Peplink's admin interface looks like it was designed by network engineers for network engineers — because it was. The dashboard gives you a clean overview of WAN status, tunnel health, and throughput. That part is fine. The problems start when you need to actually configure anything. The outbound policy editor — where you define which traffic gets bonded — has an "Expert Mode" toggle that unlocks 15+ match criteria (source IP, destination IP, protocol, DSCP, domain, application, MAC address, VLAN) with three routing modes (enforced, priority, weighted balance) and the ability to chain rules in priority order. It's incredibly powerful. It's also the kind of interface where you can spend an entire Saturday building a policy stack, only to realize that rule #4 is silently overriding rule #7 because of how Peplink evaluates the chain. SpeedFusion tunnel settings are similarly deep: WAN smoothing levels, FEC profiles (static at 13.3% or 26.7% overhead, or adaptive at 6.7–20%), packet duplication toggles (Normal = 2×, Medium = 3×, High = all WANs), cut-off latency thresholds (the Packet Jitter Buffer defaults to 150 ms — this is the literal latency floor SpeedFusion adds to bonded traffic — and Peplink recommends keeping bonded WANs within 150 ms of each other for the same reason), health check intervals, bonding algorithm selection — each with meaningful performance implications and minimal in-UI guidance on what the right settings are. FEC and duplication apply per-tunnel rather than per-flow, so once a level is set, the overhead applies to every packet that rides the bonded tunnel — including the ones that don't need protection. InControl 2 (the cloud management platform) is genuinely good for monitoring and fleet management, but it adds another layer of interface on top of the local admin panel, and the two don't always surface the same settings in the same way.

Hardware. Peplink only. SpeedFusion is embedded in Peplink firmware and can't be installed on third-party routers. The relevant models for home/SMB bonding are the B One 5G (~$500), Balance 20X (~$350, no built-in modem), and the MAX Transit series (~$800-1,200 for mobile). Enterprise models go much higher.

Relay server. SpeedFusion requires a relay endpoint to terminate the bonding tunnel. You've two options:

SpeedFusion Cloud (hosted by Peplink): Peplink operates relay nodes across AWS, Azure, and DigitalOcean in dozens of locations worldwide. Your router can auto-select the nearest node or you can pin to a specific one. Pricing is per-device and varies by router tier. For the B One 5G, mid-tier data plans run around $20/month, and truly unlimited is $99/month or $850/year. After exceeding a data cap, service drops to 10 Mbps rather than cutting off entirely. Plans are tied to a single device serial number and don't stack. All active PrimeCare subscriptions include a basic SpeedFusion Cloud data allowance, but it's typically small (500 GB on entry-tier hardware) and runs out quickly under real bonding use.

FusionHub (self-hosted): A virtual appliance you run on your own server or cloud VM. One-time license purchase: $299 for 100 Mbps, scaling up for higher throughput. You control the server location, the bandwidth, and the cost — but you're also responsible for uptime, maintenance, and sizing. Good option if you want to minimize relay latency by hosting in a nearby data center, or if you need bonding throughput beyond what SpeedFusion Cloud provides.

Pricing model. SpeedFusion's cost structure is layered in ways that catch people off guard. The router is the upfront cost. PrimeCare is the mandatory subscription that unlocks SpeedFusion features — without it, your $500+ router can't bond. Pricing varies by hardware class, ranging from $49/yr for entry-level routers up to $1,299/yr for enterprise hubs; the B One 5G sits in the middle at $99/yr. SpeedFusion Cloud or FusionHub is the relay cost. And you still need your own WAN connections (SIM plans, ISP lines) on top of all of that. Nothing is bundled. Every layer is a separate purchase from a separate product page.

Year-one cost estimate. Hardware-dependent, and the flexibility is the point. Entry tier: roughly $600 upfront for low-end hardware, then $49 PrimeCare with 500 GB of SpeedFusion valid for 6 months, plus $20 top-up after that — call it ~$70/yr in platform fees, but throughput is limited and the data allowance runs out fast. Mid tier (B One 5G class, what we use for Waveform Internet): $500 hardware + $99/yr PrimeCare + ~$20/mo SpeedFusion Cloud = ~$340/yr in recurring platform costs, with bonded throughput around 200 Mbps under AES-256. Enterprise tier: tens of thousands upfront for hardware that can push multi-gigabit bonded throughput, with PrimeCare and SpeedFusion costs scaling correspondingly. Connectivity (SIM plans, ISP lines) is separate at every tier.

The gotchas. PrimeCare isn't optional — SpeedFusion features are locked behind it, and if your PrimeCare license lapses, your bonding tunnel stops working even though the hardware is perfectly functional. SpeedFusion Cloud server selection defaults to "automatic," which usually picks the nearest server, but occasionally picks one that adds 50+ ms of latency for no obvious reason — pinning to a specific server is often necessary but not the default behavior. The B One 5G's CPU is the throughput bottleneck on prosumer hardware, not your WAN speeds — if you're seeing 80 Mbps bonded when your individual WANs each do 100 Mbps, the answer is almost always CPU saturation or encryption overhead, not a tunnel problem. SpeedFusion's TCP ramp-up algorithms also stall in cellular environments — carrier-side time-slicing and resource windowing at the tower mean the bonded tunnel can take up to 10 seconds to fully saturate available bandwidth on a fresh connection, which is invisible on a steady stream but very visible on a fresh page load or a new connection. Link failure detection takes around 6 seconds for SpeedFusion Cloud connections — long enough that in-flight packets on the failed WAN are lost during that window, though WAN Smoothing's duplication covers the gap if it's enabled. And when a WAN's quality degrades far enough, DWB suspends it entirely (rather than just reducing its allocation) and sends only probe data on the suspended link to detect recovery, which can look alarming on a dashboard ("why is this WAN showing no traffic?") even when the platform is working as designed. Firmware updates also occasionally change tunnel behavior in ways that require re-tuning your policies.

What people complain about most. Peplink's official documentation is consistently weak — sparse, inconsistent across firmware versions, and lacking the kind of how-do-I-actually-do-this guidance most platforms ship with. They lean heavily on the community forum instead, which is genuinely active but quality varies wildly by topic — some threads are gold, others are years-old half-answers. It works as a support model only because the community is engaged. The most common complaint themes:

• "SpeedFusion is destroying my throughput." The single most frequent complaint. Users expecting their bonded throughput to be the sum of their WANs are shocked when it's lower than either WAN individually. Concrete cases reported in the wild: 310 Mbps + 230 Mbps dual fiber bonds to only 170–180 Mbps; turning on WAN Smoothing can drop a 100 Mbps connection to 67 Mbps; under maximum FEC and duplication a 100 Mbps link can deliver as little as 20–25 Mbps. Encapsulation alone adds about 80 bytes per packet — roughly 5% on full-size 1500-byte packets, but 200% on 40-byte VoIP packets and ~19% efficiency loss on a typical mixed-traffic profile. Causes vary — CPU saturation, distant relay server, encryption overhead, misconfigured WAN smoothing — but Peplink's documentation does a poor job of setting expectations, and "my internet is slower with bonding enabled" remains a recurring forum thread.
• "I spent a weekend configuring outbound policies and something is still wrong." The complexity of the policy engine is a double-edged sword. Users frequently report policies silently conflicting, rules not matching traffic as expected, or subtle interactions between SpeedFusion routing and the regular outbound policy chain.
• "My PrimeCare expired and now bonding doesn't work." This is the one that generates genuine anger. Users who paid $500+ for a router discover that the bonding features they bought it for are locked behind a recurring subscription that wasn't obvious at purchase.
• "SpeedFusion Cloud auto-selected a terrible server." The automatic server selection usually works, but when it doesn't — picking a server 3,000 miles away when there's one 200 miles away — the latency impact is dramatic and the cause is opaque.
• "Is SpeedFusion bonding even working?" A recurring category of posts from users who can't tell if their setup is functional or if they've configured something wrong. The diagnostic tools exist (tunnel analyzer, WAN quality graphs), but interpreting them requires experience the platform doesn't teach you.

