Nvidia has a new kind of customer for its data center hardware, and it has nothing to do with GPUs. Modal, a fast-growing AI cloud computing startup, confirmed on October 19, 2026, that it is adding Nvidia’s Vera central processor to its infrastructure for hosting AI agents, the software systems that carry out extended, multi-step tasks for users. Reuters, The Economic Times, and The Information all picked up the story the same day, with outlets including Hoodline noting it marks the first time Modal has run anything other than Intel or AMD chips for this kind of workload.
The headline framing circulating online, that Nvidia “won” Modal outright, overstates what has actually been confirmed. There is no evidence of an acquisition. What is confirmed is narrower and still significant: Modal is bringing on Nvidia’s Vera CPU as a customer of the chip, running it alongside its existing Intel and AMD fleet rather than replacing either vendor. Early internal testing, according to Modal, showed the Vera CPU doubling the amount of agent work completed compared with an equivalent x86 machine. That claim, and the engineering effort behind it, is the real story here.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Modal and the Reports Actually Confirm
Strip away the speculation and a handful of verified facts remain. Modal plans to deploy Nvidia’s Vera central processor to host AI agents inside its cloud. Before this announcement, Modal’s fleet for these workloads ran exclusively on Intel and AMD silicon, both built on the x86 architecture that has dominated cloud computing for decades. Nvidia’s Vera chip breaks that pattern because it is Arm-compatible, not x86, which means Modal has to rebuild parts of its infrastructure and image build system to support it.
Modal says it will not swap one vendor for another. Instead, Vera joins a mixed fleet where Intel, AMD, and now Nvidia hardware run side by side, with workloads routed to whichever processor performs best for the job. That detail matters because it signals caution rather than a wholesale migration. Modal is hedging, not betting the business on a single new architecture on day one.
One more detail stands out: Nvidia reportedly hand-delivered Modal’s first Vera CPU server to the startup’s San Francisco office the week before the public announcement, a white-glove gesture that underscores how much Nvidia wants design wins for Vera beyond its own GPU ecosystem.
Who Is Modal and Why Its CPU Choice Carries Weight
Modal is a cloud computing startup that rents compute capacity to AI developers, with a focus on sandboxed execution environments and infrastructure for running AI agents at scale. It sits in the same broad category as other AI-infrastructure clouds that have emerged over the past few years, competing for developers who need to spin up compute quickly without managing their own hardware.
Why does one startup’s chip choice matter to the wider industry? Because Modal’s workload, AI agents executing many short, concurrent tasks, is exactly the kind of job cloud providers are racing to optimize for as agentic AI moves from demos into production. If Vera genuinely improves throughput for that workload, every cloud provider running similar agent infrastructure has a reason to pay attention. Modal becoming a visible, named reference customer gives Nvidia a concrete proof point it can point to when pitching other AI clouds on Vera, something that matters more than the deal’s dollar value, which has not been disclosed.
Inside Nvidia’s Vera CPU and the Vera Rubin Platform
Vera is the CPU half of Nvidia’s Vera Rubin AI-computing platform, pairing a custom Arm-based processor with Nvidia’s Rubin-generation GPU architecture. Nvidia also sells Vera as a standalone CPU server for customers who want the chip without the full accelerator stack, which is the configuration Modal is adopting first.
According to reporting on the chip’s design, Vera uses an 88-core custom architecture internally referred to as “Olympus,” paired with up to 1.2 terabytes per second of memory bandwidth. That bandwidth figure is the headline spec, because it directly targets the bottleneck Nvidia is trying to solve: AI agent orchestration requires constant, rapid coordination between the CPU and accelerator hardware, and a starved memory pipeline slows that coordination down regardless of how fast the GPU itself runs.
What is not yet confirmed publicly is a complete commercial release timeline covering every Vera Rubin configuration, or a finalized spec sheet for the paired Rubin GPU, including core counts and memory capacity. Nvidia has disclosed enough to explain the Modal deployment, but broader availability details remain unconfirmed as of this report.
The 2x Throughput Claim, Explained
The number driving most of the coverage is simple: Modal says its early benchmarking showed Vera delivering up to double the task throughput of an equivalent x86 machine. That is Modal’s own internal benchmark, run against real-world agent workloads under contention, meaning hundreds of concurrent tasks competing for the same resources at once, which is a reasonably realistic stand-in for production conditions.
