Antoine Chkaiban

Antoine Chkaiban asked about the feasibility, timeline, and economic outlook for space data centers, and how these might evolve over time.

Answer

Jensen Huang, President and CEO of NVIDIA, noted that while space data center economics are currently poor, they are expected to improve. He discussed the unique challenges of heat dissipation and the abundance of energy in space, highlighting imaging as a key use case for GPUs in space.

Antoine Chkaiban inquired about the feasibility, timeline, and economics of space data centers, and how NVIDIA views their evolution over time.

Answer

CEO Jensen Huang acknowledged that space data center economics are currently poor but will improve. He highlighted the abundance of energy but challenges with heat dissipation in space, requiring conduction and large radiators, as liquid cooling is impractical. He noted NVIDIA's Hopper GPU is already in space, with imaging being a prime use case for AI computation to process high-resolution data efficiently.

Antoine Chkaiban asked about the potential market size of Cisco Unified Edge, its use cases in retail, healthcare, and manufacturing, and preferred deployment strategies for edge AI compute. He also inquired about Cisco's plans to participate in the scale-up opportunity and potential partnerships.

Answer

Chuck Robbins, Chairman and CEO, expressed excitement for Unified Edge, highlighting its applicability for real-time inferencing in retail, restaurant chains, and healthcare. He sees deployment strategies as diverse, including CDN players, carriers/telcos offering it as a consumption service, and on-prem enterprise deployments. For the scale-up opportunity, he pointed to the Silicon One roadmap, stating Cisco's intent to play in that market as it transitions to Ethernet, with future developments expected.

Antoine Chkaiban inquired about the potential market size of Cisco Unified Edge relative to large-scale cloud data centers, asking for specific use cases in retail, healthcare, and manufacturing, and the preferred deployment strategy (CDN/operators vs. on-premise). He also asked if Cisco plans to participate in the scale-up opportunity and its materiality.

Answer

Chairman and CEO Chuck Robbins expressed excitement for Unified Edge, highlighting its applicability for real-time inferencing at the edge in retail, restaurant chains, and healthcare, and noted it's a revitalization of integrated compute/security at the edge. He sees deployment across CDN players, carriers offering it as a service, and on-premise. He confirmed Cisco's intent to play in the scale-up market, which is expected to transition to Ethernet, with future offerings anticipated.

Antoine Chkaiban asked if the developing relationship with OpenAI would tailor AMD's software stack development, how the collaboration works, and if it contributed to ROCm's robustness. He also inquired about the useful lives of GPUs, specifically if CSPs are planning to sweat their GPUs for longer than the typical 5-6 year depreciation period.

Answer

Dr. Lisa Su, Chair and CEO, confirmed that large customers, including OpenAI, contribute significantly to broadening and deepening the software stack, with deep collaboration planned across hardware, software, and future roadmaps. She noted that the work with OpenAI on Triton is valuable. Regarding GPU useful lives, Dr. Su observed some early indications of CSPs sweating GPUs longer, especially for inference with older generations like MI300X, alongside the demand for the latest GPUs for new infrastructure.

Antoine Chkaiban asked if the developing relationship with OpenAI is influencing the development of AMD's software stack, specifically how the collaboration works and if it has contributed to making ROCm more robust. He also inquired about the useful lives of GPUs, asking if CSPs are planning to extend their usage beyond the typical five to six years.

Answer

Lisa Su, Chair and CEO of AMD, confirmed that the OpenAI relationship, like all large customer engagements, contributes to broadening and deepening AMD's software stack, with plans for deep collaboration on hardware, software, systems, and future roadmaps, including work on Triton. She noted that feedback from a broad set of AI-native companies is strengthening ROCm, with AMD investing heavily and using AI to accelerate development. Regarding GPU useful lives, Lisa Su observed early indications of CSPs potentially extending usage, especially for inference, while still prioritizing the latest GPUs for new infrastructure.

Antoine Chkaiban asked about the progress of the Ultra Ethernet Consortium (UEC), the alignment of different voices within it, and key milestones investors should monitor going forward.