NetCloud Manager is the cloud management plane and is genuinely the strongest part of the platform — multi-tenant, policy-template-driven, with zero-touch provisioning that makes large fleet deployments tractable. The Bonded WAN Interface workflow lives under Network → Policy → Traffic Steering → Bonded WANs in the NetCloud Manager UI. Each bonded WAN is configured as either Dynamic mode (the system auto-assigns weights and balances traffic against the selected bonding algorithm) or Static mode (administrators enter percentage weights manually for each member interface, with the percentages required to total 100). Member interfaces are added by selecting WAN type — Cellular (with carrier name), Ethernet (adds all ethernet WAN interfaces), WiFi as WAN, or Device WAN Profile (matched against profile names configured on the device). Once a bonded interface is created, Traffic Steering Rules direct specific application categories or LAN segments through it, with the bonding algorithm and behavior configurable per rule. The depth and flexibility are genuine, but the experience is denser than InControl2 — there are more layers (NetCloud Manager → Secure Connect networks → Service Policies → Traffic Steering → Bonded WANs) before you reach the actual bonding configuration. For a network engineer in an enterprise or government IT shop, this is exactly what they want. For a smaller team or a non-specialist, it's a meaningful learning curve.

Hardware. Cradlepoint routers fall into two categories. The E-Series (branch and small office) includes the E300 (single internal LTE/5G modem, branch deployments) and the E400 (single 5G SA modem with dual eSIM, Wi-Fi 7, integrated 8-hour battery backup, "small to medium office"). The R-Series (ruggedized, vehicle, field) includes the R980 (FIPS 140-3 certified, ruggedized), the R1900 (the flagship 5G ruggedized router, IP64 rated, E-Mark and SAE J1455 vehicle certified), the R2100 (vehicle/IoT), and the R2400 (5G SA for vehicles). For dual-cellular bonding, you either pair two routers or add a 5G Wideband Adapter as a captive secondary modem connected over ethernet.

Relay server. The Bonded WAN Interface terminates at a NetCloud Exchange Service Gateway, which can be deployed three ways: fully cloud-delivered as part of NetCloud SASE (no infrastructure on the customer side), customer-hosted as a virtual appliance in an on-premises datacenter or private cloud (typical for enterprises with strict data residency or regulatory requirements), or as the SG4000 pre-loaded hardware appliance for organizations preferring local deployment without standing up their own VM infrastructure.

Pricing model. Hardware is sold exclusively as a bundle with NetCloud subscription — there's no standalone hardware purchase path. The platform offers tiered NetCloud licensing (Essentials, Advanced, NetCloud Exchange) with bandwidth aggregation living at the NetCloud Exchange tier. Cradlepoint doesn't publish a price list directly, but resellers do — the R1900 typically runs $1,500–3,500+ depending on subscription term (1-year vs 3-year bundle), reseller margin, and which NetCloud tier is included. 3-year bundles are common and lower the per-year cost, but raise the upfront commitment substantially. The Service Gateway and NetCloud Exchange add-on are separate line items.

Year-one cost estimate (single-device platform cost). R1900 (~$1,500–3,500 depending on subscription term) + NetCloud Exchange SD-WAN advanced tier (varies by partner; typically several hundred per device per year on top of the base bundle) = roughly $1,800–4,000+ in platform fees, meaningfully higher than SpeedFusion at a similar capability level. Multi-year bundles lower the per-year cost in exchange for higher upfront commitment. Connectivity (SIM plans) is separate.

The gotchas. The biggest one is the licensing layering: buyers who purchase a Cradlepoint router expecting to bond two cellular connections frequently discover that the bonding capability lives behind an additional NetCloud Exchange subscription tier on top of the base NetCloud bundle their hardware came with. "I bought a Cradlepoint and it won't bond" is a real procurement surprise. The second is the architectural caveat that Cradlepoint themselves publish: the aggregated path adopts the latency of the slowest constituent link, and Cradlepoint's own guidance is that "bandwidth aggregation works best when using connections that have the same latency profile (such as two cellular modems)." That's a sharper constraint than SpeedFusion (which Peplink recommends within 150 ms) and structurally limits Cradlepoint as a satellite-plus-cellular bonding solution.

What people complain about most. TCO is the recurring theme in MSP-focused comparisons. The mandatory subscription model and tiered licensing structure mean that Cradlepoint's effective price point at scale is meaningfully higher than Peplink's. NetCloud's complexity is the second-most-frequent complaint — it's a more capable platform than InControl2, but the additional capability comes with a steeper operational learning curve, and shops without a dedicated network engineer struggle to extract full value. The third is the rebrand and ownership transition under Ericsson — the "Ericsson Cradlepoint" terminology is still settling, documentation has been migrating, and some customers have observed shifts in pricing and packaging since the acquisition closed.

Bondix's UI experience varies by host hardware, and it's worth understanding the differences. On Teltonika RutOS — currently the best-supported platform — you install Bondix as an external client, which adds a dedicated Bondix entry to the menu. Native RutOS integration ships mid-2026 and should collapse the configuration into a single one-click activation. The Bondix panel itself is clean: tunnel status, per-link bandwidth, latency, packet loss, and packet flow are all rendered in real time, with bonding mode chosen from a dropdown (speed, seamless backup, load balancing, packet duplication). The catch on Teltonika is that you're managing two interfaces — RutOS handles WAN configuration, firewall, and routing, while Bondix handles the tunnel — and you must disable Teltonika's built-in load balancing or the two will fight over packet routing.

GL.iNet hardware running Bondix's new universal OpenWrt client (beta mid-2026) is the cleanest setup path, but the v1 client doesn't include the local monitor — you'll need the cloud-side dashboard for telemetry. Digi routers ship Bondix as "Digi WAN Bonding" inside Digi Remote Manager, but Bondix themselves describe the Digi integration as the weakest of the three; we'd avoid it unless you're already standardized on Digi.

Beyond mode selection, the QoS preset system is where Bondix earns its rating. You define traffic classes by source IP, destination IP, port, or DSCP value, assign a numeric importance score (Bondix's examples: 1 for default browsing, 3,500 for latency-critical), choose whether to duplicate packets across links for that class, pin the class to specific channels, and set per-class bandwidth caps. App auto-detection works — a Microsoft Teams session in our demo got identified, prioritized, and duplicated across both WANs automatically. Presets are configured server-side and pushed to the router. It's not as granular as SpeedFusion's outbound policy chain (no per-domain or per-MAC matching, no expert-mode chained rules), but it's a real selective-bonding system.

There's a structural distinction worth understanding here: the QoS preset system governs how traffic is handled inside the bonded tunnel. The decision about what enters the tunnel in the first place — versus what bypasses it on a single WAN — is handled by mwan3, the standard OpenWrt multi-WAN routing package. mwan3 does basic policy-based routing fine, but it's a thinner toolset than Peplink's outbound policy chain, which makes both decisions in a single unified system. This is the real reason Bondix's selective bonding is weaker than SpeedFusion's — not the in-tunnel toolset, but the routing layer around it.

Hardware. Multi-vendor and multi-architecture. Teltonika is the best-supported platform today (RUTX, RUTM, RUT9M, RUTC, RUTE series; native RUT OS integration mid-2026). GL.iNet via Bondix's new universal OpenWrt client (beta mid-2026) is the standout for price-to-performance — the Flint 2 (~$150) does ~800 Mbps; the Spark does ~500 Mbps. Digi (natively integrated via Digi Remote Manager), Advantech, Mediaport Systems, and generic OpenWrt devices are all supported. Bondix's client is 6 MB and runs on any Linux device. A new Teltonika business router with built-in 5G and a quad-core ARM CPU (1 Gbps bonded) is expected in the US in late 2026.

