AXG — FY2026 Financial Report Analysis

Revenue up almost tenfold, but it is mostly resold cloud computing at a thin margin

Solowin Holdings, a Hong Kong broker that rebranded itself in fiscal 2026 as a "digital asset and AI" platform, reported revenue of $28.0 million for the year ended March 31, 2026, up from $2.8 million (+895%). Almost all of that jump came from one new line, AI infrastructure services ($22.2 million, 79% of revenue), which arrived with AlloyX, a company bought in September 2025 from the group's own chairman and CEO in exchange for about 130 million new shares. Costs grew faster than the new revenue could cover: the operating loss widened to $13.0 million from $8.1 million, and the net loss to $13.3 million from $8.5 million (+56%).

At a glance

• $22.2M of AI revenue against $21.3M of AI cloud costs. The AI business keeps about 4 cents of each revenue dollar after paying its cloud supplier, and still lost $5.1 million before tax once staff and overheads are counted.
• Loss per share fell to $0.11 from $0.53, but only because the share count went up almost eightfold. The weighted-average share count rose from 16.1 million to 124.9 million, mainly from the AlloyX deal. The total loss grew 56%.
• $15.3M of cash burned in operations, funded by $21.9M of new money. That was $16.9 million of equity from investors and $5.0 million of convertible notes. Year-end cash was $16.8 million.

Key figures

All figures in US dollars. Solowin reports in USD, although its Hong Kong subsidiaries operate in Hong Kong dollars.

Operating margin is operating profit or loss as a share of revenue. Here it is negative: for every dollar of revenue, the company spent about $1.46 to run the business.

What actually drove revenue

AI Tokens segment: $22.2 million, new this year. Solowin sells enterprise clients a bundled cloud and AI service (design, cloud deployment and ongoing support) and buys the underlying computing from outside cloud suppliers. The filing says the company acts as the principal, meaning it books the full price the client pays as revenue and the supplier's bill as a cost. A reseller that acted as agent would record only its fee. That choice is why revenue looks so large next to the costs:

The business also only runs through part of the year. AlloyX's results are consolidated from July 3, 2025, the date the filing says Mr. Ling Ngai Lok gained common control of both companies. The 2025 comparison year therefore contains no AI revenue at all, so the 895% growth rate compares a new business against zero rather than measuring growth in the old one.

Digital Asset Tokens segment: $5.84 million, up 107%. Most of the increase was also new. Tokenization service fees of $2.72 million (turning assets such as fund units into blockchain tokens) came from AlloyX. Without that line, the legacy businesses (brokerage, asset management and consultancy) brought in $3.12 million, against $2.82 million a year earlier. Within that:

• Virtual-asset trading income rose to $305,000 from $15,000, and referral income of $490,000 appeared after none the year before. Securities brokerage commissions fell 48% to $57,000 "due to lower frequency of trading activities in the U.S. market."
• Investment advisory fees fell 34% to $680,000 "due to a reduced client base." Asset management fees fell 9% to $593,000 on lower performance fees from its Solomon Capital fund.

Operating figures for the new platforms grew from small starting points: $226 million of payment volume on AX ONE, $52 million of value tokenized across 10 projects on FERION, and $1.04 billion of stablecoin and fiat trading volume (up from $210 million). None of these is revenue. They show activity on the platforms, and the filing does not say how much revenue each one produces.

