Nvidia Vera Could Deliver 2x the Throughput of the AMD and Intel CPUs Modal Currently Uses in Preliminary Tests, but Detailed Benchmark Data Is Still Unavailable and Modal Will Continue Using a Hybrid x86 CPU Fleet
Modal’s preliminary testing indicates that Vera delivers twice the throughput of the AMD and Intel CPUs it currently uses. However, the company has not disclosed the CPU models or testing details, so it is not yet possible to conclude that Vera outperforms them in every workload.
Modal will continue using x86 CPUs from AMD and Intel while operating a heterogeneous server fleet, because migrating software to the Arm ISA requires substantial system modifications.
Where Vera Fills the Gap
Agentic AI is increasing the role of CPUs because they must run software, manage data, and handle numerous subtasks that GPUs are not well suited to perform entirely on their own.
Vera is positioned on the server side of AI infrastructure, handling these processing workloads while GPUs remain well suited to computational acceleration. This gives cloud providers such as Modal more options for allocating CPUs according to workload characteristics.
What the Available Vera Data Shows
The source contains no direct product images of Nvidia Vera and provides no details about the chip’s physical design. Its appearance, size, and component layout therefore cannot yet be determined.
If illustrations are used, they should be limited to images confirmed to be related to Vera or the server systems mentioned in the report. Details about the chip itself should not be invented based on unavailable information.
When AI Workloads Make CPUs a Bottleneck
AI infrastructure providers must support AI agents that perform complex tasks continuously over extended periods. Each workload competes for CPU cores, cache, and memory bandwidth while also processing software and data that GPUs are not well suited to handle.
Modal therefore tested Nvidia Vera against the AMD and Intel CPUs it currently uses. The preliminary results showed that Vera delivered twice the throughput on the company’s specific workloads, although detailed information about the test setup and the CPU models used for comparison remains unavailable.
Where Vera Fits in Nvidia’s Business Strategy
The deal with Modal suggests that Nvidia is moving beyond AI accelerators into standalone server CPUs to compete directly for market share in the server industry.
Nvidia expects Vera to generate at least $20 billion in sales during the current fiscal year, estimates the total addressable market for this class of CPUs at approximately $200 billion, and is targeting sales of at least 4 million units in fiscal 2026. These figures indicate that Vera is not merely a supplementary CPU in AI systems, but another major business pillar for Nvidia.
From AMD and Intel to Vera: What Changes
Modal previously used CPUs from AMD and Intel to run software and process data for agentic AI. However, preliminary internal testing found that Nvidia Vera delivered more than twice the throughput on the company’s workloads.
The AMD and Intel CPU models used for comparison, along with the benchmark configuration and performance-per-watt results, have not been disclosed. It is therefore not yet possible to conclude that Vera outperforms them in every situation. At the same time, Modal will continue using x86 CPUs and will need to make further software changes to support Vera’s Arm architecture.
How Vera’s Features Translate into Real-World Results
Vera uses Olympus cores based on Arm v9.2, capable of decoding and processing up to 10 instructions per cycle. This makes it suitable for agentic AI workloads involving multiple independent subtasks competing for CPU cores, cache, and memory at the same time.
Its 1.2 TB/s memory bandwidth benefits workloads that need to move large amounts of data during processing, such as running software and managing AI-agent data. However, it is still impossible to determine whether the Olympus cores, memory system, or other architectural factors are primarily responsible for Vera’s higher throughput compared with AMD and Intel.
How It Compares with Competitors When the Data Is Incomplete
Modal’s preliminary evaluation states that Nvidia Vera delivers twice the throughput of the AMD and Intel CPUs it currently uses. However, the test compared Vera only with currently available models, not the next-generation products from either company.
Modal has not disclosed the CPU models used, the benchmark configuration, or the performance-per-watt results. It is therefore unclear where Vera’s advantage comes from, and the results should not be treated as a final verdict that Vera is superior to AMD and Intel in every case.
Strengths and Limitations Observed in Modal’s Use
Nvidia Vera is well suited to agentic AI workloads that involve running software, processing data, and managing multiple tasks simultaneously. Modal’s preliminary testing found that it delivered twice the throughput of the AMD and Intel CPUs currently in use, making it a new option for specialized servers.
The limitation is that Vera uses the Arm ISA, requiring Modal to modify its software and take on additional engineering work before migrating its systems fully. Modal will continue using x86 CPUs while evaluating new processors from AMD and Intel before making a long-term decision.
The Costs That Do Not Appear on Vera’s Price Tag
Vera’s cost extends beyond the CPU price itself. Modal must modify its software to support the Arm ISA and maintain a server fleet containing both Arm and x86 systems. Engineering and systems-management work are therefore hidden costs that must be considered.
Modal must also evaluate multiple CPU categories, including the current version of Vera and future processors from AMD and Intel. The source provides no per-unit pricing or energy-cost data, so its cost-effectiveness cannot yet be quantified.
What Modal’s Decision Says About the Future of AI CPUs
Choosing Vera shows that CPUs still play an important role in agentic AI, especially for running software and processing data—tasks that GPUs are not always well suited to handle. However, migrating to the Arm ISA requires significant software changes, so the transition cannot happen immediately.
Modal will not stop using AMD and Intel. Instead, it will operate a heterogeneous server fleet, selecting CPUs according to workload characteristics while monitoring new processors from both companies and evaluating future options.