Vultr (Constant)
Usage-based and reserved-capacity rental of compute (VMs, bare metal), GPU instances/clusters, storage, and networking across 33 cloud regions; self-serve SMB/developer base plus growing enterprise AI contracts
Vultr has exactly one priced equity round in its history (Dec 2024, $333M at $3.5B). It was bootstrapped 2014-2024 with no outside equity (though it used debt, e.g. a $150M Constant credit facility in Apr 2021, which carried no disclosed valuation). The Jun 2025 $329M package is debt/lease financing and does not re-price the equity. The trail therefore has a single true pricing point carried forward; the 2026-07 entry is a carry-forward marker, not a new round.
Earnings, margins, COGS & capex
Vultr is private and publishes no financial statements. The verifiable skeleton: founded 2014 by David Aninowsky (now Executive Chairman; J.J. Kardwell is CEO), no outside equity for a decade (it did use debt, e.g. a $150M credit facility in 2021) while building a profitable low-cost developer cloud (~$150M ARR per The Information as of Oct 2023). The AI pivot changed the capital model: $333M equity at $3.5B (Dec 2024, LuminArx + AMD Ventures — its first outside equity), then $329M of credit/lease financing (Jun 2025), then a reported $901M capital investment in a 50MW, 24,000-GPU AMD MI355X supercluster in Springfield, Ohio (announced Dec 2025; go-live timing not formally disclosed), plus an HPE-built Nvidia GB300 NVL72 selection announced at HPE Discover (Jun 2026, terms and timelines undisclosed). Revenue mix is shifting from granular self-serve cloud toward lumpy, contract-driven AI capacity.
Revenue trend
Margins
structurally pressured as GPU depreciation and power enter COGS vs. asset-light VM-era mix
claim predates the ~$900M Ohio buildout; unverified post-2025
almost certainly negative during the 2025-2026 capex wave (debt/lease-funded)
COGS structure
Primary COGS: GPU/server depreciation (AMD Instinct MI325X/MI355X fleets, Nvidia HGX/GB300 via HPE), data center colocation and power (incl. 5C's Springfield, Ohio campus), network transit, and support. The 2024-2026 shift from commodity x86 VMs to accelerator fleets raises COGS intensity and shortens asset refresh cycles (AMD MI450/Helios already on the stated adoption roadmap).
Capex
Historically modest (leased colo, commodity servers, 32 regions built without outside equity). Now the defining line item: ~$901M reported for the Ohio supercluster (24,000+ MI355X, 50MW, announced Dec 2025), 33rd region (Milan, May 2026), HPE/Nvidia GB300 NVL72 deployments (announced Jun 2026, terms undisclosed), financed via the Jun 2025 $329M credit/lease package and the Dec 2024 equity.
Latest earnings
n/a
None. Management has signaled an IPO 'sometime in the coming years' (CMO Kevin Cochrane to DatacenterDynamics, 2025, framing the Dec 2024 LuminArx/AMD round as 'setting us upon the path'); no S-1 filed as of Jul 2026
- Last priced valuation
- $3.5B (Dec 2024)
- Total outside equity raised
- $333M (single round, Dec 2024; no outside equity 2014-2024)
- Credit financing
- $329M (Jun 2025: $255M syndicated facility + $74M lease); prior $150M facility (2021)
- Cloud regions
- 33 (Milan added May 2026; 9th in Europe)
- Ohio supercluster
- 24,000+ AMD MI355X, 50MW, ~$901M reported investment, announced Dec 2025
- Customers
- 'Hundreds of thousands of customers across 185 countries' (company claim, unaudited)
Growth drivers
- AI training + inference demand overflowing from hyperscalers to independent 'neocloud' providers
- Flagship AMD partnership — early GA on MI325X and MI355X, stated MI450/Helios adoption roadmap, AMD Ventures on the cap table — preferred allocation of supply-constrained accelerators
- Ohio supercluster (24,000+ GPUs) converting capex into large reserved-capacity AI contracts as it ramps
- Price-to-performance positioning vs. AWS/Azure/GCP for cost-sensitive SMB, developer, and international workloads
- Geographic breadth (33 regions, 6 continents — company claims customers in 185 countries) enabling data-residency and sovereign-AI deals hyperscaler-averse buyers want
- Dual-vendor GPU strategy (AMD + Nvidia via HPE) hedging silicon supply and price
Bull & bear
The disciplined neocloud: a profitable, globally distributed cloud with a decade of real customers gets early access to AMD's accelerator roadmap and converts a modest $3.5B valuation into outsized upside if the Ohio supercluster fills and an IPO reprices it against CoreWeave/Nebius multiples.