Relay server. Bondix requires a relay server (their S.A.NE server) to terminate the bonding tunnel. Three options:

Bondix Essential Cloud (hosted by Bondix): Turnkey hosted service running on Linode infrastructure with 13 PoPs in the US plus additional global coverage. Quick Connect Code activation: enter the string on your router and the tunnel comes up. Pricing starts at €21/month with 1 TB included and scales by throughput tier up to ~€100/month for 500 Mbps. Each subscription includes a dedicated IP. No port forwarding at this tier.

Partner-hosted (e.g., Unwired Networks CloudLink): Resellers like Unwired Networks (which runs Bondix on the 1,300+ European trains using the platform) and SimpliWiFi operate their own Bondix servers and sell bonding-as-a-service with regionally tuned pricing.

Self-hosted (Enterprise): Run the Bondix server as a VM on any Linux machine or cloud instance. MRSP is €310/year per 1 Gbps tunnel license. License counting is per-tunnel and daily, based on the maximum number of simultaneously active tunnels over a 24-hour window — overbooking is allowed (a 150-vehicle fleet running 100 active tunnels at peak only needs 100 licenses). Backup-server failover is included in the base license. Authentication options include certificate-based (used by European police forces who can revoke certs to kill access remotely) and API key + MAC address with an orchestrator (used by Unwired's train fleet for auto-provisioning).

Pricing model. Bondix is more transparent than SpeedFusion. The router is your upfront cost (and since Bondix runs on third-party hardware, you may already own it). The relay server is either hosted (€21–100/month) or self-managed (€310/yr per Gbps tunnel + your own VM costs). There's no equivalent to PrimeCare — the license is the license, and the router's own firmware updates are independent of Bondix.

Year-one cost estimates (platform only).

• Teltonika RUTX12 + Essential Cloud: Router (~$300–500) + Essential Cloud (~$280–1,300/yr) = roughly $580–1,800.
• GL.iNet Flint 2 + Essential Cloud: Router (~$150) + Essential Cloud (~$280–1,300/yr) = roughly $430–1,450.
• Self-hosted on a VPS: Router + VPS hosting ($60–240/yr) + €310/yr per Gbps tunnel license. Reduces ongoing software cost at the price of operational complexity.

Programmatic access. Local device API exposes per-link telemetry (latency, packet loss, throughput) in real time. Prometheus and SNMP integrations are built in for fleet monitoring. This is meaningful if you're building tooling or dashboards on top of Bondix — substantially more open than Peplink's InControl-2-mediated telemetry.

The gotchas. You must disable Teltonika's built-in load balancing before enabling Bondix — if both are active, they fight over traffic routing and performance degrades badly. This is documented in the Bondix wiki but not surfaced anywhere in the UI as a warning. The Bondix client's default behavior prioritizes low-latency WAN links automatically, which sounds smart but means an ethernet connection will always be preferred over cellular even if the ethernet link is congested and the cellular is wide open — you need to manually set priorities (or use the QoS preset system) to override this. On Teltonika hardware, RAM is the throughput bottleneck — to buffer 1 second of bonded traffic at 200 Mbps the client needs ~40 MB of available RAM, which is most of what a RUTX12 has; running other services on the same router (VPN, MQTT, container) will push throughput down further. The GL.iNet Flint 2 path sidesteps this entirely with more CPU and RAM headroom per dollar. DNS forwarding on the LAN side has to be set to a public resolver like Google — carrier DNS may not resolve correctly through the tunnel. Essential Cloud doesn't include port forwarding; if you need inbound connections, you must step up to Enterprise or self-host. The universal OpenWrt client v1 ships without a local monitor — telemetry is cloud-side only until that's added in a future release.

A note on the Digi implementation. Per Bondix's own engineering team, Digi WAN Bonding is the weakest of their integrations — Digi OS isn't optimized for the Bondix client. Use Teltonika or GL.iNet hardware unless you're already locked into Digi for other reasons.

What people complain about most. Bondix has a much smaller community than SpeedFusion, so complaints are scattered across Teltonika forums, the Bondix wiki, and webinar Q&As rather than concentrated in one place. The recurring themes:

• "My bonded speed is way below what I expected." On Teltonika hardware, RAM is the bottleneck — RUTX12 tops out at 150–200 Mbps because the Bondix client needs ~40 MB of RAM to buffer at that speed, which is most of what the router has available. Running other services on the same router pushes throughput down further. This limitation is acknowledged by Bondix in their Teltonika webinar. The GL.iNet path scales further on a similar budget.
• "Teltonika's load balancing broke my Bondix tunnel." The single most common setup mistake. Teltonika's built-in load balancing and Bondix's packet distribution fight if both are enabled. Documented in the wiki but not surfaced as a UI warning, so people discover it the hard way.
• "I can't figure out where to configure X." Settings live in two places — the host router's interface (RutOS, LuCI) and the Bondix S.A.NE panel — and the boundary between them isn't intuitive. Native Teltonika integration shipping mid-2026 should improve this.
• "Essential Cloud doesn't support port forwarding." Users who need inbound connections (hosting services, remote access, cameras) discover the affordable tier doesn't include this feature. You need Enterprise or self-hosted, which significantly changes the cost equation.

Speedify's desktop app is the best-designed interface of any platform in this guide — by a wide margin. The main screen shows your active connections as colored bars (green for healthy, yellow for degraded, red for down), your current bonding mode, and real-time throughput. Switching bonding modes (Speed, Redundant, Enhanced Streaming) is one click. Connection priority (Primary, Secondary, Backup) is a dropdown per connection. There's a built-in speed test that measures each individual connection and the bonded result side by side. Settings are organized in categories that make sense: privacy, transport mode, connection behavior, data limits. There's a dark mode. It feels like a consumer app because it's one — and that's both its strength and its limitation. There are no routing policies, no per-application rules, no traffic shaping, no tunnel parameters to tune. You get modes and priorities, and that's it. For its target user — someone who wants bonding on their laptop without thinking — that's exactly right. For anyone who needs control over what gets bonded and how, it's a dead end.

Hardware. Two distinct paths. Per-device: any device running Windows, macOS, Linux, iOS, or Android. Speedify is also available as an embedded SDK ("Powered by Speedify") for hardware vendors — the Miri Technologies X510 is one implementation. Router-wide: Speedify for Routers installs on OpenWrt 19.07–23.05 and GL.iNet firmware on x86_64 or aarch64 hardware (Raspberry Pi 4/5, Banana Pi R3, Intel/AMD mini-PCs). Installation runs through a single SSH script (wget -q https://get.speedify.com -O- | sh) which deploys the Speedify package and the LuCI web interface plugin. Stock consumer routers (TP-Link, Netgear, ASUS) aren't supported — you need OpenWrt-compatible hardware. The earlier community-built SmoothWAN project is officially sunset, but Connectify's first-party Speedify for Routers product is its supported successor.

Relay server. Speedify operates its own global network of relay servers (~500+ in 34 countries). The relay is included in your subscription — there's no separate server cost, no self-hosting option, and no ability to choose your own relay infrastructure. You pick a server location from a list within the app, or let Speedify auto-select the nearest one. For most users, this is the right model — zero server management. The tradeoff is that you're entirely dependent on Speedify's infrastructure: if their servers in your region are overloaded or experiencing issues, your bonded performance degrades and you have no alternative. Speedify publishes guidance on which connections bond well together — notably, they recommend against bonding two connections from the same provider, and warn that traditional satellite internet's high latency will drag down overall performance.

Pricing model. Two parallel pricing structures depending on which deployment path you pick. Per-device plans: $14.99/month on a monthly plan, $2.99/month on a yearly plan, and as low as $1.50/month on a 3-year commitment. Family and team plans cover multiple seats. A free tier exists with 2 GB of monthly data — enough to test the concept, not enough to use for anything real. Router-wide plans (Speedify for Routers): priced by monthly data allocation rather than seat count — 100 GB at $10.75/month, 500 GB at $15/month, 1 TB at $25/month, and 3 TB at $37.50/month, all on annual billing. Each router plan includes one complimentary individual user account that can be used on up to five additional devices, so a household with a Speedify router can also have Speedify on five laptops/phones at no extra cost. Dedicated and self-hosted Speedify Servers are available as paid add-ons for static IPs, port forwarding, and customer-controlled relay infrastructure. The platform's biggest pricing watch-out is that router plans count every device on the LAN against the data tier — a 100 GB plan disappears quickly on a household with multiple streaming devices and is genuinely only useful for very light use.

Year-one cost estimate. $18-$180/year per device on per-device plans, or $129-$450/year on router-wide plans (depending on monthly data tier). No hardware cost on the per-device path; on the router-wide path you'll need OpenWrt-compatible hardware (~$60-200) plus the subscription.

The gotchas. Speedify's "Secondary" priority means the connection won't be used for bonding unless the Primary connection is fully saturated or unavailable. If your Primary is a 100 Mbps WiFi link, a Secondary cellular connection will sit idle even if the WiFi is experiencing 10% packet loss — you need to manually set both to Primary to get real bonding. The free tier's 2 GB monthly cap is functionally useless for anything beyond a quick test. Speedify routes all your traffic through their relay servers, and their privacy policy is vaguer than you'd want — they log connection timestamps, IP addresses, and data volumes. If you're using Speedify for work VPN traffic, that's worth thinking about. Speedify's own documentation also publishes two important warnings worth knowing before you commit: connections with 10× latency or speed mismatch "might not combine well" and Speedify recommends setting the slower link to backup mode in that case, and traditional GEO satellite is explicitly not recommended for bonding with terrestrial links. And the inconsistency in performance reviews isn't just reviewers being bad at testing — Speedify's bonding performance is genuinely variable depending on server load, distance to the relay, and how different your connections' characteristics are.

What people complain about most. Speedify reviews are polarized in a way that no other platform in this guide matches. The complaints:

• "Bonding mode made my internet slower, not faster." VPNOverview's review described bonding mode as having "absolutely nuked" their dual-WiFi test setup. Other users report similar results — two connections that are each 50 Mbps bonding to 30 Mbps. The issue appears to be highly dependent on the latency match between connections, server distance, and current server load. When it works, it works well. When it doesn't, it actively hurts.
• "It's a VPN that happens to bond, not a bonding tool." Multiple security reviewers have flagged that Speedify uses a proprietary protocol (not WireGuard or OpenVPN), AES-128 encryption (not AES-256), and logs more data than users expect from a "privacy" product. Cloudwards reported it failed their encryption testing. For users who chose Speedify because they needed bonding AND privacy, this is a rude surprise.
• "SmoothWAN stopped working and now I'm stuck." Users who deployed SmoothWAN — the community-built OpenWrt wrapper that made Speedify work at the router level before Connectify shipped a first-party version — are discovering that the project is sunset and may break with future Speedify updates. The migration path is Connectify's official Speedify for Routers, which supports a similar OpenWrt/GL.iNet hardware list and is now the supported way to run Speedify network-wide. SmoothWAN itself isn't getting fixed; users should plan to move to Speedify for Routers.
• "The free tier is useless." 2 GB per month is exhausted in under an hour of normal use. It exists as a marketing funnel, not a functional product.

OpenMPTCProuter uses the standard OpenWrt LuCI web interface with a custom "OpenMPTCProuter" settings page bolted on. If you've used OpenWrt before, you'll feel at home. If you haven't, you'll be looking at a dense, text-heavy interface with minimal visual hierarchy and a lot of options that require background knowledge to understand. The OpenMPTCProuter-specific settings page is where you enter your VPS server IP and keys, assign WAN interfaces as "Master" or "Enabled" for MPTCP, toggle SQM (smart queue management), and choose between Shadowsocks and V2Ray as the TCP proxy. There's no visual tunnel dashboard, no real-time bandwidth graph per link, and no bonding mode selector — you configure the behavior through a combination of MPTCP settings, proxy selection, VPN choice, and TCP congestion control algorithm (BBR2 vs Cubic vs BBR, each with different performance characteristics on different link types). Monitoring requires either SSH-ing into the router or checking your VPS metrics separately. It's functional. It's not friendly. But for the audience that chooses OpenMPTCProuter, "friendly" isn't the priority — capability and transparency are.

Hardware. Raspberry Pi 4B/5 (most common), x86/x86_64, Linksys WRT3200ACM/WRT32X, Teltonika RUTX12, Banana Pi BPI-R2, GL.iNet MT2500/MT3000/MT6000, NanoPi R4S. Requires separate WAN devices (USB LTE modems, hotspots, or ethernet-connected routers). The Raspberry Pi path is cheapest; the x86 path (Intel NUC or mini PC) gives you more CPU headroom for higher throughput. Note that the Raspberry Pi 3B is a hard cap at roughly 90 Mbps because its onboard Ethernet is 100 Mbps — even if you have CPU headroom and fast WANs, you can't get more bonded throughput than the Ethernet port can carry. Pi 4B and 5 have gigabit Ethernet and lift this ceiling.

Relay server. OpenMPTCProuter is self-host only. There's no hosted cloud service — you must provision your own VPS and run the OpenMPTCProuter server script on it. This is both the platform's greatest strength and its highest barrier to entry.

The server script installs on any KVM-based VPS running Debian or Ubuntu. (OpenVZ VPS instances won't work — this is a common early mistake.) Popular providers include Vultr, DigitalOcean, Hetzner, and Linode, typically costing $5-20/month depending on location and bandwidth allocation. Provision the VPS uplink at roughly 120% of your total bonded throughput to handle protocol and crypto overhead — bonding two 50 Mbps connections needs at least a 120 Mbps VPS uplink, not 100 Mbps. The VPS location determines your baseline tunnel latency — pick a server in your region. The VPS public IP becomes your persistent IP, which means you get a real, routable public IP with full port forwarding capability — a significant advantage for users behind CGNAT on cellular, and something that hosted relay services like Bondix Essential Cloud and SpeedFusion Cloud don't offer at their basic tiers.

The server script handles all configuration automatically (MPTCP setup, proxy installation, VPN configuration, key generation), but you need to be comfortable with SSH, basic Linux commands, and reading terminal output. When the server needs updating — which happens with each OpenMPTCProuter release — you run the script again. A practical setup guide by the community covers the end-to-end process for Raspberry Pi deployments.

Pricing model. Free. There's no license, no subscription, and no vendor. The only costs are hardware (your router), the VPS ($5-20/month), and your WAN connections. This makes OpenMPTCProuter the cheapest bonding platform by a wide margin, but the hidden cost is your time. Setup takes a full day for someone new to OpenWrt and Linux. Troubleshooting, when things go wrong, requires reading GitHub issues and hoping someone has encountered your exact problem before. There's no support line to call.

Year-one cost estimate (platform only). OpenMPTCProuter the software is free. Your costs are: router hardware (Raspberry Pi 4/5 ~$60–80, or x86 mini-PC for higher throughput) + VPS to run the relay ($60–240/yr) = roughly $140–320 in year one. USB LTE modems or separate hotspots are extra if you don't already have them, but those are connectivity hardware rather than platform cost. The cheapest bonding platform by a wide margin — the hidden cost is your time.

