Summary

  • VNET reported 1,007MW of wholesale capacity in service, 970MW committed under effective agreements and 744MW utilized at 30 June 2026. The disclosed totals leave 226MW between commitment and customer use.
  • The gap widened from 182MW in March because in-service and committed capacity each grew by about 100MW during Q2, while utilized capacity rose by 57MW. That is an activation sequence, not evidence that contracts disappeared.
  • VNET plans RMB10bn-RMB12bn of 2026 capital expenditure while depreciation is already pressing GAAP gross margin. The next useful receipt is conversion of committed MW into utilized capacity, revenue and collected cash.

One denominator gives VNET two very different occupancy pictures. In its second-quarter filing, the company reports 1,007MW of wholesale capacity in service. It also reports 970MW committed to customers under effective agreements. That produces a 96.3% commitment rate, an unusually full-looking estate.

The same filing says customers were utilizing 744MW. Dividing that by the same 1,007MW base produces 73.9%. Commitment and utilization are both legitimate operating measures. They are not synonyms.

Subtracting the disclosed totals leaves 226MW between the two states, or 22.4 percentage points between their rates. VNET does not call that 226MW backlog, vacancy, stranded power or lost revenue. It is simply the arithmetic difference between capacity under an effective customer agreement and capacity in use. That distinction is enough to change how the quarter should be read.

A contract is earlier than a move-in

VNET defines committed capacity narrowly enough to matter: it is capacity committed to customers under effective agreements. Utilized capacity is the amount customers are using. Between those definitions sit physical and commercial events that a headline commitment rate cannot settle.

Management made the bridge unusually concrete in the earnings-call transcript. Asked about move-in, it named chip supply, the speed at which customers iterate their models and the progress of their projects. It expected move-ins to become marginally faster in the second half as domestic chip production ramps. That is a useful operating explanation and a forward-looking expectation, not a receipt for completed activation.

The Q2 sequence shows the distinction. In-service wholesale capacity rose from 907MW in March to 1,007MW in June, an addition of 100MW. Committed capacity rose from 869MW to 970MW, an addition of 101MW. Utilized capacity rose from 687MW to 744MW, an addition of 57MW. The commitment-minus-utilization difference therefore widened from 182MW to 226MW.

This is not, by itself, a deterioration. The company was adding delivered capacity and contracts faster than customers occupied it. Nor is it proof of future revenue. The analytical object is a moving corridor: facilities enter service, agreements attach to them and customer equipment then begins drawing service. The cash and profit clocks start according to contract terms and performance, which VNET does not disclose by MW cohort.

VNET says the quarter's 57MW increase in use came mainly from N-HB Campus 03 and N-OR Campus 01. It does not publish the price, power density, gross margin or collections associated with those MW. A megawatt is a capacity unit, not a standardized unit of revenue.

The portfolio contains two speeds

The Q2 presentation divides the in-service wholesale estate into mature and ramp-up facilities. Mature capacity represented 66.7% of the mix and was 92.5% utilized. Ramp-up capacity represented 33.3% and was 36.6% utilized.

Those labels deserve careful handling. VNET classifies wholesale data centres as mature when utilization is at least 80% and as ramp-up when it is below 80%. The categories therefore describe current utilization states; they are not independent predictions of eventual demand. A site can move between them as occupancy changes.

The split nevertheless locates the economics. VNET's mature estate is already heavily occupied. Most of the distance between a 96.3% commitment rate and 73.9% utilization lies in the newer, lower-use cohort. The investment case for that cohort depends less on signing another headline contract than on getting customer systems installed and revenue-producing on schedule.

That is why long leases do not close the argument. VNET says more than 90% of wholesale revenue is recurring and reports a seven-year weighted-average remaining lease term for contracted capacity. Duration helps visibility after the contract is operating. It does not identify the move-in date for the 226MW difference, the billing ramp or the cost borne before activation.

