{"slug": "nvidia-wins-modal-cpu-deal-claims-2x-ai-agent-speed-2026", "title": "Nvidia Wins Modal CPU Deal, Claims 2x AI Agent Speed [2026]", "summary": "Modal will deploy Nvidia's new Arm-compatible Vera CPU alongside its existing Intel and AMD fleet, the cloud-computing startup confirmed in a blog post titled \"NVIDIA Vera CPUs double throughput for Modal Sandboxes,\" after Nvidia hand-delivered its first Vera CPU server to Modal's San Francisco office last week. Modal reported that initial sandbox tests showed Vera roughly doubling the amount of AI agent work completed versus its prior x86 hardware under heavy concurrent load, a vendor-reported early result not yet replicated by independent benchmarks. Modal chief executive Erik Bernhardsson said the company is not abandoning its existing chip suppliers, giving Nvidia a third seat in a server CPU market Intel and AMD have controlled for decades.", "body_md": "Nvidia has picked up central processor business from Modal, a cloud-computing startup that rents out computing power to AI developers and until now built its entire fleet on chips from Intel and AMD. [Reuters first reported the deal on October 19, 2026](https://www.fidelity.com/news/article/company-news/202610191010RTRSNEWSCOMBINED_L6N465081_1), and it was quickly picked up by [The Economic Times](https://economictimes.indiatimes.com/tech/technology/nvidia-wins-fast-growing-ai-startup-modal-for-new-central-processors/articleshow/135051728.cms?from=mdr) and other financial outlets, marking one of the clearest public signals yet that Nvidia’s multi-year push into the server CPU market is gaining real customers rather than just lab demos.\n\nModal confirmed the move in a company blog post, saying it plans to deploy Nvidia’s new Vera central processor to help host AI agents, the increasingly popular class of autonomous software that can plan, browse, and execute multi-step tasks without constant human prompting. The announcement lands at a moment when demand for agent-hosting infrastructure is climbing fast, and it puts a new, credible data point behind Nvidia’s attempt to crack a CPU market that Intel and AMD have controlled for decades.\n\nDon't miss new tech stories on Google\n\nAdd Shattered once in the Google app and our stories show up in your news suggestions.\n\n[Add now](https://www.google.com/preferences/source?hl=en&q=https%3A%2F%2Fshattered.io%2F)\n\n## What Modal Actually Announced About Nvidia’s Vera CPU\n\nAccording to [Modal’s own blog post, titled “NVIDIA Vera CPUs double throughput for Modal Sandboxes”](https://modal.com/blog/nvidia-vera-sandboxes), the company said Nvidia hand-delivered its first Vera CPU server to Modal’s San Francisco office last week. Modal described Vera as Nvidia’s new Arm-compatible CPU, built primarily for the Vera Rubin GPU platform but also sold as a standalone CPU server that does not require an attached GPU. That distinction matters: it means Nvidia is positioning Vera not just as a companion chip bundled with its GPUs, but as a product that can compete head-to-head with general-purpose Intel and AMD server CPUs on its own merits.\n\nModal’s chief executive, Erik Bernhardsson, was quoted in reporting as saying the company is not abandoning its existing chip suppliers. Instead, Modal says it will run Nvidia CPUs alongside its established Intel and AMD fleet in what the company calls a mixed hardware approach, assigning each workload to whichever processor handles that specific job most efficiently. That is a meaningfully different story than “Nvidia wins Modal and replaces Intel and AMD” — it’s closer to “Nvidia earns a third seat at a table that used to have two.”\n\nThe throughput numbers reported alongside the announcement are notable but still framed as early-stage results. Modal said its initial sandbox tests, the isolated compute environments it spins up to run AI agent workloads safely, showed Vera roughly doubling the amount of agent work completed compared with the company’s prior x86 hardware under heavy concurrent load. No independent, third-party benchmark has yet replicated that figure, and Modal has not published a full methodology breakdown, so the 2x number should be read as a vendor-reported early test result rather than an audited industry benchmark.\n\n## Why This Matters: Nvidia’s Long Road Into Server CPUs\n\nNvidia built its trillion-dollar valuation almost entirely on graphics processors for AI training and inference. The CPU market is a different battlefield, one where Intel’s x86 architecture has dominated data centers since the 1990s and AMD has spent the past decade clawing back share with its EPYC server line. Nvidia’s answer has been Grace, and now Vera, both Arm-based chips designed to pair naturally with Nvidia’s own GPUs inside a single rack, cutting the latency and power overhead that comes from mixing an Nvidia GPU with an Intel or AMD CPU built on a different interconnect philosophy.\n\nWhat makes the Modal deal significant is not the size of the contract, which has not been disclosed, but the type of customer. Modal is not a hyperscaler buying chips for its own massive internal workloads the way Google or Amazon might. It is an independent, fast-growing AI infrastructure startup making an open-market purchasing decision about which CPU delivers the best price-to-performance for a very specific, modern workload: hosting autonomous AI agents at scale. That is precisely the kind of third-party validation Nvidia needs if it wants Wall Street and enterprise buyers to see Vera as a genuine Intel and AMD alternative rather than an in-house curiosity bundled only with Nvidia’s own GPU racks.