The more interesting part of the claim is the latency behavior. Modal reported that individual task latency on Vera measured similar to, or in some cases lower than, the x86 comparison machine, even while handling roughly twice the concurrent load. In plain terms, Vera was not just doing more work by sacrificing responsiveness per task, it appeared to do more work without making any single task noticeably slower. If that holds up under independent testing, it is a meaningfully different result than simply adding more cores and calling it a throughput win.
It is worth being precise about what this is not. The 2x figure is not an independently verified industry benchmark, and it has not been replicated by a third party as of this writing. It is a vendor-adjacent claim from the customer itself, published on Modal’s own blog, which makes it a useful early signal but not conclusive proof of real-world performance at scale.
Why the Jump to Arm Architecture Is a Bigger Deal Than It Looks
Cloud infrastructure has run almost entirely on x86 chips from Intel and AMD for most of its history, with Arm making inroads mainly through chips like AWS’s Graviton line. For a cloud company like Modal, adding an Arm-compatible processor is not a drop-in hardware swap. It means rebuilding container images, adjusting the build pipeline, and retesting software that may have assumed an x86 instruction set somewhere in its dependency chain.
Modal has been explicit that this requires real engineering work across its infrastructure and image build system. That a startup is willing to absorb that cost for a CPU that is not yet widely available, rather than wait for a more mature ecosystem, is itself a signal of how much weight it is placing on the throughput numbers it is seeing in testing.
A Mixed Fleet: Vera Alongside Intel and AMD
Nothing about this deal removes Intel or AMD from Modal’s infrastructure. The company has been explicit that Vera is being added to, not swapped in for, its existing hardware. The practical effect is a routing decision: workloads get sent to whichever processor, Intel, AMD, or now Nvidia’s Vera, performs best for that specific job.
This mixed-fleet approach is a sensible hedge for any cloud provider dealing with a new, unproven architecture. It lets Modal capture upside from Vera’s reported throughput gains on agent workloads without exposing the whole business to a single vendor’s roadmap, supply, or pricing. For Intel and AMD, it is also a less alarming outcome than an outright replacement, though it still represents a crack in what had been exclusive share of Modal’s CPU spend, as outlets including Yahoo Finance and Reuters reported on October 19.
Routing Agent Workloads by Performance
The agent-workload angle is worth dwelling on because it explains why Modal moved on Vera specifically rather than for its general-purpose compute. AI agents tend to run many short, bursty tasks rather than one long, steady job. That pattern rewards a chip that can juggle high concurrency without latency spikes, which is precisely the behavior Modal says it measured. A CPU optimized for that pattern will not necessarily look better on every workload Modal runs, which is exactly why a mixed fleet, rather than a full migration, is the rational first move.
What Modal Is Saying About the Switch
Modal’s own public statements, published on its company blog and on social media, give the clearest picture of the company’s reasoning. Modal said: “We got our hands on @nvidia’s new Vera CPU and put it to the test. We built a benchmark using real-world agentic workloads to see how it performs under contention. In early tests, Vera delivered up to 2x the throughput of comparable x86 hardware, maintaining low latency even as hundreds of concurrent workloads competed for resources.”
On the engineering cost of the switch, Modal was candid: “We’d need to make changes across our infrastructure and image build system to support an entirely different CPU architecture, but the benchmarks are impressive: a 2x increase on throughput.” The company also described the delivery itself, noting that “last week, NVIDIA hand-delivered Modal’s first Vera CPU server to our SF office,” and summarized its overall read on the chip by saying “with this, the NVIDIA Vera CPU becomes an incredibly compelling option for powering Modal Sandboxes.”
In a more detailed technical comparison on its blog, Modal added: “Our early benchmarking runs show up to double the task throughput on Vera compared with an equivalent x86 GNR machine, despite individual task latency measuring equal to, or sometimes even lower than, the x86 machine.” No other named individual, at Modal, Nvidia, or elsewhere, has an on-record quotation attached to this story as of October 19, 2026.
How Nvidia Got Into the CPU Business
Nvidia’s push into CPUs is not new, though this deal gives it fresh momentum. The company has spent several years building out server-grade Arm processors to pair with its GPUs, aiming to control more of the data center stack rather than ceding the CPU socket to Intel or AMD by default. Vera is the latest and most agent-focused entry in that lineage, designed specifically to pair with the Rubin GPU generation under the Vera Rubin platform name.
