
Summarise this article with:
Tor and a VPN both hide your IP address, but they do it through architecturally opposite approaches with very different security properties. Tor distributes trust across three volunteer relays so that no single entity can see both your identity and your destination. A VPN concentrates that trust in one server you control through your provider choice. Understanding the architecture is what lets you match the tool to the actual threat.

For the use-case decision guide (when to pick which tool in practice), see Tor vs VPN: Which One Do You Actually Need?. This post focuses on the technical architecture, encryption, threat model coverage, and honest speed numbers.
Architecture at a Glance
| Dimension | Tor | VPN |
|---|---|---|
| Hops | 3 relays (guard, middle, exit) | 1 server |
| Encryption layers | 3 nested layers, one per relay | 1 tunnel, AES-256-GCM or ChaCha20-Poly1305 |
| Trust model | Decentralized: no single relay knows both endpoints | Centralized: provider sees all traffic |
| Average throughput | ~5 Mbps (volunteer-limited) | Hundreds of Mbps on premium servers |
| Added latency | 250-400 ms typical | 20-50 ms typical |
| Traffic scope | Tor Browser only (by default) | All apps on the device |
| Cost | Free | Paid ($3-$13/month for reputable providers) |
| Exit node risk | Exit sees destination + unencrypted content | VPN server sees destination + content |
| Defeats global passive adversary | No | No |
| Defeats single subpoena to provider | Yes | No (depends on no-logs audit) |
How Tor's Onion Routing Works
Onion routing was designed in the 1990s at the US Naval Research Laboratory by Paul Syverson, David Goldschlag, and Mike Reed. The core idea: route a message through multiple relays, wrapping it in a separate encryption layer for each hop, so that peeling one layer reveals only the next hop, never the full path.
Circuit construction. When your Tor client builds a circuit, it negotiates a separate symmetric key with each of the three relays independently using Diffie-Hellman key exchange. The message is then encrypted three times: first with the exit relay's key (innermost), then the middle relay's key, then the guard node's key (outermost). Each relay decrypts only its own layer and forwards the rest.
The three relay roles:
- Guard node (entry). Knows your real IP address. Strips the outer encryption layer. Sees only the next hop (the middle relay). Does not know your destination.
- Middle relay. Knows neither your IP nor your destination. Strips its layer and passes the packet to the exit relay.
- Exit node. Strips the final layer and sends your request to the destination. Sees the destination and, if the connection is plain HTTP, the content. Does not know your IP.
No single relay holds both ends of the conversation simultaneously. That compartmentalization is the structural basis for Tor's anonymity guarantee.
Circuit lifespan. Tor builds a fresh circuit every ten minutes by default, and a new circuit is created for each new domain you visit. Each fresh circuit may route through a completely different set of relays and a different exit IP.
How a VPN's Encrypted Tunnel Works
A VPN creates a single encrypted tunnel between your device and a server operated by your provider. All traffic from all apps travels through that tunnel. The server forwards requests to the destination using its own IP, so websites see the server's address, not yours.
Modern protocol stack. WireGuard, now the dominant protocol used by most premium VPNs, runs as a kernel module and uses ChaCha20-Poly1305 for symmetric encryption and Curve25519 for the Diffie-Hellman key exchange. It performs handshakes with ephemeral keys (perfect forward secrecy): if a session key is ever compromised, past sessions remain protected. NordVPN's NordLynx wraps WireGuard in a double NAT system to avoid storing user IP assignments on the server, addressing one of WireGuard's original privacy concerns.
What the ISP sees. An encrypted stream to a known VPN server IP. Nothing else. Your ISP cannot see which sites you visit, but can see that you are using a VPN.
What the VPN provider sees. Everything that exits the tunnel, including the destination and, for plain HTTP, the content. A no-logs policy limits what they store. NordVPN completed its sixth independent no-logs audit (by Deloitte Lithuania, November-December 2025), confirming that no user activity or connection metadata is retained. An audit reduces risk but does not eliminate the structural fact that the provider is technically capable of seeing your traffic.