The gotchas. The "master" WAN interface designation is critical and fragile: if the interface you set as master is down when the router boots, MPTCP won't establish and you lose connectivity until that specific interface comes back — even if your other WANs are fine. This catches people who assign their least reliable connection as master by mistake. The proxy choice (Shadowsocks vs V2Ray) matters more than you'd think: Shadowsocks only proxies TCP, so UDP traffic goes through a separate VPN tunnel (adding TCP-over-TCP overhead); V2Ray can handle both TCP and UDP but requires more CPU. The VPS location is your relay server — if you pick a cheap VPS in a different region to save $3/month, you'll add 30-50 ms of baseline latency to every bonded packet. BBR2 congestion control gives the best throughput on clean links but degrades badly on lossy connections; Cubic is slower but more resilient to packet loss. MTU also has a sharp edge: tunneled bonding adds 60–80 bytes of overhead per packet, and most cellular carriers silently filter ICMP "Fragmentation Needed" messages, which means Path MTU Discovery black-holes traffic with the don't-fragment bit set. The fix is to clamp your LAN-side MTU to around 1340 bytes — without that, you'll see specific applications hang on large packets while small packets work fine, which is one of the more confusing failure modes to diagnose. And MPTCP itself can be filtered by ISPs that strip TCP options from headers — OpenMPTCProuter works around this with an additional VPN layer per WAN, but that adds overhead and complexity.

What people complain about most. The GitHub issues tracker is the primary forum, and it's active — over 3,800 issues filed to date. The recurring pain points:

• "My speed is way below what I expected — 3 modems but only 100 Mbps." The most common issue. Causes range from VPS throughput limitations (cheap VPS instances often have throttled bandwidth) to suboptimal congestion control algorithms (BBR2 on lossy links, Cubic on clean links — the wrong choice tanks performance) to the Raspberry Pi's CPU maxing out under encryption load. There's also a structural reason MPTCP underperforms on cellular: TCP's three-DUPACK rule treats out-of-order delivery as packet loss and halves the congestion window, so even 1% of spurious loss from path reordering on an 80 ms LTE link caps per-flow throughput at about 14.6 Mbps (Mathis equation: T ≤ MSS / (RTT × √p)). A dual-LTE MPTCP study at highway speeds saw MPTCP throughput drop 37% while single-path TCP only dropped 10% — reordering penalties compound on multipath. The current state-of-the-art schedulers (BLEST and ECF, Ferlin et al., 2016) achieve roughly 62% of aggregate bandwidth on heterogeneous paths, which is the realistic ceiling rather than the sum-of-WANs that users expect. Solutions usually combine VPS upgrade, congestion control tuning, and adjusted expectations about what MPTCP can deliver over cellular.
• "Netflix/streaming services block me because they see a VPN." Because all traffic routes through your VPS, streaming services detect the VPS IP as a proxy/VPN and block access. The OMR-Bypass feature is supposed to let you route specific traffic (like Netflix) outside the tunnel, but it's fragile — domain-based bypass works inconsistently, and MAC-address-based bypass requires per-device configuration. This is one of the most frequently filed issues on GitHub.
• "Setting up the VPS was a nightmare." Users unfamiliar with Linux regularly report multi-hour struggles with SSH, VPS provisioning, broken pipes during installation, and confusion about KVM vs OpenVZ requirements. The server script itself works well — the problem is everything leading up to running it.
• "My internet goes down when one WAN reboots." The master interface cold-start problem. If the interface designated as "master" for MPTCP is down when the router powers on, connectivity doesn't establish even though other WANs are up. This is a fundamental MPTCP behavior that catches people who didn't read the documentation carefully.
• "UDP performance is terrible for gaming/VoIP." The TCP-over-TCP overhead for non-TCP traffic is a known architectural limitation, not a bug. But users who chose OpenMPTCProuter for gaming or VoIP without understanding this distinction are consistently disappointed.

Dejero ships with two UIs: Control, the cloud platform for fleet management (device list, usage tracking, concentrator placement, license management), and a per-device web UI on the gateway itself. The web UI is one of the better ones in the category. The dashboard renders real-time blended throughput as a green line representing the estimated total pipe, with each individual carrier's contribution layered underneath as overlapping bands so you can see at a glance which links are doing the work. Per-cell latency, RSSI, and dropped packets are all surfaced inline. A built-in speed test runs against the actual blended path and shows your real end-user latency rather than just the latencies of individual carriers — a small thing, but a useful one when you're trying to explain to a customer why their connection feels the way it does.

Flow Rules is the feature that most directly distinguishes Dejero's approach from SpeedFusion's outbound policy chain. Each rule defines a traffic class by port and protocol, then sets behavioral parameters: target latency, importance level, packet ordering behavior (e.g., "first come first served" vs "ordered delivery"), and which SIMs are eligible to carry that class of traffic. UDP traffic, for example, can be tagged as real-time/interactive with a 150 ms target latency and the system will prioritize delivery accordingly. The traffic flow inspector shows which rules matched which flows, which is genuinely useful for debugging.

The UI supports skinning — customer logos can be applied so end users see a branded view — and read-only profiles for customers who want visibility into their device without admin access. Both are useful for managed-service deployments where you want customers to see something other than a vendor admin panel.

Hardware. Dejero's gateway lineup, current and forthcoming:

• Titan Command (current flagship): 3× 5G radios, built-in antennas, 4× 2.5 GbE RJ45, sensor and vehicle integration ports, IP67 rated with optional external IP enclosures, extreme operating temperature range. ~$3,730 one-time including warranty.
• Titan Mobile (new): 3× 5G, IP54, 6× RJ45 ports, requires externally mounted antennas. ~$3,995.
• Titan Scout (Q3 2026): 4× 5G fully aggregated portable hotspot with 2-hour built-in battery and full UI access on-device. Pricing not yet published.
• Titan Ghost (Q4 2026 / Q1 2027): software-only license to run the Dejero VM on your own hardware. Removes hardware lock-in for customers who want it.

Dejero previously partnered with Celerway to use Celerway's dual-5G hotspot form factor — running CelerwayOS underneath, but with Celerway's bonding software ripped out and Dejero's blending layered in. The partnership has since shifted toward Dejero's own hardware lineup, but it's a useful data point on Dejero's view of Celerway's bonding software.

Relay server. Every Dejero deployment requires a concentrator — their term for the cloud-side aggregator that reassembles blended packets and provides the persistent IP. The concentrator is a fully managed service, with a dedicated concentrator provisioned per customer in any Azure region globally. Customers can place the concentrator close to their gateway (typical) or close to the destination (advantageous for path protection in some configurations). Self-hosting is available with a $450/yr discount on the Core license. There's no equivalent to FusionHub for fully customer-controlled deployments without ongoing license fees — you're either paying Dejero for managed hosting or paying Dejero a (smaller) ongoing license fee to self-host.

Pricing. License-based with three speed tiers, all annual:

• CoreMax — 850×850 Mbps performance, $5,000/yr MSRP
• Core — 50×50 Mbps base with token-system bursting up to 850 Mbps, $3,300/yr MSRP (~$2,900/yr if self-hosted)
• Core Light — 25×25 Mbps base with bursts to 100 Mbps, $1,100/yr MSRP, aimed at fleet, mobile data terminals, and sensor deployments

Multi-year commitments earn 5% off (3-year) or 10% off (5-year). The token-based bursting on the Core tier is unusual: instead of paying for peak bandwidth, you provision baseline capacity and consume tokens during bursts up to 850 Mbps. Worth understanding before committing if your workloads are bursty.

Year-one cost estimate. Hardware ($3,730–3,995) plus a managed Core license ($3,300) puts most realistic deployments at $6,000–7,000 in year one. A CoreMax-tier configuration with hosted concentrator runs closer to $8,500–9,000 in year one. Multi-year discounts and self-hosting can pull this down somewhat. Testing periods of 2–30 days are available as proof of concept or paid rental.

