 10 of which are unicorns.

Ukraine’s IT sector had been growing for years. But after 2022, the pace changed — not because of investment or market conditions, but because the stakes did. Engineers learned to build without a margin for error, with redundancy baked into everything, because failure had real consequences. That kind of pressure doesn’t just accelerate development. It changes what gets built and how.

This path was shaped under pressure, which is precisely why the solutions it produced have already passed the test of reality.

The case for systemic partnership

In AI, these two markets occupy different layers. Japan operates at the hardware layer (chips, robotics, industrial AI), Ukraine at the engineering layer (NLP, computer vision, GovTech architecture). Combining them closes a structural gap that neither country can close alone.

The zones where this fit holds are already visible. The most immediate is semiconductors and physical AI. Japan’s manufacturing precision meets Ukrainian software and algorithm engineering.

Also Read: To become better at prompt engineering, learn how to think like a manager

A natural next layer is robotics. Japan produces 38 per cent of the world’s industrial robots, and Ukraine has engineers who have built and deployed autonomous systems and tested them in difficult real-life conditions.

Joint R&D is another. Ukrainian teams are already embedded in Japanese industrial projects, but this is still point cooperation, not a systemic research pipeline.

The same logic applies to talent development — shared programs, structured internships, and long-term contracts that build pipelines rather than one-off engagements. And at the product level, both countries have something the other needs.

Japan has the hardware and the market access, Ukraine has the speed and the engineering culture to build globally scalable AI products.

On April 8 in Tokyo, we, as AI House with support from Roosh Investment Group, convened a panel discussion. The panel brought together government officials, business leaders, and researchers from both countries to examine a question: what Ukraine’s experience building a digital ecosystem under pressure actually looks like in practice, and where it connects with Japan’s own trajectory.

Such meetings are important not for what happens during them, but for what remains after — a shared understanding of where there is something real to build. That requires not one-off collaboration, but systemic engagement as long-term contracts, joint education programs, and structured exchanges.

The institutional groundwork is already in place. In 2023, Ukraine’s Ministry of Digital Transformation and Japan’s Digital Agency signed a memorandum on digital cooperation — covering cybersecurity, e-government exchange, and digital infrastructure. Yet the areas where both sides have something to offer go well beyond what that memorandum covers.

The trajectory of the AI economy is becoming clearer. More models mean more code, more compute, more chips, and more engineers. Demand at both layers will only grow. This partnership is about investing in people and systems.

As an investor, I look for structural logic — not just opportunity. Japan brings precision, depth, and the physical infrastructure of AI. Ukraine brings speed, adaptability, and engineers who learned to build without a margin for error. Individually, these are powerful. Together, they close a gap that neither country can close alone. That is the alignment I rarely see. Japan and Ukraine are exactly that case.

—

The post Can Ukraine’s engineers help solve Japan’s tech talent crisis? appeared first on e27.