AnswersPublished13 sources
Modal Plans Nvidia Vera CPUs for AI Agents After a 2× Throughput Test
Modal plans to add Nvidia Vera CPUs to its Sandboxes service after its own 50 task test showed up to twice the throughput of an x86 system under concurrent load. Vera adds an Arm based option to a fleet that has relied on Intel and AMD CPUs; Modal says supporting the new architecture will require changes to its infr...
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Modal plans to add Nvidia Vera CPUs to its Sandboxes service after its own 50 task test showed up to twice the throughput of an x86 system under concurrent load.
- Vera adds an Arm based option to a fleet that has relied on Intel and AMD CPUs; Modal says supporting the new architecture will require changes to its infrastructure and image build system.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did AI cloud startup Modal announce on October 19 about its first adoption of Nvidia’s Vera CPUs for hosting AI agents, how does the Ar. Article summary: On October 19, 2026, Modal announced plans to use Nvidia’s Vera CPUs for hosting AI agents—its first move beyond the Intel and AMD processors it had previously used for that work. The reported advantage is higher capacit. Topic tags: general, news, general web, user generated, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, water
Modal said on October 19, 2026, that it plans to use Nvidia’s Vera CPUs to host AI agents in its Sandboxes service. The move would add an Arm-based processor to a workload that Modal had previously run on Intel and AMD chips. 1 Its headline result—up to twice the throughput in a concurrent-load test—is a company-reported benchmark, not evidence that every individual agent task will finish twice as quickly.
7
What Modal tested—and what “2× throughput” means
Modal’s Agentless Agent Bench uses 50 Terminal-Bench reference solutions and their verifiers. Modal says it removed network calls and ran the work in prewarmed, concurrent Docker containers to focus on CPU tasks such as compilation, file searching, Git operations and device emulation.
In those concurrent tests, Modal reported up to twice the task throughput on Vera compared with an equivalent Intel Granite Rapids x86 machine. Throughput means the amount of work completed across a set of jobs over time. It is different from single-task speed: in lighter-load tests, Vera was often faster, but the lead was not consistent or large. 7
The result is specific to Modal’s benchmark and workloads. Modal published the test, and the supplied reporting says the results have not been independently checked. 7 They should not be read as a general guarantee for other AI workloads or deployments.
Why memory bandwidth may matter
Sandbox workloads can involve many containers doing CPU work at once. Modal points to Vera’s memory bandwidth as a factor that can support more containers running concurrently. Nvidia’s published specifications list up to 1.2 TB/s of LPDDR5X memory bandwidth for Vera.
That provides a plausible technical explanation for sustaining concurrent work, but it does not independently establish what caused Modal’s benchmark result. The measured claim remains Modal’s reported throughput for its selected tests.
Vera’s architecture and Modal’s integration work
Vera is an Arm-based CPU associated with Nvidia’s Vera Rubin platform and is also described as available in standalone CPU servers. For Modal, adding it means supporting a processor architecture beyond its previous Intel- and AMD-based fleet. 1
Modal says it will need to make changes across its infrastructure and image-build system to support the different CPU architecture. That is a meaningful part of the adoption: the benchmark result alone does not make existing software images and systems automatically compatible.
A new CPU option, not a full supplier switch
Modal plans to keep its relationships with Intel and AMD while adding Vera. 13 The announcement therefore points to a mixed fleet, not an exclusive switch to Nvidia. A measured advantage on one concurrent workload does not establish that a single processor is best for every task; the available reporting does not detail how Modal will route workloads among its CPUs.
7
13
The deal gives Nvidia an opening in CPU workloads at a cloud provider that had relied on Intel and AMD for agent hosting. 1 But the immediate news is narrower: Modal plans to add Vera, its benchmark showed a throughput gain under load, and the company still has infrastructure work to do before supporting another architecture.
1
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Modal Plans Nvidia Vera CPUs for AI Agents After a 2× Throughput Test"?
Modal plans to add Nvidia Vera CPUs to its Sandboxes service after its own 50 task test showed up to twice the throughput of an x86 system under concurrent load.
What are the key points to validate first?
Modal plans to add Nvidia Vera CPUs to its Sandboxes service after its own 50 task test showed up to twice the throughput of an x86 system under concurrent load. Vera adds an Arm based option to a fleet that has relied on Intel and AMD CPUs; Modal says supporting the new architecture will require changes to its infrastructure and image build system.
Sources
rmb.reuters.comNvidia wins fast-growing AI startup Modal for new central processors
thenextweb.comModal says Nvidia's Vera CPU doubles throughput for its AI sandboxes
ca.investing.comNvidia wins CPU business from Modal, previously exclusive to AMD and Intel By Investing.com
english.aawsat.comManus Raises More Than $500 Million after Meta Exit