Orders and reservations sit on other clocks

The largest number in the quarter belongs to another perimeter. VNET won 345MW of new wholesale orders in Q2 from an unnamed leading cloud-service provider at N-OR Campus 03. Together with 517MW announced in Q1, four wholesale orders totalled 862MW in the first half.

The presentation maps those orders to an expected delivery schedule: about 287MW in 2026, 345MW in 2027 and 230MW in 2028 and beyond. Those three amounts sum to 862MW. They should not be added to the 970MW committed in-service total. The order ledger describes future deliveries; the commitment-rate denominator is capacity already in service.

VNET also reported 355MW of reservations, making 1,217MW of orders plus reservations. Reservations reflect customers securing capacity in advance, but the presentation does not put them inside the definition of committed in-service capacity. Combining the three ledgers would count unlike states as if they were one backlog.

Construction adds a fourth state. VNET had 585MW under construction with a 94.2% pre-commitment rate. It planned about 333MW of delivery in the second half of 2026 and 252MW in the first half of 2027. Pre-committed construction is better evidence than an unallocated land parcel, but it is not yet in service and cannot be part of current utilization.

Beyond that sits more than 4GW of reported wholesale resources: the operating estate, construction, capacity held for future development and a newly secured land bank. VNET defines secured land bank as locations with power quotas or resources earmarked by local governments. Power and land create an option to build. They do not establish engineering completion, customer acceptance or revenue.

Depreciation arrives before full occupancy

The financial statements show why the state sequence matters. Q2 wholesale revenue increased 29.3% year on year to RMB1.105bn, helping total revenue rise 14.2% to RMB2.779bn. Adjusted EBITDA increased 25.4% to RMB918.3m, while its margin rose from 30.1% to 33.0% as operating expenses fell.

GAAP gross profit moved the other way. It fell 7.8% to RMB505.2m, and gross margin declined from 22.5% to 18.2%. VNET attributes the reduction mainly to higher depreciation associated with rapid capacity expansion. Once an asset enters service, its accounting cost can begin before the new cohort reaches mature utilization. The 226MW corridor is therefore also a margin-timing problem.

VNET's adjusted cash gross profit excludes depreciation, amortization and share-based compensation. It rose 9.4% to RMB1.162bn, but the adjusted cash gross margin still fell from 43.6% to 41.8%. That measure can help isolate current facility cash costs; it cannot make depreciation economically irrelevant or turn adjusted EBITDA into cash flow.

Q2 operating activities generated RMB218.1m of cash, while purchases of property and equipment used RMB1.513bn. The company also borrowed RMB2.176bn and repaid RMB1.689bn during the quarter. Those selected lines are funding evidence, not a complete free-cash-flow calculation.

The next build is larger. VNET spent RMB3.55bn of capital expenditure in the first half and kept a full-year guide of RMB10bn-RMB12bn to support 450MW-500MW of 2026 delivery. If the guide is achieved exactly, RMB6.45bn-RMB8.45bn remains for the second half. At June it reported RMB7.214bn of cash, restricted cash and short-term investments, RMB8.497bn of unused credit and RMB23.42bn of debt plus convertible notes.

Liquidity makes the delivery plan possible; it does not make conversion automatic. Capital must be staged before customer use, while debt, depreciation and lease costs follow their own schedules.

The next quarter needs three receipts

The first receipt is operational: how many of the 970 committed MW become utilized, and how quickly the 585MW under construction enters the in-service denominator. A rising utilization rate would show activation outrunning new delivery. A flat rate alongside rapid delivery could still produce revenue growth, but would extend the time before the newer estate resembles the mature cohort.

The second receipt is financial: whether wholesale revenue and gross profit catch up with depreciation as move-ins progress. EBITDA growth alone cannot answer that question because it excludes depreciation and can benefit from lower operating expenses.

The third is concentration. VNET says its top 20 customers generated 59.4% of total revenue and does not identify the buyer behind the 345MW order. Long contracts can stabilize cash flows while increasing dependence on a small number of large deployment schedules. The useful disclosure is not merely another order total; it is the pace and economics of customer activation.

Sources