The gotchas. Port-based traffic classification means you can't identify traffic by domain or by application name in the way SpeedFusion's outbound policy chain can — if you want to pin "all YouTube traffic to a specific SIM," you need to enumerate every port YouTube uses, the same problem we hit with SpeedFusion when DPI isn't available for bonded traffic. The token-based bursting on the Core tier is fine if your workloads are predictably bursty but can produce unpredictable monthly costs if you exceed token budgets frequently. Health monitoring on the gateway requires an IP-level check rather than just interface state — this is the right behavior, but you have to configure it explicitly rather than rely on defaults. The Titan Ghost software-only option is the most interesting upcoming development but isn't shipping until late 2026 at the earliest, so anyone evaluating Dejero today is committing to their hardware.

What people complain about most. Dejero's customer base is concentrated in broadcast news, public safety, and emergency services rather than the home/SMB market, so the public complaint footprint is genuinely smaller than SpeedFusion's. The recurring themes from what does exist:

• "It's expensive." The $6,000+ year-one cost is well above SpeedFusion (which starts under $1,000 in platform costs on prosumer hardware) and Bondix's $430–1,800. For users coming from those tiers, the price step is significant. Dejero's positioning is enterprise-grade, FedRAMP-certified, and broadcast-quality — the price reflects that, but it's not for everyone.
• "Hardware lock-in." Until Titan Ghost ships, you're buying Dejero's appliances. The hardware is well-built (IP67, extreme temperatures, sensor ports) but the lock-in is real, and the per-device cost adds meaningfully to the deployment.
• "No DPI for traffic classification." Application-aware routing without DPI requires manual port enumeration. This is a deliberate architectural choice, but it's a friction point for users coming from platforms with domain-aware policies.

Viprinet's management interface is accessed through a setup tool for initial configuration and a web interface for ongoing management. The interface reflects the platform's enterprise orientation — dense, detailed, and designed for network engineers who already understand VPN tunnel architecture. QoS and traffic shaping settings are granular, with per-tunnel and per-WAN-module controls. SNMP and Syslog integration are built in for monitoring. It's not a UI you'd describe as "modern" or "intuitive" — it's a UI you'd describe as "complete." For its target audience (IT teams managing multi-site deployments), that's the right tradeoff.

Hardware. Viprinet only. Proprietary routers and hubs. Router 310 (3 WAN module slots, 100 Mbps bonded), Router 2620 (6 slots, 400 Mbps bonded), and a mobile-specific variant for vehicles. WAN modules are hot-swappable — you can add or remove LTE, DSL, or ethernet modules without powering down the router.

Relay server. Viprinet uses proprietary hub hardware — there's no cloud relay option and no self-hosting on generic VMs. You need a physical Multichannel VPN Hub: the Hub 2030 (500 Mbps bonding capacity, for small/medium networks) or the Hub 5010/5020 (up to 2 Gbps / 5 Gbps, for large ISP-scale deployments supporting 250+ sites). The hub must be colocated in a data center or hosted by a Viprinet partner. Some resellers offer hosted hub services, but the pricing is opaque and typically requires a sales conversation.

Pricing model. Everything is capital expenditure plus mandatory subscriptions. Router hardware ($2,000-6,000+), Hub hardware ($2,000-10,000+), and Viprinet Lifetime Maintenance (VLM) subscription (required for firmware updates, support, and RuggedVPN access — pricing through sales channel). The VLM subscription is non-optional in any practical sense: without it, you can't run RuggedVPN firmware (which delivers the best performance), you get no updates, and you get no support.

Year-one cost estimate (single site, platform only). Router ($3,000–6,000) + Hub or hosted hub service ($2,000+ for hardware or ongoing hosting fees) + VLM subscription = $5,000–10,000+ depending on configuration. Connectivity is separate. Not a meaningful comparison to the other platforms in this guide.

The gotchas. The VLM subscription isn't optional in practice — without it, you get no firmware updates and no support, and Viprinet's RuggedVPN firmware (which delivers the best performance) strictly requires an active VLM. The Hub is a hard requirement, not a nice-to-have — you can't use Viprinet with a generic cloud VPS. You need to either colocate a physical Hub in a data center or use a Viprinet-hosted hub service, both of which add significant ongoing cost. And Viprinet's product names — Multichannel VPN Router 310, 2620, Hub 5010, Hub 5020 — give you no indication of what any of them do, which makes the initial research process more confusing than it needs to be.

Mushroom Networks' web GUI is functional and straightforward — you can monitor per-link performance, total bonded throughput, and data usage in real time. The application-aware routing works automatically, which means the interface is less complex than SpeedFusion's because you don't need to build policy rules — the device classifies traffic and optimizes routing on its own. That's a genuine advantage for organizations that want bonding without a network engineer on staff. The downside is that you have limited ability to override the automatic behavior when it makes decisions you disagree with.

Hardware. Mushroom Networks only. Proprietary appliances. The Portabella line (cellular bonding, $1,300-7,500) ranges from the 2000i (2 SIM slots) to the 8000i (8 SIM slots). The Truffle line (wired broadband bonding, $4,000-19,000) handles fiber, cable, DSL, and other fixed connections — the flagship Truffle EX bonds up to 16 simultaneous WAN transports and handles up to 10 Gbps aggregate routing throughput, which is the high end of any platform reviewed here. Portabella-Home (~$1,300) is the consumer-adjacent product that bonds a phone's LTE hotspot with a wired ISP.

Relay server. Mushroom Networks Cloud Relay is required — there's no self-hosting option. The Cloud Relay terminates bonding tunnels and provides the application-aware optimization that's the platform's key differentiator. Alternatively, if your headquarters runs a Truffle appliance, remote sites can terminate tunnels there instead of the cloud (for site-to-site bonded VPN). Cloud Relay pricing is subscription-based but not publicly listed — expect $50-200/month based on throughput and features.

Pricing model. Hardware upfront plus mandatory cloud subscription. There's no free tier, no trial, and no self-hosted alternative to the relay. The pricing reflects Mushroom Networks' enterprise positioning — this isn't designed to be cost-competitive with SpeedFusion or Bondix. It's designed to be the best-performing option for organizations where cost is secondary to reliability.

Year-one cost estimate (Portabella cellular bonding, platform only). Portabella device ($1,300–7,500) + Cloud Relay subscription (estimated $50–200/month, ~$600–2,400/yr) = roughly $1,900–10,000+. Connectivity (SIM plans) is separate.

The gotchas. Cloud Relay is a hard dependency, not an option. If Mushroom Networks' cloud service has an issue, your bonding tunnel goes down — there's no self-hosted fallback. The Portabella-Home concept (bond your phone's hotspot with wired ISP) sounds elegant but requires you to keep your phone's hotspot enabled and within WiFi range of the device at all times, which drains your phone battery and ties up your phone's cellular connection. And the upgrade path from Portabella-Home to multi-SIM Portabella isn't incremental — it's a completely different (and much more expensive) product line.

VeloCloud Orchestrator is the cloud-hosted management plane — multi-tenant, policy-template-driven, with role-based access for distributed IT teams. The pitch is that Smart Defaults eliminate most of the manual policy work that earlier-generation SD-WAN required: 2,500+ applications come pre-classified with default steering behavior and priority levels, so a branch deployment "just works" once the Edge is connected and authenticated. The reality is that enterprise SD-WAN policy editors are still complex once you move past defaults — custom apps, hybrid SaaS-and-on-prem flows, branch-to-datacenter routing, and per-tenant business policies all require operator skill and partner engagement to tune. Less complex than SpeedFusion's chained outbound policy expressions because Smart Defaults handle the majority of traffic automatically, but more abstract: you're managing policy templates and business intent rather than directly editing per-packet steering rules. Orchestrator visibility into per-tunnel and per-link performance is genuinely strong — DMPO's continuous measurement feeds detailed loss/latency/jitter telemetry into the dashboard.

