Omar Hassen

Co-founder & CPO @EriduAI
Opinions are my own and do not represent Eridu
Reposts ≠ Endorsement

[Omar Hassen @omarhassen]

https://t.co/NVb4dBXxcR

[Omar Hassen @omarhassen]

I'm speaking at AI Infra Summit next month on what is needed to solve one of the critical challenges in AI infrastructure--the network bottleneck. I'll discuss how high-radix switches address that bottleneck, improving utilization and performance in AI clusters.
Sept 15th @ noon

[Omar Hassen @omarhassen]

@EriduAi is transforming the AI Data Center network!

[Eridu @EriduAi]

Network switching has struggled to keep up with the demands of AI, making the network wall a critical challenge for AI deployments at scale.

Join this educational webinar on July 28 at 9am PT / 12pm ET to hear experts from @OpenComputePrj, @650Group and Eridu discuss why high-radix network switching will play a critical role in current and future AI data centers.

Register to save your spot at https://t.co/BeNXDN5Dhl

[Eridu @EriduAi]

Coming up next month: learn how high-radix network switching addresses critical challenges in AI data centers from experts in the field.

Join @AlanWeckel from @650Group and Cliff Grossner from @OpenComputePrj together with Promode Nedungadi and Jay Gill from Eridu for a discussion on scaling challenges in AI networks today, and how to use high-radix switching to build flatter, more scalable networks and larger clusters for AI training and inference.

Register now to save your spot!

https://t.co/BeNXDN5Dhl

[Eridu @EriduAi]

The rapid evolution of AI continues to increase the importance of high bandwidth, low latency data movement. Larger domains and clusters, expanding context windows, disaggregated processing and agentic workflows (just to name a few) all depend on being able to deliver data at speed and scale to GPUs and XPUs.

At the recent OCP Summit in Barcelona, Eridu's @Mike_Capuano took a closer look at the implications of these changes on AI network infrastructure, and at how high-radix switching is a critical part of what's needed to keep up with the demands of AI.

Watch the full video at https://t.co/rt969Y5wFn

[Eridu @EriduAi]

"Higher radix switches deliver decisive advantages in AI networking. They consolidate equipment counts, drive down costs and power demands, and enable vastly larger, efficient clusters." --Alan Weckel, 650 Group

As AI clusters and data centers continue to grow exponentially, networking bandwidth is on a much slower, incremental rate of advance. The result is that the share of power, cost, and space consumed by the network continues to grow on an unsustainable trajectory.

High radix switching offers a solution to this critical challenge, making it possible to keep up with the growing demands of scale up and scale out clusters.

Read the analysis by 650 Group at https://t.co/OnnBpENVQj

[Eridu @EriduAi]

Eridu co-founders @DrewDPerkins and @omarhassen sat down with @tfir_io to discuss why traditional network switching has not been able to keep with the demands of AI, the impact of network latency on the performance and economics of AI data centers, and the architectural shift to higher-radix solutions that enable flattening of network tiers.

https://t.co/M79s4U1QpW

[Eridu @EriduAi]

A new article from 650 Group highlights why higher radix switches are critical to next-generation AI data centers.

For large-scale GPU clusters, traditional network architectures create bottlenecks that compound quickly. Higher radix addresses this by:
• Reducing the number of network tiers required at scale
• Improving latency and overall data transfer time through fewer hops
• Driving better power efficiency and TCO with simplified topologies

High radix changes the equation, enabling more direct connectivity and better alignment between compute and network performance.

Read the full analysis: https://t.co/OnnBpENVQj

[Eridu @EriduAi]

AI infrastructure is hitting a simple constraint: when the network can't keep up, GPUs sit idle.

The challenge facing organizations isn't just about getting more GPUs, it’s about ensuring the network can feed them at scale.

Even with large investments in compute, AI clusters underperform because data isn’t delivered fast enough. The result is lower utilization, slower training and inference, and wasted investment.

That's why Eridu is rethinking networking infrastructure from the ground up: silicon, systems, and architecture.

[Omar Hassen @omarhassen]

Join us on this amazing journey!

[Eridu @EriduAi]

We’re hiring to expand our ASIC engineering team!

AI clusters have pushed beyond what traditional networking was designed to handle. We're solving the resulting network bottleneck by building a clean-sheet network switch designed from the ground up for AI.