- Entry multiple is low for the category: $3.5B (Dec 2024) vs. GPU-cloud peers that trade at far richer revenue multiples — CRWV and NBIS repriced the whole comp set upward after listing
- Capital discipline is differentiated: one equity round + $329M of bank debt vs. peers' tens of billions of leverage; less dilution and less refinancing risk if the cycle turns
- AMD partnership is a genuine structural edge: investor-supplier alignment, early GA on MI325X/MI355X, stated MI450/Helios pipeline — supply access without Nvidia's allocation politics or pricing
- Real, diversified demand base: hundreds of thousands of paying cloud customers (company claim) provide baseline utilization for GPU fleets that pure-play AI clouds must win contract-by-contract
- 33 regions is a moat neoclouds can't quickly copy — sovereign-AI and data-residency deals structurally favor distributed footprints
- Stated IPO path plus bulge-bracket banking relationships (J.P. Morgan, BofA, Goldman, Citi already lenders) gives a credible liquidity catalyst in 2026-2027
A ~$150M-ARR-scale company (last verified 2023) making a leveraged ~$900M bet on AMD silicon in the most competitive infrastructure market in tech; if ROCm adoption lags or GPU rental prices deflate, the Ohio fleet becomes fast-depreciating overcapacity and the $3.5B mark does not survive its next pricing event.
- Scale mismatch: last verified revenue (~$150M ARR, Oct 2023) is small against a single ~$901M capex commitment; even strong growth leaves the balance sheet carrying concentrated asset risk
- No disclosed contract backlog: unlike CoreWeave's multi-year committed contracts, Vultr's AI demand cover is unverified — utilization risk sits with Vultr
- AMD bet is double-edged: MI355X price-performance claims remain contested, CUDA lock-in persists across enterprise AI stacks, and resale/redeployment value of AMD fleets in a downturn is weaker than Nvidia's
- Competitive pincer: hyperscalers cut GPU prices and offer custom silicon above; CRWV/NBIS/Crusoe flood the dedicated-AI market beside; Hetzner/OVHcloud undercut the classic VPS business below
- GPU depreciation treadmill: MI450/Helios already on the stated roadmap within roughly 18 months of the MI355X buildout compresses the revenue window on the Ohio asset
- Debt is small but real and covenanted; a soft 2026-2027 AI-capacity market could force a down-round or asset sales before any IPO window opens
- The information vacuum itself is a diligence red flag at this capex intensity — no audited revenue, margin, or utilization disclosure exists to test the equity story
What it is worth
Last priced round + public-comp revenue-multiple triangulation (no disclosed financials, so all scenario math is assumption-labeled)
$1.5B-$2.5B (assumption-driven)
GPU rental price deflation or weak ROCm adoption leaves the Ohio fleet under-utilized against debt service; down-round or delayed IPO reprices toward developer-cloud (DOCN-like) multiples on the core business
$3.5B-$5B (assumption-driven)
steady core-cloud growth plus partial AI-capacity fill; next round or IPO modestly above the Dec 2024 mark
$6B-$8B+ (assumption-driven)
Ohio supercluster substantially contracted, AMD MI355X/MI450 economics validated, revenue scales toward a $500M+ run-rate, IPO prices at premium neocloud multiples (CRWV/NBIS comps hold)
Anchor: $3.5B post-money (Dec 2024, LuminArx + AMD Ventures — the only priced round). The last verified revenue datapoint is ~$150M ARR (The Information, Oct 2023); if 2024 revenue was in a $150M-$250M assumed range, the round implied roughly 14x-23x revenue — inside the post-CRWV/NBIS neocloud band but rich vs. DOCN (~4-5x). The next pricing event (IPO or new round) will mark it against 2026 AI-cloud multiples and, critically, against whatever utilization the Ohio supercluster demonstrates. All scenarios below are assumption-driven, not disclosed-financials-driven.
