Karrie's AI Revenue Is Real — the Vera Rubin Story Is the Part to Check

Karrie International is the sort of company the AI trade makes famous without most investors ever learning its name: a Hong Kong-listed maker of the metal and plastic boxes that hold the world's most expensive computers. In its June quarter, the AI slice of that business roughly septupled, and management now says revenue and profit are heading for records. The stock, already up 158% in a year, is betting the step-up holds. Before climbing into the same trade, it pays to separate what is actually showing up in the filings from what the company is telling you it expects.

The toll booth, not the boom

Karrie is a chassis and enclosure maker — server boxes, air- and liquid-cooled — whose customers are the branded server names rather than end buyers: Dell, Lenovo, HPE, and the contract manufacturers behind them. Its own marketing claims at least one in ten commercial servers shipped worldwide carries a Karrie-made chassis. This is a toll booth on the AI buildout. Every NVIDIA rack and every hyperscaler custom-silicon rack still needs a box around it, so money flows past regardless of which chip wins.

For most of the AI cycle, that toll revenue barely moved. For the financial year that ended in March 2026, group revenue rose a modest 7.4% to HK$3.47 billion, even as profit climbed at least 30% to roughly HK$271 million. The AI boom was real; Karrie's exposure to it just had not shown up. In the three months to June 30 it did. Total revenue rose 35% year over year, the general-and-AI server-enclosure line grew 63%, and AI-related products were up about sevenfold.

Read the driver, not the headline

Now the attribution, because it inverts the headline. The upbeat coverage would have you believe Vera Rubin — NVIDIA's new platform — carried the quarter. Read the company's own filing language and the documented drivers of that sevenfold jump are something else: higher shipments of ASIC servers, the custom-silicon racks cloud giants design for themselves, plus the start of deliveries of an "AI exchanger" product to another AI customer.

Vera Rubin is not what moved the June numbers. The company line that first Rubin-platform products are now delivering is a forward claim with no independent volume or customer evidence behind it yet, and management separately flags record revenue and profit this quarter with more growth into the fourth quarter.

That is the exact opposite of the story the headline sells.

To be fair, the surrounding evidence points the same direction as Karrie's hope. NVIDIA has guided that Vera Rubin will be roughly a fifth of its data-center revenue by the third quarter — the fastest product ramp in its history — and it has pushed through server price increases north of 15% on soaring memory costs. Both trends raise the dollar value of the mechanical content inside each system, which is the part Karrie sells.

The per-box reality, and what is already priced

But the economics are where the discipline lives. Karrie sells the enclosure — a sheet-metal-and-plastic slice of a rack that fetches millions — not the silicon inside it. That is not a technology business; it is a volume business running a net margin near 8% and an operating margin near 12%. Worse for the margins that matter, the ASIC channel actually driving this growth is the one where hyperscalers who design their own chips and racks are best positioned to squeeze a chassis supplier's price over time.

So the interesting question is not whether the AI step-up is real — it is — but what is already in the price. At a market value near HK$4 billion against trailing profit of about HK$271 million, the stock trades around 15 times trailing earnings, pays a roughly 3% dividend, and has more than doubled in a year. Management itself felt the need in July to tell shareholders their share price "may not fully reflect" fundamentals — a stress, in a small-cap with thin trading, that the narrative has outrun the audited numbers.

The honest summary: Karrie is now a genuine, verifiable-derivative play on the AI buildout — and for a retail investor that is actually the useful part, because most names in this chain are pure narrative. But the easy repricing has already happened. The two things that could break the trade are the two things hardest to verify: whether hyperscaler ASIC volume stays at this pace, and whether the Rubin story ever becomes more than a company statement. Buying today means paying for the part of the story that is not yet in the filings.