Hardware. VeloCloud Edge appliances range from the Edge 50 (~50 Mbps, smallest branch) through mid-range models (200–400 Mbps) up to multi-gigabit appliances for larger branches and datacenters. As of the Arista acquisition, the appliance line is being rebranded but the underlying hardware design and tiers continue.

Relay server. VeloCloud Gateways are hosted at internet PoPs and run by Arista (formerly Broadcom/VMware). They terminate DMPO tunnels and provide cloud-onramp for SaaS traffic. For enterprise customers wanting branch-to-datacenter or branch-to-branch topologies, a Hub Edge can be deployed at the corporate datacenter to act as the tunnel termination point. Self-hosted Gateway deployments aren't the typical path.

Pricing model. Enterprise sales channel only. Quoted via partners with annual subscription commitments — typically per-Edge per-year licensing on top of hardware, with bundled Gateway access. There's no published price list, no trial, and no self-service tier. Partner discounts vary. The all-in cost depends on Edge tier, contract length, and which features are activated (basic SD-WAN versus advanced security and SASE add-ons). The Broadcom era pushed customers toward longer-term enterprise contracts and bundled pricing; the Arista transition is too early to know how that motion will evolve.

Year-one cost estimate. Not meaningfully comparable to the other platforms in this guide. Enterprise distributed-branch deployments typically run thousands of dollars per Edge per year, all-in, with the actual number depending heavily on partner negotiation. If you're a single-site buyer asking "how much will VeloCloud cost me," you're likely not the target customer.

The gotchas. The biggest unknown in 2026 is the ownership transition. Arista closed the VeloCloud carve-out in July 2025, and the deal was structured as an asset-plus-talent transfer where roughly half of VeloCloud's ~1,000 employees moved over — primarily core engineering. Most sales and marketing roles weren't transferred, meaning the partner channel and customer success organizations are mid-rebuild as Arista integrates the platform into its broader product line. Documentation is still hosted on Broadcom's techdocs site as of early 2026, transitional. Expect more reshuffling as Arista consolidates branding, packaging, and channel partnerships through 2026 and into 2027.

What people complain about most. Encrypted-traffic classification can fail on novel or custom applications, requiring manual policy intervention — Smart Defaults are excellent for the long tail of mainstream SaaS but break down on internal apps and unusual traffic patterns. Policy template inheritance can produce unexpected behavior when tenant policies override profile defaults in non-obvious ways. Edge appliance hardware refresh cycles tend to be tied to Broadcom (now Arista) channel pricing rather than buyer preference. And as with most enterprise SD-WAN, the platform assumes you have an IT organization or managed service partner to operate it — the cloud Orchestrator simplifies deployment but doesn't eliminate the need for ongoing tuning.

CelerwayOS has a web interface that's clean by industrial router standards. The Phantom VPN configuration is reasonably intuitive — you define hub instances, assign WAN interfaces, and the system handles load balancing across them dynamically. Where Celerway's UI genuinely shines is Nimbus, their cloud fleet management platform. Nimbus provides real-time monitoring of every deployed router, with quality indicators, analytics, alerts, and remote configuration push. For an organization managing 50+ routers in vehicles or field locations, Nimbus is a genuinely good product. For a single home user, it's irrelevant. The local router interface is functional but clearly designed for the deployment engineer who sets it up once and then manages it remotely through Nimbus.

Hardware. Celerway only. Proprietary ruggedized routers: Arcus (flagship, modular, up to 3 modem slots, 7 WAN connections, MIL-STD-810G certified), Stratus (dual-modem, up to 8 WANs), Cumulus (budget, single modem), and Go (portable, battery-powered). All are built with impact-resistant materials rated for extreme temperatures, shocks, and vibrations. Wide voltage input (9-36V DC) for vehicle and field deployment.

Relay server. Celerway Phantom VPN can terminate at a self-hosted hub instance (on any hardware or VM) or at Celerway-hosted cloud endpoints. Unlike SpeedFusion and Bondix, Phantom is built on WireGuard, which gives it excellent performance and modern security — but WireGuard's design means tunnel endpoints have fixed configurations, so changing your relay server requires reconfiguring the tunnel rather than seamlessly migrating. Nimbus cloud management handles fleet-wide monitoring and configuration, but the bonding tunnel itself runs through Phantom, which is separate from Nimbus.

Pricing model. Hardware through reseller channels only — no published pricing, no direct purchase. Expect the Stratus and Arcus to be in the €1,000-3,000+ range depending on modem configuration. CelerwayOS and Phantom are included with the hardware — there's no separate software license. Nimbus fleet management may have per-device or per-organization pricing depending on scale. The pricing model is opaque by design: Celerway sells to integrators and fleet operators through a partner channel, not to end users browsing a website.

Year-one cost estimate. Not meaningfully comparable. Hardware through reseller channels, pricing not public. The total cost depends on your reseller, your configuration, and your scale.

The gotchas. Celerway's Phantom VPN is built on WireGuard, which is fast and modern — but WireGuard's design means each tunnel endpoint has a fixed IP, so if you need to change your relay server, you're reconfiguring the tunnel rather than seamlessly switching. The open-source CelerwayOS is a selling point on paper, but in practice you can only run it on Celerway hardware — you can't flash it onto a Teltonika or a Raspberry Pi. And buying Celerway hardware requires going through a reseller, which means no Amazon Prime two-day shipping, no published pricing, and a sales conversation before you can even get a quote.

The two most direct competitors in cellular bonding. Both deliver true packet-level bandwidth aggregation with FEC, packet duplication, and policy-based traffic steering. The differences are positioning. Peplink ships everything from $300 home routers to enterprise hubs and sells transparently through resellers; PrimeCare is the only mandatory subscription, and it's a flat per-device annual fee. Cradlepoint is enterprise-first — every router is sold as a NetCloud bundle with mandatory subscription, and bandwidth aggregation lives behind the higher-tier NetCloud Exchange SD-WAN license. Cradlepoint's edge is government and public-safety credentials: Verizon Frontline, AT&T FirstNet, T-Mobile Connecting Heroes, FedRAMP-authorized cloud management. Peplink's edge is total cost and breadth — comparable bonding capability for meaningfully less, with hardware that scales from prosumer to enterprise on one platform. For US federal, public safety, or defense buyers, Cradlepoint. For anyone else evaluating cellular bonding, SpeedFusion is usually the better-value answer.

Three tiers of the same problem. SpeedFusion is our pick — most capable bonding, deepest selective-bonding policy engine, integrated 5G modem and Starlink management, transparent pricing through resellers. Cradlepoint is the more-expensive enterprise alternative with stronger US carrier-program credentials (FirstNet, Frontline, Connecting Heroes) and FedRAMP-authorized cloud management; pick it if those certifications matter to procurement, otherwise it's overkill. Bondix is the cheaper alternative running on third-party hardware (Teltonika, GL.iNet, Digi, generic OpenWrt) — fewer policy knobs than SpeedFusion, but real bonding on a $150 GL.iNet Flint 2 instead of a $500+ Peplink router.

Different products despite the surface similarity. Bondix is true packet-level bonding — your bonded throughput is the sum of your WANs (minus overhead), with selective-bonding policies, FEC-equivalent stream-fragmentation handling, and per-class QoS. Speedify for Routers is a router-level deployment of the Speedify per-device VPN, which means simpler setup, less granular control, AES-128 instead of AES-256, and data caps tied to your monthly subscription tier. Hardware-wise, both run on OpenWrt boxes including the GL.iNet Flint 2. If you want the simplest possible router-wide bonding and your throughput needs fit inside Speedify's data tiers, Speedify. If you want real packet-level bonding with selective routing, custom QoS, and no monthly data cap, Bondix.