SWOT
Strengths
- Decade of profitable operation with no outside equity — rare capital discipline among GPU clouds that are burning billions
- Deep AMD alignment (investor + supplier) — early access to Instinct GPUs when Nvidia allocation favors bigger buyers
- 33-region global footprint gives latency, data-residency, and sovereign-cloud reach most neoclouds (CoreWeave, Lambda) lack
- Large diversified self-serve customer base (hundreds of thousands, per company) cushions against single-contract concentration
- Price-performance brand credibility with developers/SMBs built over 10+ years
Weaknesses
- No disclosed financials — last verified revenue scale (~$150M ARR as of Oct 2023) is small relative to a $3.5B valuation and a ~$900M single-site capex bet
- Balance-sheet firepower is a fraction of CoreWeave's (tens of billions in contracts/debt) or the hyperscalers'; must be selective, cannot win scale auctions
- Profitability claim predates the AI buildout — GPU-fleet economics (depreciation, utilization risk) are unproven at Vultr's new scale
- AMD-heavy fleet bets on ROCm ecosystem maturity; most enterprise AI tooling still defaults to Nvidia CUDA
- Thin public information and no audited disclosure raises diligence friction for large enterprise and government buyers
Opportunities
- Enterprise inference wave — cheaper-than-hyperscaler AMD capacity is well matched to cost-sensitive inference at scale (the stated focus of the HPE/Nvidia deal too)
- Sovereign/regional AI programs across its 6-continent footprint (e.g., Milan region for EU data residency)
- IPO window: a public listing could crystallize value and fund expansion if 2026 AI-infra sentiment holds
- AMD MI450/Helios generation could close the Nvidia performance gap and validate the dual-source strategy
- Upsell of 10+ years of self-serve cloud customers into managed AI/inference services (higher-margin layer)
Threats
- GPU price/oversupply cycle — if AI capacity outruns demand, spot GPU pricing collapses onto Vultr's freshly-financed fleet
- CoreWeave, Nebius, Crusoe, Lambda and hyperscalers all adding capacity faster and with cheaper capital
- Nvidia's own cloud ambitions (DGX Cloud) and hyperscaler custom silicon squeeze independent clouds from both ends
- Debt-funded capex meets unproven contract cover — no disclosed backlog equivalent to CoreWeave's committed-contract book
- Rapid GPU generation turnover (MI355X to MI450 within roughly a year) compresses useful life of the Ohio fleet
Moats, dependencies & bottlenecks
Moats
Years and hundreds of millions to replicate; underpins data-residency and latency-sensitive wins vs. single-site neoclouds
Real supply and roadmap access, but AMD also courts Oracle, Crusoe, TensorWave — preferred, not exclusive
10+ years of trust in the DigitalOcean/Linode-adjacent segment; sticky but price-sensitive by definition
weak-to-moderate Hundreds of thousands of accounts (company claim), but VPS workloads are portable; switching costs are low vs. hyperscaler ecosystems
24,000 GPUs is meaningful but an order of magnitude below CoreWeave/hyperscaler fleets
Dependencies
supplier + strategic investor Instinct GPU supply, roadmap (MI450/Helios), software stack (ROCm) and cap-table alignment; ROCm ecosystem maturity is the single biggest technical dependency
HGX/GB300 fleets for CUDA-demanding customers; allocation priority likely below larger buyers
supplier/integrator Builds Vultr's Nvidia GB300 NVL72 clusters with Spectrum-X networking and direct liquid cooling (announced HPE Discover, Jun 2026; terms/timelines undisclosed)
data center partner Hosts the flagship 50MW Springfield, Ohio supercluster — single-campus concentration for the AI bet
BofA, Wells Fargo, Citi, Goldman Sachs, KeyBank) $329M credit/lease package funds the capex wave; covenants and refinancing terms undisclosed
50MW draw exposes the supercluster to power pricing and interconnection timelines
Advantages