Threat Model Coverage
This is the most important table if you are choosing between the two tools:
| Threat | Tor | VPN |
|---|---|---|
| ISP sees which sites you visit | Blocked | Blocked |
| ISP sees you are using a privacy tool | Detectable (mitigated by bridges) | Detectable |
| Destination site logs your real IP | Blocked | Blocked |
| VPN/relay provider compelled by subpoena | Resists (no single relay has full picture) | Depends on no-logs policy and jurisdiction |
| Exit node reads unencrypted HTTP | Vulnerable (use HTTPS) | Vulnerable at VPN server |
| Global passive adversary (timing correlation) | Vulnerable | Vulnerable |
| Browser fingerprinting | Strongly mitigated by Tor Browser | Not mitigated |
| Account-level tracking (you log in) | Not mitigated | Not mitigated |
The global passive adversary caveat. Tor explicitly does not defend against an adversary that can monitor traffic at both the entry and exit of the network simultaneously. By correlating timing patterns at the guard node and exit node, such an adversary can deanonymize circuits without breaking any encryption. This is a documented, acknowledged limitation. State-level actors with broad network visibility represent the realistic version of this threat. VPNs have the same limitation: an adversary who controls both sides of the single-hop tunnel has an even easier correlation task.
The provider trust asymmetry. A VPN funnels everything through one legal entity in one jurisdiction. That entity can be compelled, hacked, or acquired. Tor distributes that exposure across three volunteer-operated relays, typically in different countries and under different legal jurisdictions. No realistic adversary can simultaneously compel all three to cooperate for a single investigation.
Speed: The Mechanism, Not the Marketing
Tor's speed limitation is structural, not incidental. Your packets travel through three relays, often in different countries, each of which must decrypt a layer, process the packet, and re-encrypt before forwarding. The bandwidth available to you is bounded by the slowest and most congested relay in your circuit, and all relays run on donated volunteer hardware.
Measured numbers from Tor Metrics. In 2025, typical user-level throughput on Tor was around 5 Mbps download. A 20 ms connection to a nearby server becomes 250-400 ms through a typical Tor circuit. This makes streaming video nearly impossible and video calls impractical. File downloads that would take seconds take minutes.
In my testing, Tor is perfectly usable for reading articles and loading static pages. Anything interactive or media-heavy is a poor fit.
VPN speed overhead. Modern WireGuard-based VPNs add 20-50 ms of latency on well-placed servers and impose minimal throughput overhead compared to unprotected connections. The kernel-space packet processing eliminates the context-switching penalty that slowed older protocols like OpenVPN. You can measure your actual connection speed with our Speed Test before and after connecting to quantify the difference.
Tor Browser's Built-in Protections
The Tor Browser is a hardened fork of Firefox that does more than route traffic through the Tor network. It deliberately standardizes signals that websites use for fingerprinting:
- Screen dimensions are rounded to coarse buckets rather than reporting exact pixel values.
- All Windows installs report the same Windows version; all macOS installs report the same macOS version.
- Canvas and WebGL reads are blocked or return identical spoofed values across all users.
- Tor Browser 11.5 (released July 2022) enabled HTTPS-Only mode by default, which prevents exit nodes from reading plaintext content on most sites.
The practical effect: all Tor Browser users look similar to websites rather than unique. This is the opposite strategy from a VPN, which does nothing to change your browser fingerprint. You can see what a website observes about your browser at our Browser Fingerprint tool.
Pluggable Transports: When Tor Itself Is Blocked
In countries where ISPs block connections to known Tor relay addresses (China, Russia, Iran), Tor offers pluggable transports that disguise the traffic pattern:
- obfs4 wraps Tor packets so they appear as random encrypted bytes. Even a censor scanning the entire IPv4 address space cannot distinguish an obfs4 bridge from any other encrypted service.
- Snowflake routes traffic through short-lived WebRTC proxies running in volunteers' browsers. To the censor, the connection looks like a WebRTC video call (the same protocol used by Zoom, Google Meet, and Discord). Blocking it would break those services too.