What makes the Modal deal different from Nvidia’s earlier CPU efforts is the use case. Earlier Arm server pushes from Nvidia and others were pitched broadly at data center efficiency and power savings. Vera’s pitch to Modal is narrower and more specific: it is built to remove CPU-side bottlenecks in AI agent orchestration, a workload category that barely existed at meaningful commercial scale a few years ago and is now central to how AI clouds differentiate themselves.
Nvidia Vera vs Intel Xeon vs AMD EPYC: A Side-by-Side Look
Direct spec-for-spec comparisons are difficult right now because Nvidia, Intel, and AMD disclose different details for different workloads, and not every number for every chip has been independently verified for this specific use case. The table below lays out what is confirmed versus what remains general industry positioning.
| Processor | Architecture | Vendor | Role at Modal | Reported Agent-Workload Edge |
|---|---|---|---|---|
| Vera | Arm (custom “Olympus” design) | Nvidia | New addition, agent-workload focus | Up to 2x throughput vs. x86 in Modal’s early tests |
| Xeon (current generation) | x86 | Intel | Existing fleet, general workloads | Mature ecosystem; not separately benchmarked in this report |
| EPYC (current generation) | x86 | AMD | Existing fleet, general workloads | Mature ecosystem; not separately benchmarked in this report |
The honest takeaway from this table is that Vera’s advantage, as reported, is specific to one workload type under one company’s internal testing. It says nothing conclusive about how the three chips compare on database workloads, web serving, or other general-purpose cloud tasks where Intel and AMD have had decades to optimize.
Modal’s CPU Fleet, Before and After Vera
The clearest way to see the actual scope of this news is to compare Modal’s infrastructure before and after the announcement, since the “Nvidia acquires Modal’s CPU business” framing circulating in some headlines overstates what changed.
| Aspect | Before October 2026 | After October 2026 Announcement |
|---|---|---|
| CPU vendors in Modal’s fleet | Intel and AMD only | Intel, AMD, and Nvidia (Vera) |
| Architecture support | x86 exclusively | x86 plus Arm (first Arm move) |
| Agent workload routing | Routed across Intel/AMD only | Routed across Intel, AMD, and Vera by performance |
| Reported throughput for agent tasks | Baseline x86 performance | Up to 2x throughput reported on Vera in early tests |
| Infrastructure/build system | Built for x86 only | Being rebuilt to also support Arm |
Market Impact: What This Means for Intel and AMD
For Intel and AMD, the immediate financial impact of one startup adding a third CPU vendor is close to nothing. Modal is not a hyperscaler, and no contract value has been disclosed. The more important impact is reputational and strategic. Both chipmakers have built their data center businesses on being the default, unquestioned choice for cloud CPU workloads. A visible, well-covered customer choosing to add Nvidia silicon, even in a limited, workload-specific way, chips at that default status.
That pressure arrives at an awkward moment. Intel has already raised CPU prices multiple times this year as it works through its own cost pressures, a pattern tracked in tech-insider.org’s coverage of Intel’s third 2026 price hike, and framed by Global Banking & Finance as part of Nvidia’s broader push to secure AI startups as central-processor customers. A credible, if early, challenger to x86’s hold on agent-workload hosting adds a new variable to that pricing conversation, especially if other AI clouds follow Modal’s lead and ask Intel and AMD to justify their pricing against a chip that claims double the throughput on a growing workload category.
For Nvidia, the calculus is different. The company’s hardware ambitions and spending commitments have drawn scrutiny over how its enormous AI infrastructure bets connect to actual demand, a tension examined in tech-insider.org’s reporting on Nvidia’s $500 billion AI bet. A real, named customer adopting Vera for a production workload, rather than a pilot or a press-release partnership, is a tangible data point Nvidia can use to answer that scrutiny, even if the deal’s financial scale is modest next to its GPU business.
The Competitive Landscape: Modal and the AI Cloud Pack
Modal operates in a crowded field of AI-infrastructure clouds that rent compute to developers building and running AI applications. Available reporting does not provide a verified, head-to-head comparison of Modal against named rivals in this category, so no specific customer counts or market-share figures for those competitors can be stated here. What can be said is that Modal has reportedly raised a $750 million Series D round led by Accel and reported more than $1 billion in annualized recurring revenue, according to figures cited in recent coverage, though founder details and a confirmed valuation were not substantiated in the available reporting.