Answer

Jayshree Ullal, Arista Networks' CEO and Chairperson, confirmed that UEC published its first specification (1.0) in June 2025, and Arista's Ethernet portfolio is entirely UEC-capable. She discussed augmenting UEC with the ESUN specification, expanding UEC into scale-up configurations, and the modular framework for Ethernet.

Antoine Chkaiban asked for an update on the progress of the Ultra Ethernet Consortium (UEC), inquiring about the alignment among its members and key milestones investors should monitor going forward.

Answer

Jayshree Ullal, CEO and Chairperson, reported that UEC published its first specification (1.0) in June 2025, with Arista's Ethernet portfolio being fully UEC-capable and compliant. She highlighted the ESUN specification and OCP ESUN agreement as crucial for expanding UEC into scale-up configurations, leveraging a modular framework for Ethernet.

Antoine Chkaiban from New Street Research requested clarification on the requirements for "scale up" networking, how it differs from "scale out," and the potential evolution of the competitive landscape in that segment.

Answer

Chairperson & CEO Jayshree Ullal explained that "scale up" is a new requirement for high-bandwidth, low-latency interconnects within a single AI rack, currently dominated by proprietary NVLink. She stated that Arista sees an incremental market opportunity as open standards like scale-up Ethernet and UA link emerge, similar to how Ethernet is displacing InfiniBand in scale-out back-end networks.

Antoine Chkaiban asked for Arista's latest views on co-packaged optics (CPO), particularly whether NVIDIA's recent CPO switch announcement has altered its perspective on CPO adoption in AI networks.

Answer

CEO Jayshree Ullal stated that NVIDIA's announcement had no impact, as Arista still views CPO as a very early-stage technology with historical reliability issues. She affirmed Arista remains technology-agnostic and will support whatever customers adopt—be it pluggable optics, co-packaged copper, or CPO—but considers CPO not yet ready for production.

Antoine Chkaiban asked what Arista is hearing from classic service providers regarding AI-driven demand and network build-outs.

Answer

CEO Jayshree Ullal responded that Arista has not yet seen a significant uptick in AI-driven demand from classic service providers. However, she highlighted substantial activity from a new class of 'neo clouds' and Tier 2 specialty cloud providers, including sovereign cloud initiatives, that are investing heavily to provide AI as a service.

Antoine Chkaiban requested an update on the major AI trials, asking about their progress, the timeline for moving to production, and the potential scale of these deployments.

Answer

CEO Jayshree Ullal revealed that Arista is now engaged in five major AI trials, up from four. She reported that three are progressing well and are expected to scale to 50,000-100,000 GPU clusters in 2025. A fourth trial has begun with a customer historically focused on InfiniBand, while a fifth is moving slower than anticipated due to GPU availability and infrastructure challenges.

Antoine Chkaiban requested details on the planned rollout, economics, and deployment pace for the Lyft robotaxi deal, and asked for more color on the cloud-enhanced ADAS win with a Korean OEM.

Answer

Executive Daniel Galves outlined a multi-stage robotaxi rollout process, from testing to full commercial service. CEO Amnon Shashua reiterated the economic model of a one-time payment plus a recurring license fee. On the ADAS win, Galves explained it's a high-volume project crucial for their data ecosystem strategy, with a rollout planned for the 'near future'.

Antoine Chkaiban of New Street Research asked about the collaboration with current SuperVision customers like Polestar and VW, and for feedback on the DXP framework that allows OEM customization.

Answer

CEO Amnon Shashua explained that development focus has shifted from the mature first-generation SuperVision to a complete software rewrite for the next-gen EyeQ 6 platform, which is already in pre-production testing. He confirmed that the DXP framework is in 'full use' with current SuperVision partners and is a critical part of the development stack.

Antoine Chkaiban asked about Zeekr's shift to its in-house software on NVIDIA hardware in China, inquiring about the performance of that system compared to Mobileye's. He also asked for more detail on the challenges of building out its REM mapping data in China.

Answer

CEO Amnon Shashua stated that Zeekr's in-house system costs twice as much for similar performance and believes the change was a strategic decision driven by a top-down push in China for local technology. He noted Mobileye still benefits from the relationship through data collection and collaboration on a parking system. An executive, likely Nimrod Nehushtan, added that Mobileye works with local map providers to ensure its technology remains fully compliant with Chinese regulations.