Neither pluggable transport defeats a sufficiently resourced nation-state adversary, but both raise the cost of blocking significantly compared to blocking known Tor relay IPs.
Tor Over VPN: Combining Both
Connecting to a VPN first and then opening Tor Browser creates a "Tor over VPN" configuration:
Your device > VPN server > Tor guard > Tor middle > Tor exit > destination
What this adds: Your ISP sees a VPN tunnel, not Tor traffic. The Tor guard node sees the VPN server's IP rather than your home IP. This configuration is useful if Tor usage itself is monitored or suspicious in your network environment.
What this does not add: It does not defeat the guard node if the VPN provider and the guard node cooperate. The VPN provider knows you are using Tor. The speed penalty of Tor is compounded by VPN overhead.
The reverse, "VPN over Tor," routes through Tor first and exits through a VPN server. It hides your IP from the VPN provider but is far harder to configure, slower, and supported by very few providers.
For most people, the combination is unnecessary. The speed penalty of Tor alone is already steep; stack both only if you specifically need to hide Tor usage from your ISP.
You can verify your current IP exposure and check whether your VPN is leaking your real IP through DNS or WebRTC with our VPN Leak Test and DNS Leak Test tools.
What Neither Tool Defends Against
Both tools stop working the moment you voluntarily identify yourself. Logging into your Gmail account through Tor hands Google your identity regardless of your exit IP. Logging into your bank through a VPN hands the bank your identity regardless of which server you used.
Neither Tor nor a VPN protects against:
- Account-based tracking (you log in with credentials tied to your identity)
- Malware on your device (the traffic is captured before it enters the tunnel)
- Legal compulsion against the destination site (the site can be asked what your account did)
- Browser exploits delivered through the exit node (use HTTPS, keep Tor Browser updated)
FAQ
What is the difference between Tor and a VPN technically?
A VPN creates one encrypted tunnel between your device and a single server. Tor wraps your traffic in three encryption layers and routes it through three separate volunteer relays, each stripping one layer. The structural difference is that no single Tor relay knows both who you are and where you are going. With a VPN, the provider server sees both.
Is Tor more secure than a VPN?
It depends on your threat. Tor provides stronger anonymity against an adversary who can subpoena or hack a single provider, because there is no single entity holding the full picture. A VPN provides better protection against most everyday threats (ISP surveillance, public Wi-Fi sniffing, ad tracking) with far less friction and much faster speeds. Neither is unequivocally "more secure" across all threat models.
Can Tor exit nodes spy on my traffic?
Yes, if the traffic is unencrypted HTTP. The exit node decrypts the final layer of onion encryption and forwards your request in plaintext if the destination only supports HTTP. Tor Browser 11.5 and later enables HTTPS-Only mode by default, which blocks connections to HTTP-only sites and substantially reduces this risk. On HTTPS connections, the exit node sees the destination domain but not the content.
Does a VPN protect against the same threats as Tor?
For most common threats, yes. A VPN hides your traffic from your ISP, protects you on public Wi-Fi, and prevents websites from logging your real IP. Where it fails relative to Tor: the VPN provider is a single point of trust that can be subpoenaed or hacked, and a VPN does nothing to protect against browser fingerprinting. For an ISP or a casual advertiser, a VPN is sufficient. For a threat model that includes a provider being compelled to log traffic, Tor is the stronger choice.
How slow is Tor compared to a VPN in practice?
Tor averages around 5 Mbps download throughput for individual users, with latency typically running 250-400 ms to servers that might otherwise be 20 ms away. A modern WireGuard-based VPN adds 20-50 ms of latency and delivers hundreds of Mbps on well-placed servers. The gap is roughly fifty times in throughput in independent speed tests. Tor is adequate for reading text and loading pages; streaming, video calls, and large downloads are impractical.
Sources
WhatIsMyLocation Team
Our team of network engineers and web developers builds and maintains 25+ free networking and location tools used by thousands of users every month. Every article is reviewed for technical accuracy using real-world testing with our own tools.
Related Articles
Try Our Location Tools
Find your IP address, GPS coordinates, and more with our free tools.