That scale matters for context. A company reporting nine-figure annual revenue is not a toy startup experimenting with exotic hardware for publicity. Its CPU decisions reflect real production traffic, which is part of why Nvidia appears to have invested effort, including the hand-delivered server, in getting this specific deployment right. Other AI clouds watching this space, including those offering GPU and CPU bundles similar to Modal’s, now have a public benchmark claim to react to, whether by testing Vera themselves or by publicly defending their existing x86 commitments.
Nvidia has also been building Vera’s presence with infrastructure providers beyond Modal, a trend visible in tech-insider.org’s earlier coverage of CoreWeave’s plans to offer Nvidia Vera and in reports that server maker HPE has already shipped Vera-based systems, covered in tech-insider.org’s report on HPE’s Vera CPU server debut. Modal’s announcement adds a cloud-native, developer-facing name to that growing list.
What This Means for Enterprise AI Buyers
For enterprises evaluating where to run AI agent workloads, this deal is a useful early data point rather than a reason to change vendors immediately. A 2x throughput claim from one customer’s internal benchmark is not the same as an independently audited result across varied workloads, and the price difference between running on Vera versus Intel or AMD capacity at Modal has not been disclosed.
What buyers evaluating AI cloud providers should watch for is whether other providers publish their own Vera benchmarks, and whether Modal itself follows up with more granular, third-party-verifiable numbers as its Vera rollout matures. Cloud buyers comparing GPU and CPU pricing across providers, a comparison tracked in tech-insider.org’s AWS vs Azure vs GCP GPU pricing breakdown, now have one more variable to factor in if Arm-based CPU options start showing up in more providers’ catalogs.
Predictions: Where the Vera Bet Goes From Here
- Expect more named AI-cloud customers to confirm Vera trials over the next two to three quarters, following the pattern set by Modal, CoreWeave, and HPE’s server rollout.
- Independent benchmarking firms are likely to attempt to replicate or challenge Modal’s 2x throughput claim once more Vera hardware becomes available outside Nvidia’s direct partners.
- Intel and AMD will likely respond with their own agent-workload-specific performance marketing rather than price cuts alone, since the Vera pitch is about architecture fit, not just cost.
- More cloud providers will adopt mixed-fleet strategies similar to Modal’s rather than full migrations, given the engineering cost of supporting a second CPU architecture.
- Nvidia will likely use Modal as a reference case in future enterprise sales conversations for Vera, given the scarcity of named, production-scale customers disclosed so far.
Frequently Asked Questions
Did Nvidia acquire Modal?
No. Available reporting confirms Modal is adopting Nvidia’s Vera CPU as a customer, not that Nvidia acquired the company. Headlines describing Nvidia “winning” Modal refer to winning its CPU business, not an acquisition.
What is Nvidia’s Vera CPU?
Vera is an Arm-compatible central processor Nvidia designed as the CPU component of its Vera Rubin AI-computing platform, which also includes Nvidia’s Rubin-generation GPU architecture. Nvidia sells Vera as a standalone CPU server as well.
Is Modal dropping Intel and AMD?
No. Modal says it will run Vera alongside its existing Intel and AMD processors in a mixed fleet, routing workloads to whichever chip performs best.
How much faster is Vera than x86 chips?
Modal reported its early internal testing showed up to 2x the task throughput on Vera compared with an equivalent x86 machine, with similar or lower per-task latency. This is Modal’s own benchmark and has not been independently verified by a third party.
What is Modal?
Modal is an AI cloud computing startup that rents compute capacity to AI developers, including infrastructure for hosting AI agents and sandboxed execution environments.
Why does Modal need to rebuild infrastructure for Vera?
Vera uses an Arm-compatible architecture, while Modal’s existing fleet runs on x86 chips from Intel and AMD. Supporting a different CPU architecture requires changes to the image build system and other infrastructure components.
Has Nvidia disclosed full specs for Vera and the Rubin GPU?
Not completely. Reports describe Vera as using an 88-core custom design with up to 1.2 terabytes per second of memory bandwidth, but a full commercial release timeline for every Vera Rubin configuration and complete Rubin GPU specifications have not been confirmed in available reporting.
Does this affect Intel and AMD’s stock or revenue?
No specific financial impact has been disclosed. The deal is significant mainly as a reputational and competitive signal rather than a confirmed revenue shift, since Modal has not disclosed contract value or deployment volume.
