Everything About AI
NextFin News - Hong Kong is betting its artificial-intelligence ambitions on a data center complex that would multiply the city's computing power 36-fold, but the project's HK$20 billion ($2.6 billion) loan is hitting a global credit wall: banks that funded the AI infrastructure binge are now full, and the financing that would make the city's largest computing facility real has yet to be secured.
Range Intelligent Computing Technology Group Co., the Shenzhen-listed operator behind the Range (Hong Kong) Sandy Ridge data facility cluster, is in early talks with lenders over a one-year facility of around HK$20 billion, according to people familiar with the matter. The loan could take the form of a syndicated facility or multiple bilateral financings, the people said, adding that discussions remain preliminary and are subject to change.
The ask is large even by the standards of a sector that has just posted its biggest year on record. The project is a HK$23.8 billion commitment covering construction and the first three years of operation, tied to a 50-year land grant the company won from the Hong Kong government in March for a premium of $581 million. Construction began on March 28, 2026, and operations are expected to commence within 42 months.
The stakes for Hong Kong are explicit. The Innovation, Technology & Industry Bureau says the site — more than 110,000 square meters at Lot No. 953 in Demarcation District No. 89, Sandy Ridge, North District — will deliver a total gross floor area of 250,000 square meters, with 88% dedicated to high-tier data center use, and provide 180,000 peta-floating point operations per second of computing power by 2032. That is 36 times the city's current capacity, a figure officials cite as the backbone of Hong Kong's "AI+" push. The project is also expected to generate about $4.6 billion in economic output and create around 180 skilled positions in its first three years of operation. It will require roughly 220 megawatts of power.
Here is the tension: the city broke ground in March, but the debt that underwrites the build remains uncommitted as banks across Asia-Pacific report they have reached their data center lending limits. The financing test is not just Range Intelligent's problem — it is a stress test of whether Hong Kong can convert a policy blueprint into built infrastructure when the traditional lenders are stepping back.
The Credit Wall: Why Banks Are Full
The funding squeeze is a direct consequence of the AI infrastructure debt boom. At least US$334.5 billion in bonds and loans has been issued so far in 2026 to fund AI infrastructure projects, nearly double the US$185.5 billion raised across all of 2025. In Asia-Pacific alone, data center loans have reached nearly US$29 billion since the start of last year, with a record US$15 billion arranged so far in 2026 — a 27% jump from the same period a year earlier.
That volume has run into balance-sheet arithmetic. Credit Agricole is seeking to sell down about HK$150 million of a HK$1.6 billion loan it and other banks extended to ESR Group for a data center in Hong Kong's Kwai Chung district, after hitting its own lending cap for the sector. Morgan Stanley has explored shifting some of its data center portfolio risk to investors through a significant risk transfer. When the lenders that underwrote the boom start distributing exposure rather than holding it, new borrowers face a narrower door.
Andrew Ashman, who heads Barclays' Asia-Pacific loan syndicate team, put the constraint plainly: Asia-Pacific lenders "are reporting to be heavily exposed to the data centre sector," and as a result banks are likely to favor familiar borrowers or deals that offer higher returns. Richard Satchwell, Barclays' head of capital markets financing for the region, went further.
"We're looking at ways in which clients can recycle that capital into long-term bond markets like ABS and project bonds, as has been done globally," Satchwell said. "The bank market can't absorb the quantum of liquidity that's likely to be required in the data-center space."
The mechanism is straightforward. A one-year facility backing a multi-year construction program is precisely the kind of exposure banks are trying to shorten or offload. If Range Intelligent's facility lands, it will likely be priced to compensate lenders for holding a sector they already consider crowded — or it will be structured with distribution in mind from the start. The scale of what is being asked becomes clearer against the regional ceiling: Singapore-based DayOne Data Centers is in talks to double an existing loan to as much as US$7 billion, which would be Asia's largest data center debt financing. Range Intelligent's HK$20 billion ask sits within that order of magnitude, in a market where lenders are already telling borrowers to look beyond bank balance sheets.
Cyclical or Structural: A Regime Shift in Who Funds Infrastructure
Is the funding constraint cyclical — a temporary shortage of appetite that will revert once the current wave of deals clears — or structural? The evidence points to structural, and getting this call wrong flips the entire conclusion.
A cyclical squeeze would show up as a pricing dislocation that normalizes when deal flow slows. What is visible instead is a change in the funding model itself. Banks are not merely raising spreads; they are telling borrowers to access capital markets. The shift from bank balance sheets to asset-backed securities and project bonds is a permanent change in the plumbing of infrastructure finance, and it reorders the competitive landscape. Sales of asset-backed securities in the United States rose last year to the highest level in a decade, with offerings by fiber-infrastructure companies and data center owners jumping 86% to about US$23 billion — the recycling channel that Barclays is already pitching to Asian clients.
That matters for Hong Kong because the city's data center ambitions were built on an assumption of readily available bank credit. If the bank market cannot absorb the required liquidity, the projects that get built will be the ones that can be packaged for institutional investors: larger, pre-leased, sponsored by borrowers with track records and investment-grade-adjacent credit. First-time entrants, speculative sites, and developments without anchor tenants face a higher hurdle. Internal bank concentration limits are not a mood that passes; they are governance rules that persist across credit cycles.