Join our stellar ASIC engineering team to play a key role in building that solution. You'll get to:
➡️ Work on performance-critical infrastructure for next-generation AI clusters
➡️ Collaborate with a world-class team with deep experience in silicon, packaging, and hyperscale systems
➡️ Help us deliver a step function leap in network switching throughput, latency, and efficiency

Your contributions will directly impact the next generation of AI infrastructure solutions, transforming the performance of AI data centers.

Learn more and apply online at https://t.co/VFeCXzhYyt.

[Eridu @EriduAi]

What's behind the name Eridu?

Eridu was the name for what is often described as one of the world’s first cities. Founded between 5300 and 5500 BCE in ancient Mesopotamia, Eridu was a place associated with early systems, structure, and the foundations of organized society.

That aligns with what we’re building today. AI is still in its early stages, and what we're developing is foundational infrastructure for AI.

Like the first cities, progress in AI will depend on the strength of that foundation, shaping what can be built on it.

At Eridu, we're defining and building the networking foundation for modern AI. Learn more at https://t.co/7wB6LJewu3.

[Omar Hassen @omarhassen]

Extremely grateful for the great support from TSMC!

[Eridu @EriduAi]

To build the next generation of network switches, it's critical to leverage leading-edge technology throughout the stack.

That's why we're working with TSMC to use their advanced technologies for our custom-designed networking ASICs.

[Omar Hassen @omarhassen]

Eridu is addressing the network inefficiencies in AI data centers.

[Eridu @EriduAi]

Networking is falling behind the rapid advances in GPUs and AI algorithms--while AI compute is advancing at a near-exponential rate, networking is improving only incrementally.

Adding more switches and optics to current architectures won't solve it. The bottleneck remains:
⚠️ GPUs sit idle waiting for data
⚠️ Latency continues to increase
⚠️ Inefficiency grows in power, cooling and cost

Eridu is building a clean-sheet design to create a new frontier in network switching.

Read more about our vision: https://t.co/aUfwLuERLI

[Eridu @EriduAi]

This week's discussions at GTC and OFC made one thing clear: the transition from copper to optical interconnect is critical to meeting AI demands.

Scale-up domains are expanding from 72 GPUs to 144, 576 and beyond. Copper can't keep up with these large, multi-rack domains needed for advanced training and inference.

At Eridu, we're delivering an order-of-magnitude advance in optical network switching to meet the demands of expanding scale-out and scale-up domains.

Visit us online to learn more.

[Omar Hassen @omarhassen]

The network is key to achieving AI productivity and efficiency metrics!

[Eridu @EriduAi]

At #GTC26 this week, Jensen Huang reframed data centers as AI factories. Those factories are now measured by:
• Tokens per watt (efficiency)
• Throughput at scale (productivity)

Both depend on the network. The network is a critical constraint that determines:
• Whether GPUs sit idle or stay utilized
• How much power is wasted waiting on data
• The real output of the cluster and factory

Eridu is building a frontier networking switch designed to remove the network bottleneck and unlock AI factory performance with an order-of-magnitude advance--in performance, radix and efficiency.

More at https://t.co/5ATfwP0p6I

#AIinfrastructure #GTC26

[Omar Hassen @omarhassen]

If you like to work on groundbreaking technology, join us..

[Eridu @EriduAi]

Eridu is hiring! We're expanding our team as we accelerate our product development following the recent announcement of our $200M in Series A funding. We're growing our ASIC, software and systems teams in the Bay Area and in India.

If you're interested in building the solutions that will power the future of AI, join us.

Read more about our announcement in @svbizjournal and check out our open roles on our website at https://t.co/5ATfwP0p6I.

[Eridu @EriduAi]

Eridu is hiring! We're expanding our team as we accelerate our product development following the recent announcement of our $200M in Series A funding. We're growing our ASIC, software and systems teams in the Bay Area and in India.

If you're interested in building the solutions that will power the future of AI, join us.

Read more about our announcement in @svbizjournal and check out our open roles on our website at https://t.co/5ATfwP0p6I.

[Eridu @EriduAi]

The network wall is a critical problem for both scale-out and scale-up networks in AI datacenters.

We spoke with @sdxcentral about our announcement this week of our $200M in Series A funding to continue our work building a next-generation networking switch designed to break through that wall.

Read the article at the link below. #ainetworking #aidatacenter

[Omar Hassen @omarhassen]

A clean-sheet design is required to achieve the network scale required in AI data centers

[Sean Kerner @TechJournalist]

AI networking startup @EriduAi launches with $200M in funding to redesign AI data center networking using a clean-sheet network switch and customer silicon. https://t.co/nd4lvHIJ7E