The capital keeps flowing toward Asia's contract manufacturers even as the marquee AI customers keep walking away from China. Luxshare Precision, the Shenzhen-based assembler best known as Apple's largest China-based iPhone partner, just priced the largest Hong Kong IPO of 2026 -- a $3.1 billion raise anchored by sovereign investors Temasek and GIC -- explicitly to fund its shift from smartphone assembly into AI data center hardware 1 2. Five months earlier, OpenAI had already pulled its first AI hardware manufacturing order away from Luxshare and handed it to Foxconn, citing concerns about assembling in mainland China 3. Both things are true at once, and that tension is the story for anyone pricing Asia's AI supply chain right now: the money is still pouring in, but the most China-sensitive customers are pricing an exit at the same time.

A pivot that predates the IPO

Luxshare did not become an AI infrastructure company overnight. Its communications and data center segment generated RMB24.57 billion in 2025, up 33.81% year over year, with gross margin improving two points to 18.4% -- a materially richer margin than the company earns on consumer electronics assembly 7. The business is no longer just cables. Luxshare's self-developed 224G KOOLIO CPC/NPC connector and its Intrepid NEXUS backplane and cable-cartridge interconnect systems are already running commercially inside AI clusters in China and abroad, and the company is layering in liquid cooling and rack-level power distribution up to 800V -- the full stack around the GPU, not the GPU itself 4. That is a deliberate bet that the scarcest capacity in AI data center buildouts over the next two years will not be silicon but the interconnect, cooling, and power hardware wrapped around it, and Luxshare wants to own more of that bill of materials per rack than any single competitor.

The first-quarter numbers back up the shift. Revenue rose 35.77% to RMB83.9 billion, net profit rose 20.24%, and the company has now guided first-half 2026 net profit to RMB7.84-8.11 billion, up 18-22% year over year 5 6. None of that growth is coming from a slowing smartphone business; Apple's own device volumes have been roughly flat this cycle. It is coming from data center and automotive electronics layered on top of the old assembly base.

The IPO is a war chest, not a bailout

A company growing net profit north of 18% does not need a $3.1 billion primary raise to survive. It needs one to build capacity fast enough to keep pace with hyperscaler capex, which is running at roughly $775-800 billion globally in 2026 and increasingly rotating out of pure semiconductor names into the server, cooling, and power suppliers around them 8. Temasek and GIC did not back this listing as a value play on iPhone assembly margins; they backed it as a bet that Luxshare converts a meaningful share of that capex rotation into recurring, higher-margin data center revenue. The Hong Kong listing also hands Luxshare a non-yuan currency and a second capital-raising venue outside the mainland exchanges -- useful for a company whose growth increasingly depends on customers who are actively trying to reduce their China exposure.

The customer that said no anyway

That is what makes the OpenAI reversal worth more than a footnote. OpenAI's first consumer AI hardware device, developed with former Apple design chief Jony Ive, was originally slated to be built by Luxshare. By January, the order had moved to Foxconn instead, with assembly reportedly shifting toward Vietnam or the US as part of a broader push to keep the device's supply chain outside mainland China 3. Foxconn is also positioned to pick up OpenAI's cloud-side AI server orders, extending its reach across both the data center and device sides of a relationship Luxshare lost before it had really started.

The lesson is not that Luxshare's data center strategy is failing -- the growth numbers say otherwise. It is that Asia's AI hardware buildout is fracturing into two supply chains that both happen to run through the same city blocks in Shenzhen and Zhengzhou: one for customers, largely Chinese hyperscalers and non-US brands, who will keep buying from mainland-based assemblers, and one for US customers increasingly unwilling to have their newest AI products carry a made-in-China label, regardless of engineering quality or cost. Luxshare's own numbers suggest the first supply chain is currently large enough on its own to fund a $3.1 billion IPO and 18%-plus profit growth. But the ceiling on that first supply chain is real, and every OpenAI-to-Foxconn-style decision narrows it further.

What investors should watch

The near-term signal to track is not Luxshare's headline growth rate -- it is the customer mix behind the data center segment's next earnings breakdown. If communications and data center revenue keeps compounding above 30% while the disclosed customer list stays dominated by Chinese hyperscalers rather than US names, that confirms Luxshare is winning the domestic side of a bifurcating market rather than closing the gap with Foxconn on Western orders. Watch too whether the $3.1 billion in IPO proceeds gets deployed into Southeast Asian capacity -- Vietnam, Thailand, or Indonesia -- rather than further mainland expansion; that would be the clearest sign Luxshare itself is trying to build a second, non-China-labeled supply chain to compete for the customers Foxconn is currently winning. For investors positioned in the broader Asian EMS complex, the more durable trade is not picking a single winner between Luxshare and Foxconn, but recognizing that both are now expanding capacity simultaneously to serve supply chains that are diverging by customer geography rather than consolidating into one.