- Only major neocloud with a decade-long profitable operating history and a diversified pre-AI revenue base
- Early GA on successive AMD Instinct generations — cost-per-token pricing edge for inference
- Global footprint (33 regions / 6 continents) unmatched by any independent GPU cloud
- Low historical dilution: founder-owned into Dec 2024, one outside round — clean cap table into any IPO
- Dual-silicon (AMD + Nvidia) and dual-motion (self-serve + enterprise contracts) hedges
Weaknesses
- Opaque financials at exactly the moment its capex intensity demands scrutiny
- Sub-scale in AI capacity vs. CoreWeave, Nebius, and hyperscalers
- AMD-weighted fleet fights CUDA gravity in enterprise AI tooling
- Core VPS segment is commoditized with structurally low switching costs
- No disclosed long-term committed AI contracts to underwrite the Ohio buildout
Bottlenecks
- GPU allocation and delivery timing (AMD MI355X/MI450 ramps; Nvidia GB300 via HPE queue)
- Data center power: 50MW Springfield draw plus future expansion in a grid-constrained market
- Capital: ~$660M of disclosed equity + credit vs. multi-billion peer war chests limits how many superclusters Vultr can build concurrently
- ROCm software maturity gating enterprise adoption of the AMD-heavy fleet
- Enterprise sales motion: converting a self-serve SMB machine into a large-contract AI sales org
Top signals & trends
Top signals
Supplier-investor doubling down with capacity commitments is strong validation
Six major banks underwriting a private neocloud implies diligence passed on real cash flows
Signals valuation confidence and dilution aversion — but adds fixed obligations against uncontracted GPU demand
Hedges the AMD concentration and targets enterprise inference; financial terms, timelines, and locations undisclosed
'Coming years' framing since the Dec 2024 round; the window is open (CRWV/NBIS precedent) but Vultr hasn't filed
Either discipline or an unwillingness to test the $3.5B mark — unverifiable from outside
Core cloud business still growing organically alongside the AI pivot
Trends
Creates public comps and an IPO template for Vultr
Inference is price-sensitive and geographically distributed — both play to Vultr's footprint and AMD cost positioning
Vultr is the most leveraged independent cloud to this outcome, in both directions
Rental price deflation would hit recently-financed fleets hardest
Multi-region independents win deals hyperscalers and single-country neoclouds cannot
Compresses the merchant-GPU cloud price umbrella over time
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Instinct GPUs (MI325X/MI355X, MI450 roadmap), EPYC CPUs; also a strategic investor via AMD Ventures
HGX and GB300 NVL72 GPU systems, Spectrum-X networking
Server systems, direct liquid cooling, and integration for the Nvidia GB300 buildout
GPU server systems used in prior Instinct deployments
Springfield, Ohio campus hosting the 50MW supercluster
Networking silicon common across AI cluster fabrics
Hundreds of thousands of accounts across 185 countries (company claim) — the revenue base
GPU cluster and inference customers; individual names not disclosed
Served through the 33-region footprint
The scale leader in dedicated AI cloud; vastly larger contracted backlog and fleet
Fast-scaling AI cloud with public currency and large Nvidia fleets
Price cuts and Trainium silicon squeeze independents on big AI deals
Enterprise AI default via the OpenAI ecosystem
TPU alternative economics for large inference buyers
Aggressive GPU-cloud pricing and its own large AMD Instinct deployments
Closest public comp on the developer/SMB cloud side; also pushing GPU droplets
Direct developer-cloud rival with edge distribution
Private GPU cloud focused on AI developers; reported IPO candidate
Private, energy-advantaged AI data center builder; also an AMD deployer
European low-cost cloud; overlaps in the EU price-sensitive segment
Private German provider undercutting on classic VPS/bare-metal pricing