VNET: A Data Center Developer Dressed Up as an AI Company — What's It Actually Worth?
Three REIT deals, a $14 billion customer, and a leveraged subtraction problem with a 10x multiplier
VNET Group is a third-party, carrier-neutral data center operator based in China, listed on Nasdaq via an ADR structure. Last quarter, wholesale revenue grew 58% year-over-year. The AI narrative is in full swing. Underneath the headline growth numbers though, valuing VNET is a subtraction problem — big number minus big number equals small number — and small numbers are extremely sensitive to small changes in the inputs.
This article ignores sell-side consensus and builds entirely from public data and completed transactions to break down VNET’s business model and valuation framework.
This Is Not a Tech Company
VNET sells power capacity and physical space. Someone buys GPUs and suddenly needs to plug them in, cool them down, and turn them on. VNET is selling that somewhere. They don’t design chips. They don’t build algorithms. They don’t write software. VNET is a landlord renting out factory floors to AI training clusters.
Except it’s not your typical landlord. In a data center, ~20-30% of total construction cost is spent on the concrete shell. The other ~70-80% is mechanical and electrical; transformers, UPS systems, precision cooling, chillers, backup generators. Equipment with 10-15 year useful lives. Annual D&A of roughly RMB 24-25 billion eats most of the EBITDA.
ROIC hovers around 9-13%, leaving only 1-3 percentage points of excess return above the cost of capital. Each additional MW of capacity requires the same capex. This isn’t a tech business benefiting from scale economies. It’s a toll road. It’s a power plant.
What VNET Looked Like Before AI
From 2021 to 2023, revenue grew 2-3% per year. EBITDA margins declined from 28.3% to 26.5%. Cabinet utilization sat at 55-59%. The wholesale business barely existed. A mediocre company with no moat.
Then AI happened. Wholesale revenue grew from 19% of total in Q1 2024 to 39.5% by Q1 2026, growing 58% year-over-year. Wholesale capacity expanded from less than 200MW to 907MW. EBITDA margin recovered back up to 33%.
This transformation was 100% driven by an external demand shock. AI compute demand exploded, creating a supply-demand gap, and VNET happened to have some land, power quotas, and a relationship with at least one major customer. It’s not selling irreplaceable capability — it’s selling a time arbitrage. Self-building takes 18-30 months; leasing from VNET takes 3-6. Time arbitrage has an expiration date.
A One-Customer Story
Of the 517MW in new wholesale orders signed year-to-date in 2026, approximately 510MW came from a single “leading internet customer.” Over 98% of incremental demand from one buyer. This isn’t a broad market undersupply story — it’s one specific customer’s explosive demand landing on VNET’s doorstep.
What Three REIT Deals Reveal
VNET completed three data center securitization transactions in 2025-2026 totaling roughly RMB 7.2 billion in issuance, with underlying assets in a tier-1 city, Taicang in Jiangsu (~210MW), and Ulanqab in Inner Mongolia.
Here’s the data point that jumped out at me: in late 2024, strategic investor Dajia Holdings paid to acquire a 49% interest in the Taicang project at an implied valuation of RMB 5.74 billion for 210MW, or approximately 10.1x EV/EBITDA. You can back into EV/MW directly (~RMB 27.3 million) without needing to estimate EBITDA margins or pick your own multiple. This isn’t a hypothetical; an actual institutional buyer paid real money for an equity stake. By the time the REIT listed in March 2026, management said the multiple had risen to 13-14x. That’s 30-40% expansion on the identical asset in eighteen months, reflecting rising institutional appetite for data center assets.
GDS’s P-REIT revealed its full economic waterfall: EV of RMB 2.9 billion to ~RMB 1.2 billion project-level debt (~41% LTV) to RMB 1.7 billion equity consideration, with GDS retaining 30%, leaving just RMB 500 million cash received upfront and RMB 700 million contingent on milestones. A headline EV of RMB 2.9 billion ultimately translated into roughly RMB 500 million of net cash for the parent — a 17% conversion rate. VNET’s management guided total 2026 REIT-related cash proceeds of “no less than RMB 2 billion” against RMB 6.36 billion in total issuance — roughly 31%. The capital recycling is real, but far slower than the headlines suggest.
The 10x Leverage Amplifier
VNET’s total operating asset base is valued at roughly RMB 30-45 billion. Add up all the senior claimants ahead of common equity — bank loans, finance leases, convertible bonds, minority interests — and you get approximately RMB 26-29 billion. Common equity isn’t a large number. It’s the razor thin difference between two very large numbers.
If you move from EV/EBITDA of 10x to 17x, it looks like a ~70% change on the surface. Account for leverage amplification, and the value of common equity can differ by a factor of 5-7x. This isn’t a hard math problem — anyone can subtract. It is however a great example of extreme sensitivity to input assumptions.
And the single most impactful variable in the entire debate — wholesale segment EBITDA — has never been separately disclosed. The weakest piece of evidence drives the largest swing in output.
Steady-State Earnings Under GAAP
What if you skip adjusted EBITDA and start from GAAP?
Annualized EBITDA is roughly RMB 3.57 billion. Once you account for D&A (~RMB 2.4-2.5 billion), interest (~RMB 0.8-1.2 billion), taxes, and minority interests, net income attributable to common shareholders is approximately zero. FY2025 GAAP net loss was RMB 133 million. EBITDA looks decent; net income is negative. The gap is entirely consumed by depreciation and interest.
Even assuming all capacity under construction is completed and fully leased at 90% utilization, company-wide steady-state GAAP net income comes to roughly RMB 7-8 billion per year. But the wholesale segment won’t turn genuinely profitable until around 2028 — new capacity triggers immediate D&A and interest while revenue waits for customer move-in. Once it crosses breakeven, though, operating leverage hits hard: a 50% revenue increase could drive a 400-500% profit increase. The reverse is equally true.
Two Anchors
I price VNET from two completely different directions.
The defensive valuation asks: what if the REIT window closes, AI growth falters, and VNET is left collecting rent from existing assets? Using GAAP steady-state net income in a DCF that incorporates gradual construction delivery and the profit inflection, total equity value comes to roughly RMB 6-7 billion. About $3/ADS. This is the floor — paying only for cash flows that have already occurred or can be verified, with zero premium for expectations.
The offensive valuation asks: what if the REIT window stays open and multiples hold? Anchoring EV/MW to the Pre-REIT transaction price, adjusting for liquidity discount and multiple expansion, and probability-weighting across scenarios, then subtracting all claims, common equity comes to roughly RMB 16 billion. About $8/ADS. This requires everything to go right — a hot REIT market, stable multiples, a loyal anchor customer, and sustained AI demand.
What $9.23 Is Paying For
$3 to $8 is the range that public data can support. $8 already requires the REIT window to stay open, multiples to hold, and Pre-REIT liquidity discounts to fully correct. The current price of $9.23 sits above even the offensive estimate — meaning the market is paying for the 1,056MW future development option, continued REIT arbitrage, or higher-than-observed multiples.
All of that might materialize. But the pre-AI track record — 2-3% growth, declining margins, 55% utilization — shows exactly what this company looks like when the tailwind stops. And the capital structure guarantees that any disappointment gets amplified, not cushioned.
$3 is the floor. $8 is the ceiling with full tailwinds. Above $9, you’re paying for faith.
Disclaimer: This article reflects personal research notes only and does not constitute investment advice of any kind. The author may or may not hold positions in the securities mentioned. All data is sourced from public filings (SEC, Shanghai Stock Exchange), company earnings calls, and publicly available news reports. Accuracy and completeness are not guaranteed. Investing involves risk. Please exercise independent judgment and assume full responsibility for your own decisions. This article does not represent the views of any institution.