SpeedFusion has a deeper routing policy engine — per-domain, per-MAC, chained expert-mode rules — and a more mature FEC plus packet-duplication implementation for lossy cellular links. Bondix takes a different architectural approach: it dissolves traffic into stream fragments inside the tunnel and uses a real-time control channel to retransmit only what's lost, which is why it doesn't use FEC. Bondix is also more hardware-flexible — it runs on Teltonika, GL.iNet, Digi, generic OpenWrt, and any Linux device, while SpeedFusion is locked to Peplink hardware. Setup is simpler too (Quick Connect Code activation vs SpeedFusion's multi-step tunnel configuration). If you need the most granular per-application routing control, SpeedFusion. If you want bonding without Peplink lock-in — especially on inexpensive hardware like the GL.iNet Flint 2 — Bondix.

Different tools for different people. OpenMPTCProuter delivers excellent TCP bonding throughput for free — users regularly report near-perfect bandwidth aggregation. But UDP and real-time traffic (gaming, VoIP, some VPN protocols) are second-class citizens due to TCP-over-TCP overhead. SpeedFusion handles all traffic types equally well and has superior performance on lossy cellular connections, but starts around $700 in year one on entry hardware (and runs into the thousands on mid- and enterprise-tier hardware) and requires Peplink hardware. If your primary use case is high-throughput TCP workloads on stable connections and you enjoy Linux, OpenMPTCProuter. If you need reliable real-time application bonding on cellular, SpeedFusion.

Yes. Speedify started as a per-device VPN application for Windows, macOS, Linux, iOS, and Android, and that path still exists, but Connectify also ships Speedify for Routers — a first-party install for OpenWrt 19.07–23.05 and GL.iNet firmware on supported hardware (Raspberry Pi 4/5, Banana Pi R3, x86_64 and aarch64 boxes). The router-wide version turns the router into a bonding gateway for every device on the LAN, with pricing tiered by monthly data allocation rather than seat count. The earlier community-built SmoothWAN project, which packaged Speedify into an OpenWrt image, is officially sunset; users on SmoothWAN should migrate to Speedify for Routers. Note that Speedify for Routers only supports OpenWrt-compatible hardware — it doesn't install on Peplink, Teltonika, or stock consumer routers.

It depends on the platform. All numbers below are platform only — connectivity (cellular SIM plans, ISP lines) is separate. SpeedFusion runs from ~$700 in year one on entry-tier Peplink hardware (router + PrimeCare + a small SpeedFusion data plan) up to $10,000+ on enterprise hardware that pushes multi-gigabit bonded throughput. Mid-tier (B One 5G class, what we use for Waveform Internet) is around $840 in year one — $500 hardware plus ~$340/yr in recurring platform fees. OpenMPTCProuter is free software — you pay for hardware (~$60-80 for a Raspberry Pi) and a VPS ($60-240/yr), totaling roughly $140-320 in year one. Bondix runs $430-1,800 in year one depending on hardware path (GL.iNet Flint 2 + Essential Cloud at the cheap end, Teltonika RUTX12 + Essential Cloud at the higher end). Speedify is $18-180/year per device on the per-device path, or $129-450/year for router-wide plans depending on data tier. Cradlepoint is $1,800-4,000+. Enterprise platforms run $5,000-10,000+ for Viprinet, $1,900-10,000+ for Mushroom Networks, and $6,000-9,000 for Dejero. Alternatively, Waveform Internet provides managed SpeedFusion bonding starting at $65/month for backup or $125/month for full service, with hardware from $499.

Not quite, and the failure mode is worth understanding. FEC works by sending parity packets so the receiver can recover from random packet loss without retransmission, and it does this well when losses are independent across links. The problem with cellular: losses on two SIMs from the same carrier — or even different carriers connecting through the same tower — tend to correlate. A congestion event, a tower handoff, or an RF burst affects both connections simultaneously, so what looks like 5% loss on each link individually can show up as moments where both links lose packets at the same time. When that happens, even maximum-overhead FEC (26.7% on SpeedFusion) may be insufficient because the parity packets get lost in the same burst as the data packets they're supposed to protect. Packet duplication has the same vulnerability — sending the same packet on two correlated cellular links doesn't help if both lose it.

There's also a breakeven below which FEC is wasted bandwidth: roughly 20 ms RTT and 0.5% loss. Below that threshold, normal TCP retransmission resolves losses quickly enough that the overhead of FEC parity packets exceeds the latency savings. Above it — or for deadline-sensitive UDP traffic where retransmission isn't an option — FEC is strictly better. SpeedFusion's adaptive FEC (6.7–20% based on observed loss) is essentially an attempt to track this breakeven automatically, raising overhead when conditions warrant and dialing it down when they don't.

Mitigations: bond cellular SIMs across two genuinely different carriers (not two SIMs on the same carrier), and pair cellular with at least one wireline backup where available. The platforms that handle this best are the ones that combine FEC with active path measurement and per-flow scheduling, so they can route loss-sensitive traffic onto the link that's currently performing best rather than relying on duplication or parity to paper over correlated failures.

You can, but expect compromise. Bonding works by splitting packets across paths and reassembling them at the relay, and that reassembly requires waiting for the slowest packet — so the bonded tunnel's effective latency tracks the slowest active link, not an average. A 10 ms fiber bonded with a 70 ms Starlink connection produces a tunnel with roughly 70 ms of latency, even when most packets travel the fiber. SpeedFusion handles this with a Packet Jitter Buffer (default 150 ms) that smooths the reordering, and Peplink explicitly recommends keeping bonded WANs within 150 ms of each other. Cradlepoint is sharper still — their own documentation says bandwidth aggregation "works best when using connections that have the same latency profile (such as two cellular modems)." If your two paths differ by more than ~150 ms, you'll usually get better real-world performance using selective bonding (route latency-sensitive traffic over the fast path only) or using the slower path strictly as backup rather than bonding it in.

For DIY home bonding, SpeedFusion (Peplink) is the most capable platform — it has the best selective bonding, the best lossy-link handling, and the most active community. The tradeoff is complexity and cost. OpenMPTCProuter is the best free option for technically comfortable users. For home users who want bonding without the project, Waveform Internet is a managed service built on SpeedFusion that ships pre-configured with high-priority 5G SIMs and full support — no configuration, no maintenance, no relay server to manage.

Yes — this is one of the primary use cases for WAN bonding. All platforms in this guide support cellular connections as WAN inputs. SpeedFusion on Peplink routers with built-in 5G modems (like the B One 5G) is the most common approach. Bondix on Teltonika dual-SIM routers is another option. OpenMPTCProuter works with USB LTE modems or separate cellular hotspots. The key consideration for cellular bonding is how well the platform handles the jitter, packet loss, and congestion inherent to cellular networks — SpeedFusion's FEC and packet duplication are purpose-built for this, while OpenMPTCProuter's MPTCP-based approach struggles more on lossy links.

Yes. Waveform Internet is a fully managed WAN bonding service built on SpeedFusion and Peplink hardware. It ships pre-configured with high-priority 5G SIMs, intelligent selective bonding, proactive monitoring, and full-service support. You plug it in and it works — no tunnel configuration, no relay server management, no routing policy tuning. Plans start at $65/month for backup internet or $125/month for full service. It is, as far as we're aware, the only managed bonding service designed for homes and small businesses rather than enterprises.

Selective bonding means choosing which traffic goes through the bonding tunnel and which bypasses it. This matters because bonding adds overhead — encryption, encapsulation, and a throughput cap (roughly 150-200 Mbps on most prosumer hardware). You want your VPN, Zoom calls, and gaming sessions bonded for session protection. You don't want your Netflix stream or file downloads eating tunnel bandwidth when they could ride your fastest WAN directly at full speed. SpeedFusion is the only platform with a truly granular selective bonding implementation. Bondix, OpenMPTCProuter, and Speedify offer more limited versions. Waveform Internet's Smart tier uses SpeedFusion's selective bonding to protect critical traffic while letting everything else bypass the tunnel at full WAN speed.