The synthesis

Developer-owner role: a Gowings-led Sydney-metro build, a specialty operator running the service tier on long-tenor triple-net, land held adjacent for one staged expansion. Deployed capital lands near AUD 1.26 billion at twenty-eight megawatts on the conservative capex base. Unlevered yield runs 8.7 per cent at steady state; levered IRR runs 12.7 per cent at fifty-five per cent LTV and 6.5 per cent senior coupon. The residual asset value of the building, the cooling, the connection, and the land sits on Gowings's balance sheet, not on an incumbent's.

At gigawatt-tier merchant build the cost arithmetic refuses. A Hunter site against ERCOT in central Texas runs structurally twenty to forty per cent more expensive on fully-loaded dollars per megawatt — wholesale electricity does roughly half the work, the rest is construction, equipment, and financing depth — and no Australian counterparty pays the premium that would close it for AI training fleets.

At frontier scale the comparator is ERCOT and the math fails. At a fraction of that scale the comparator is wrong. The boutique competes against the only thing actually offered to Australian premium-tier enterprise customers: the hyperscaler regions sited inside Australia — AWS Sydney, Azure Australia East, Google Australia Southeast.

Physically Australian and IRAP-PROTECTED, but multi-tenant at scale, operated under American parent governance, indirectly subject to the CLOUD Act and the wider American extraterritorial discovery regime, contractually generic on service tier, structurally indifferent to the small customer who needs more than a console and a support ticket.

Against that comparator a boutique Australian-owned, Australian-operated, single-tenant-feeling colocation tier with deep service can sustain a ten to thirty per cent premium depending on segment. Not a sovereignty mandate; a service-tier and governance-fit premium that APRA CPS 230, the Privacy Act reforms staged through 2025-26, cyber-insurance underwriters distinguishing Australian residency from offshore, and ASIC's Royal Commission and class-action precedent on chain-of-custody will sustain through any policy window a sovereign-mandate thesis depends on.

Seven boutique segments populate this tier: mining IP, legal sensitive, medical premium, APRA-regulated, professional services, family conglomerate, government mid-sensitivity. Each is small. None individually justifies a facility. In aggregate they support a standalone build sized for them and patient enough to fill it.

Capex is anchored at AUD 45 million per megawatt for the powered shell, stress range to AUD 60M — deliberately above the AUD 35 to 45M indicative range for premium colocation at Sydney metro. Captured premium runs five to thirty-five per cent over the AUD 4.5M/MW/year hyperscaler-Sydney baseline (the mid-range AirTrunk Sydney and NEXTDC S3 wholesale rate), with eighteen per cent the central case. Anything below ten per cent retires the thesis on its own.

Move the inputs. The bound below which the thesis stops paying is roughly six per cent unlevered; the bound above which the project runs ahead of any defensible boutique base case is about twelve. Most live combinations between sit inside the recommendation envelope.

Long-tenor REIT-style equity. Investment-grade senior debt on the powered shell at fifty-five per cent LTV. A specialty operator on a ten-to-fifteen-year triple-net with rent escalators tied to power and service inflation. The residual asset — building, cooling, connection, land — underwriting the terminal value.

Customer relationships sit with the operator. The GPU obsolescence cycle that breaks the merchant case sits with the tenants. The role maps onto the real-estate development and long-tenor asset experience the firm has been building for over a century. Macquarie Park, Eastern Creek, and Erskine Park each carry connection capacity at this scale today; site selection is real but soluble, not the substance of the underwriting.

Hunter sovereign co-investment, fifty to two hundred megawatts. Silent infra slice alongside Canberra Data Centres or AirTrunk on a Defence-anchored Hunter project. Defensible at seven to nine per cent unlevered, but the customer relationships, brand, residual asset value, and operating optionality all sit with the incumbent. Worth taking if a deal of this kind lands and the diligence below answers cleanly. The pack does not organise around finding it.

Listed exposure, residual. A patient, disciplined entry into NEXTDC at a sensible price provides liquid Australian datacentre exposure in size. The residual route if conviction in the development role lapses, not the position the rest of this pack underwrites.

Before a boutique standalone clears for capital, six questions need concrete answers. Each fits on a single line, or the project is not yet underwritable.