What the headline numbers hide

• Five AI clients account for about three-quarters of revenue. The top five customers made up roughly 25%, 23%, 15%, 6% and 5% of revenue, all of it from AI infrastructure. The three largest (Customers G, H and I) contributed $17.9 million. A single supplier ("Supplier A") accounted for $18.6 million of costs, or 47% of all expenses. In practice, the bulk of revenue depends on three client relationships and one supplier contract.
• Receivables grew much faster than sales. Receivables (money customers owe but haven't paid yet) rose to $11.3 million from $146,000. Customers G, H and I alone owed $9.8 million at year-end, equal to about 44% of the full year's AI revenue, even though the stated credit period is normally 90 days. On the other side of the balance sheet, Solowin owed its cloud suppliers $10.4 million (96% of it to Supplier A). The business is effectively a pass-through: if those clients pay slowly, Solowin still owes its supplier.
• Cash conversion was poor. Operations used $15.3 million of cash against a $13.3 million net loss. The filing names the loss and a $2.2 million drop in amounts payable to brokerage customers as the main drivers. Year-end cash of $16.8 million came from $16.9 million of new equity, $5.0 million of convertible notes and $5.3 million of cash that came with the acquired subsidiaries, not from the business itself.
• The per-share "improvement" is dilution. Weighted shares rose 674%. Shares outstanding went from 16.5 million (Class A plus Class B) to 189.0 million at year-end, including 130.1 million issued for AlloyX and 7.5 million issued for a 48% stake in Tiger Coin. More shares have followed since: 2.5 million to Streeterville Capital under a prepaid share-purchase agreement of up to $100 million, and 1.65 million (valued at about $5.4 million) to consultants in June 2026.
• Share-based pay is a cost in both years. Non-cash share awards were $3.2 million in FY2026 (in staff costs, which the filing links to 750,000 shares awarded to an employee) and $3.3 million in FY2025 (in professional fees, paid to a consultant). By our calculation, excluding them, the loss roughly doubled to about $10.1 million from about $5.2 million. That makes the deterioration look worse, not better. The company itself does not publish an adjusted figure.
• A related-party deal, booked at historical cost. Because the CEO controlled both sides, AlloyX was recorded at its book value ($5.6 million of net assets, mostly cash) rather than at market value. So the deal created no goodwill and carries no impairment risk on the balance sheet, even though it was publicly announced at a stated value of $350 million. The filing itself warns that the company "may have paid more consideration than we would have paid in an arm's-length transaction." A side arrangement could pay an entity controlled by the CEO up to $10 million if valuation milestones are met by September 2027. No expense has been booked for it yet.
• The auditors found a material weakness. This is a serious gap in internal controls: the filing says Solowin lacks enough accounting and financial-reporting staff with the necessary knowledge, and lacks comprehensive policies and procedures, for applying SEC rules. The 20-F itself arrived after a notice of late filing.
• Prior year had small one-offs. FY2025 included a $290,000 investment write-down and a $100,000 loss on selling an associate, and FY2026 had neither. These are small compared with the swing in operating costs.

Takeaway: The tenfold revenue jump is real cash flowing through the books, but Solowin keeps only about 4% of it after paying its cloud supplier, and three customers who owed $9.8 million at year-end generate most of it. At this margin, revenue growth cannot cover a $41 million cost base. The company is funding the gap by issuing shares, and the share count grew almost twelvefold in a year.

What to watch next

Management gives no revenue or profit guidance. The 20-F lays out a 12–24 month plan instead: start commercial issuance of its AXUSD and AXBHD stablecoins under the Bahrain stablecoin issuer license granted in June 2026, move bank and payment partnerships (several still non-binding memoranda) into production, and grow the AI client base.

Our read on the trajectory:

1. Whether the AI customers pay. The next interim report (six months to September 30, 2026) should show whether the $9.8 million owed by the three largest clients turned into cash, and whether the cloud-supplier payable came down with it. If receivables keep climbing, that matters more than further revenue growth.
2. Whether the AI spread widens. At about 4%, the AI line cannot cover its own staff costs. The company has to price above its compute cost or add higher-margin software (its KovaRouter model-routing product) to change that. A full year of AlloyX in FY2027 will make revenue look bigger again, so watch the margin rather than the top line.
3. Stablecoin revenue actually starting. The license is in hand, but the filing says commercial issuance remains "subject to all required approvals." Until reserve income and minting fees appear in the income statement, the Digital Asset segment is mostly the old broker plus $2.7 million of tokenization fees.
4. Dilution pace. With $16.8 million of cash, a cash burn of $15.3 million last year, a $100 million prepaid share-purchase facility and prospectus supplements filed in June and September 2026, more share issuance is the most likely way the next year gets funded.