The vacancy data underscores the demand side of the equation. CBRE reports Hong Kong's data center vacancy rate at 18% in the first quarter of 2026, down sharply from 28% in the first quarter of 2025, as leasing velocity absorbed available capacity. Demand is real. The question is whether the funding model can match it. Hong Kong hosts more than 100 data centers, ranks third globally for fiber connectivity and fourth in digital competitiveness, and is positioning itself as the gateway for cross-border data flows between the Greater Bay Area and global markets. China Mobile opened a flagship data center in March 2026 with around HK$10 billion of investment, and Equinix's HK6 facility — a US$124 million investment set to open in 2026 — is designed for high-density AI workloads with liquid cooling and 100% renewable energy. The ecosystem is being built in parallel; the missing piece is the debt.
The Counter-Thesis: The State Is Behind This, and Demand Will Clear the Deal
The strongest case against a pessimistic read is that Range Intelligent is not a typical speculative developer. It won a government tender that weighted non-premium criteria at 70% and premium at 30%, secured a 50-year land grant, and has the public backing of the Innovation, Technology & Industry Bureau. Only one tender was received, and the successful bidder met both the non-premium and premium requirements. Secretary for Innovation, Technology & Industry Sun Dong called the company the "pioneer in the city's first ever high-end data cluster" and said the government would "closely monitor the project's progress according to the service deed to ensure that the tender commitments are delivered as scheduled and with the required quality."
"The project's launch marks a milestone in Hong Kong's innovation push, and a decisive step in turning the Northern Metropolis from blueprint to reality," Sun said at the March 28 groundbreaking. "The company's capital commitment and concrete action signals a strong vote of confidence in Hong Kong's emergence as a magnet for high tech."
The counter-argument is coherent: a project this strategically important, with government land and monitoring behind it, will get funded because the alternative is a visible failure of the city's AI strategy. Banks may price it richly, but they will not let it stall. Hong Kong has strategic reasons to want this project built, and the Northern Metropolis is the centerpiece of the city's long-term development plan.
The answer is that the counter-thesis confuses political priority with bank capacity. A government that wants a project built can provide land, fast-track permits, and monitoring — but it cannot force a bank to hold a loan beyond its internal concentration limit. What state backing can do is improve the structure: credit enhancements, offtake commitments, or a bridge toward a capital-markets refinancing. The project likely gets done, but on terms that are more expensive and more complex than the original blueprint assumed. In other words, the deal clears — the cost of capital does not.
What the Funding Test Reveals About Who Gets to Build
The second-order implication is that the funding constraint acts as a selection mechanism. It does not stop AI infrastructure from being built; it decides who builds it. Operators with scale, pre-leased capacity, and access to capital markets will consolidate the sector. Smaller players and greenfield speculators will find the door closed, not because demand is absent but because the funding channel has migrated from bank relationship lending to securitized, institutional capital.
For Hong Kong specifically, the risk is a timing mismatch. The city's policy clock — the 42-month path to operations, the 2032 computing-power target — runs on government timelines. The capital-markets clock runs on deal windows, investor appetite, and the ability to package contracted cash flows into tradeable securities. If those clocks drift apart, projects slip even when demand is waiting. The 220-megawatt power requirement for the Sandy Ridge cluster is a reminder that physical constraints compound financial ones; land has been allocated, but power and financing must both arrive for the facility to energize.
Conclusion: Built, But on Different Terms
The funding test for Range Intelligent's HK$20 billion loan is a proxy for a wider question: can Hong Kong finance the infrastructure its AI ambitions require? The base case is that the facility closes, but as a distributed or partially distributed deal with pricing that reflects a crowded sector — a win for the project, but a signal that cheap bank money for data centers is over. The upside case is that state backing unlocks a credit-enhanced structure that draws institutional capital quickly, accelerating the build. The downside case is that distribution fails to find buyers at acceptable pricing, forcing the tenor shorter, the equity contribution higher, or the timeline longer.
Split by time horizon, the picture is mixed. In the short term, expect slower syndication timelines and wider spreads as lenders work through existing exposure. Over the medium term, the projects that advance will be those that can be refinanced into asset-backed securities or project bonds — the recycling mechanism Barclays is already pitching. Over the long term, the structural shift favors large, well-capitalized operators with pre-leased capacity and cross-border tenant demand; smaller players and greenfield speculators will find the door closed.
The falsifying signal is specific: if Range Intelligent's HK$20 billion facility syndicates fully with banks willing to hold the exposure at spreads close to pre-2025 data center norms, the funding-constraint thesis is wrong. Watch for the deal's structure — a heavy distribution component or a short tenor would confirm that banks are lending to sell, not to hold.
Hong Kong's data center dream is not short on land, policy, or demand — it is short on banks willing to hold the bill. The city's AI ambitions will be built, but they will be financed by capital markets, not balance sheets, and that changes who gets to play.
Explore more exclusive insights at nextfin.ai.
