MaxLinear Annapurna 224G retimer targets 224Gbps copper links for AI racks

MaxLinear announced Annapurna, a 224G scale-up retimer designed to extend copper connectivity in AI data centers, with the company positioning it for copper backplanes and Active Electrical Cables (AECs) operating at 224 Gbps per lane using PAM4 signaling. MaxLinear says the device delivers up to 1.6 Tbps of electrical connectivity and is intended to let architects keep electrical interconnects in place where they have power, latency, and cost advantages versus optical links.

MaxLinear is framing Annapurna for “hybrid optical and electrical AI architectures,” where optical interconnects remain the choice for longer-reach scale-out networks, and electrical stays attractive for scale-up networking within and between racks. The company argues that at 224G per lane, copper channels run into reach limits from attenuation, return loss at interfaces, and impairments such as skin-effect resistance, copper roughness, and inter-symbol interference, which can constrain where chips must be placed relative to the backplane.

On the silicon side, MaxLinear says Annapurna uses advanced adaptive equalization, transmit pre-emphasis, and proprietary DSP algorithms to address those 224G-per-lane electrical channel impairments. The company is offering two variants: eight-lane and sixteen-lane configurations. MaxLinear lists support for 1.6 Tbps and 3.2 Tbps applications, and calls out AEC and on-board retimer deployments as target use cases.

For AI fabrics, MaxLinear is emphasizing “ultra-low latency” for scale-up GPU-to-GPU synchronization and “high power efficiency,” without publishing numeric latency or power figures in the announcement. It also lists “comprehensive telemetry and diagnostics,” and states compliance with IEEE 802.3dj and OIF CEI-224G standards.

“Within a data center, optical excels at longer reaches for scale-out networks, while copper remains highly attractive for scale-up networking if signal integrity challenges at 224G per lane can be addressed,” said Rajneesh Gaur, Senior VP & GM, Data Center Connectivity Business Unit, MaxLinear. “Annapurna delivers the signal processing required to keep electrical connectivity viable at these speeds.”

MaxLinear also says Annapurna complements its optical DSP portfolio and supports emerging scale-up protocols including ESUN, UALink, and Ultra Ethernet, positioning it as a building block for mixed interconnect strategies across an AI fabric. In an endorsement quote, Aravind Srikumar, SVP, Product & Marketing, Upscale.AI, said, “Annapurna is an important enabler for signal integrity across our rack-scale AI systems.”

Availability: MaxLinear says the Annapurna Scale-Up Retimer is expected to be available in Q2 2026, and directs interested buyers to contact MaxLinear sales.

Source: MaxLinear

Get Data Center Engineering News In Your Inbox:

Popular Posts:

Screenshot
Five AI data centers to reach 1 GW power capacity in 2026, new analysis shows
Near-Packaged-Optics--Rethinking-the-AI-Data-Center-Interconnect
Near-Packaged Optics: Rethinking the AI Data Center Interconnect
30cf-data-center-pr
Carrier launches AquaEdge 30CF chiller to boost data center cooling reliability and uptime
Mosaic-4148-2000x1333_1_1
300mm silicon carbide wafers pitched for AI data center packaging by 2030
shine 的複本 的複本 - 36
GenerMotor launches stackable HVDC generator modules for AI data center power

Share Your Data Center Engineering News

Do you have a new product announcement, webinar, whitepaper, or article topic? 

Get Data Center Engineering News In Your Inbox: