MikroTik’s New 400GbE Switch Is Wildly Fast — and Almost Nobody Can Actually Use It

When MikroTik announces a new switch, it usually comes with two guarantees: the spec sheet will be unreasonably ambitious, and the price will make competitors uncomfortable. The newly announced MikroTik CRS804 DDQ follows that tradition perfectly. Four ports of 400GbE in a compact form factor, positioned as a next-gen core or aggregation switch, and priced low enough to raise eyebrows across the networking world.

On paper, it’s absurdly fast. In practice, it feels… early. Not bad. Not misguided. Just very ahead of where most people — even serious enthusiasts and small businesses — actually are.

And that gap between what the hardware can do and what people can realistically deploy is the entire story here.

Four Ports, 1.6 Terabits, No Waiting

Let’s start with the impressive part, because there’s plenty of it. The CRS804 DDQ is a 1.6 Tbps switch with four QSFP-DD ports, each capable of pushing 400GbE. That’s data center-class bandwidth in a box small enough to feel almost mischievous. You don’t usually see this kind of throughput without a chassis, redundant everything, and a quote process that involves several meetings.

MikroTik, as usual, skipped the ceremony and went straight to shipping something bold.

Internally, it’s built around a modern Marvell switching ASIC — the same family used in MikroTik’s larger CRS812 — which means this isn’t some experimental science project. It supports breakout modes, letting each 400G port fan out into multiple lower-speed links. In theory, this single switch could act as a dense aggregation point for dozens of 50GbE or 100GbE connections, all while sipping power compared to older gear.

If your mental model is “tiny spine switch,” you’re not wrong.

The Cost Problem Isn’t the Switch

Here’s where the story takes a turn.

The CRS804 itself is relatively affordable for what it is. Shockingly so. The problem is everything around it.

Affordable 400GbE NICs basically don’t exist yet. New cards from vendors like NVIDIA (via its Mellanox line) routinely cost more than the switch itself. We’re talking thousands of dollars for a single adapter. Used inventory is thin, and even when you find something, compatibility, firmware quirks, and optics support can turn into a weekend-long headache.

Dropping down to 100GbE helps a bit, especially on the secondary market, but now you’re already underutilizing what the switch is built to do. And once you start stacking breakout cables and multiple NICs just to “make use” of the ports, the simplicity starts to fall apart.

The switch isn’t the expensive part of a 400G setup. It’s the cheapest line item on the invoice.

PCIe Reality Check

There’s also the inconvenient matter of physics.

A single 400GbE NIC wants a full x16 PCIe Gen5 slot. Not Gen4. Not shared. Not “technically enough if nothing else is happening.” Full bandwidth, clean lanes, no compromises. That alone rules out a huge swath of desktop platforms and even many server boards that look powerful on paper but run out of lanes fast.

You can absolutely build a system around this. EPYC platforms, high-end Xeons, specialized AI or in-memory database servers — they all exist. But now we’re firmly out of “homelab curiosity” territory and into “this box has a job” land.

That’s not a criticism. It’s just context. The CRS804 is speaking fluently to workloads that already know what RDMA is and have strong opinions about latency jitter. Everyone else is overhearing the conversation.

Breakouts Are Clever, but They’re Not Magic

One of the more interesting angles with the CRS804 is its breakout potential. Each 400G port can be split into multiple lower-speed links — for example, eight 50GbE connections off a single port using the right DAC or optic.

Suddenly, this isn’t just a four-port monster. It’s a dense, half-width aggregation switch with surprisingly flexible personality.

That sounds great, and it is, but it comes with trade-offs. Breakout cables aren’t cheap. Optics add up fast. Cable management gets messy in a hurry. And you’re still building around relatively uncommon speeds like 50GbE and 100GbE that most access-layer networks haven’t standardized on yet.

It’s clever engineering. It just assumes the rest of your network has already made the same leap.

Where This Actually Makes Sense

Despite all of that skepticism, there are real use cases where the CRS804 feels perfectly at home.

High-speed storage clusters. In-memory databases. Media pipelines pushing uncompressed video. AI training nodes that need predictable, low-latency east–west traffic. Environments where RDMA, PFC, and offloads aren’t buzzwords but requirements.

In those spaces, the idea of a compact, relatively affordable 400G aggregation switch is genuinely appealing. The people building these systems already have the NICs, the servers, and the budgets. For them, the CRS804 isn’t “too early.” It’s right on time.

The tension comes from how visible this hardware is to everyone else.

The Missing Middle

One recurring frustration around MikroTik’s recent lineup isn’t that they’re going too big — it’s that the middle of the portfolio feels thin.

There’s a jump from very capable 10G and 25G switches straight to gear like the CRS804 and CRS812. If you’re trying to design a cohesive network with sensible access, aggregation, and core layers all within the same ecosystem, the pieces don’t always line up cleanly.

The CRS804 feels like a spine switch looking for leaves that don’t quite exist yet.

That’s not unusual in networking. Ecosystems evolve in fits and starts. But it does make this switch feel more aspirational than practical for a lot of buyers who would otherwise love it.

A Spec Sheet That’s Also a Statement

It’s tempting to view the CRS804 DDQ as overkill, but that misses part of the point. MikroTik has always had a habit of shipping gear that nudges its user base forward, sometimes before they’re fully ready.

This switch says, very clearly, that 400GbE isn’t some distant, hyperscaler-only future. It’s here. It fits on a shelf. You can buy it without calling a sales rep.

Whether you should is a different question.

For most people today, the honest answer is probably no. Not because it’s bad, but because the surrounding ecosystem — NIC pricing, server platforms, access-layer speeds — hasn’t caught up yet.

Early, Not Wrong

The CRS804 DDQ isn’t a mistake. It’s just early.

In a few years, when 100GbE NICs are cheap and 400GbE doesn’t sound exotic anymore, this class of switch will feel obvious. Inevitable, even. Right now, it feels like a glimpse into that future — impressive, slightly frustrating, and a little disconnected from everyday reality.

If you’re one of the few who can actually light up all four ports at full speed, this might be one of the most compelling pieces of networking hardware MikroTik has ever shipped.

Everyone else? You’re probably still window-shopping. And that’s okay.