\n\nReaders tracking Nvidia’s broader hardware roadmap have already seen the company test this strategy elsewhere. Our earlier coverage of the [HPE Vera CPU server launch](https://shattered.io/hpe-vera-cpu-server-1-2tb-bandwidth-rivals-2026/) detailed how Nvidia’s Arm-based chip entered the market promising 1.2TB/s of memory bandwidth against three established rivals. The Modal deal is the first widely reported case of a cloud-native AI company choosing Vera on its own, rather than through an OEM server partner like HPE packaging it into a pre-built system.\n\n## Modal’s Business and Why It Matters as a Customer\n\nModal positions itself as infrastructure for AI developers, letting engineering teams rent serverless compute to run everything from model training jobs to the sandboxed environments that let AI agents execute code, browse the web, and interact with external tools safely. Before this announcement, reporting indicates Modal’s fleet ran exclusively on Intel and AMD processors, a common setup across the cloud industry given how mature the software tooling, driver support, and operational playbooks are for x86 hardware.\n\nThe shift to a mixed fleet that includes Nvidia Vera reflects a broader trend across AI infrastructure providers: workload-specific hardware selection rather than a single standardized chip across the entire data center. As agent workloads multiply, they expose CPU bottlenecks around sandbox spin-up time, concurrent process isolation, and memory bandwidth that look different from the bottlenecks found in traditional web serving or batch analytics. If Nvidia’s Arm-based Vera architecture genuinely handles those specific patterns better, even by a meaningful fraction rather than a full 2x, that is a real technical edge independent of Nvidia’s dominant GPU market position.\n\n## Market Impact: What a $1.2 Trillion-Plus Wager on CPUs Signals\n\nNvidia’s stock reaction to individual customer wins like this one tends to be modest in isolation, since Modal’s compute footprint is a rounding error next to Nvidia’s roughly $5 trillion annual GPU revenue run rate from hyperscale customers. But investors and analysts watch these smaller wins as leading indicators. Reuters framed the Modal deal explicitly as “an early sign that its push into a market long controlled by those two rivals has traction,” referring to Nvidia’s broader ambition against Intel and AMD in CPUs. Financial research site [GuruFocus](https://www.gurufocus.com/news/9129981/modal-adopts-nvidias-vera-cpu-highlighting-nvdas-expanding-ai-infrastructure-role) echoed that read, framing the Modal deployment as evidence of Nvidia’s expanding footprint across AI infrastructure beyond its core GPU business.\n\nThat framing lines up with a rockier few weeks for chip stocks more broadly. Our recent report on how [chip stocks erased roughly $1 trillion in combined value](https://shattered.io/chip-stocks-slump-1-trillion-value-erased-2026/) showed how sensitive the sector has become to any signal about AI infrastructure demand, whether that signal points up or down. A CPU design win, even a modest one, gives Nvidia bulls a concrete data point to point to amid that volatility, while skeptics will note that one mid-sized startup switching part of its fleet is not the same as unseating Intel or AMD across the broader data center market.\n\nAMD, for its part, has been gaining ground against Intel in traditional x86 server and desktop CPU share, as detailed in our coverage of [AMD’s record 30.7% CPU market share and Intel’s 31-year low](https://shattered.io/amd-cpu-market-share-record-2026-intel-31-year-low/). Nvidia entering that fight as a third competitor, armed with an Arm-based architecture rather than x86, changes the shape of that rivalry rather than simply adding a third x86 player.\n\n## Nvidia Vera vs. Intel and AMD: A Side-by-Side Look\n\nBecause Nvidia has not published a full public datasheet with every architectural detail, and Modal’s own disclosures remain limited to the blog post and the figures cited in press coverage, any comparison table has to be read as directional rather than exhaustive. The table below pulls together what has actually been confirmed in reporting and Modal’s own statements, alongside the general market position of Intel and AMD’s competing server CPU lines.\n\n| Factor | Nvidia Vera (at Modal) | Intel (Modal’s prior fleet) | AMD (Modal’s prior fleet) | \n|---|---|---|---|\n| Architecture | Arm-compatible, built for Vera Rubin GPU platform | x86 | x86 | \n| Deployment at Modal | New addition, first Vera server delivered to Modal’s SF office | Existing, long-standing supplier | Existing, long-standing supplier | \n| Reported throughput on agent sandboxes | Up to roughly 2x in Modal’s early internal tests | Baseline reference point | Baseline reference point | \n| Standalone CPU-only server option | Confirmed available, not tied to a GPU purchase | Yes, standard | Yes, standard | \n| Modal’s stated hardware strategy | Workload-matched placement across a mixed fleet | Workload-matched placement across a mixed fleet | Workload-matched placement across a mixed fleet | \n| Market history with Modal | First-time deployment, announced October 19, 2026 | Multi-year incumbent | Multi-year incumbent | \n\nThe key takeaway from that comparison is that Nvidia is not winning this fight by replacing Intel and AMD outright. It is winning by convincing a sophisticated infrastructure buyer to add a third option and then route specific workloads toward it. That is a much lower bar than a full migration, but it is also a necessary first step for any new CPU vendor trying to break into a market this entrenched. Our earlier look at [Nvidia reallocating GB202 chip supply toward its RTX Pro line](https://shattered.io/nvidia-gb202-shifts-rtx-pro-5090-supply-2026/) showed a similar pattern of Nvidia carefully managing where its newest silicon lands first, prioritizing high-value commercial and enterprise customers before broader availability.\n\n## The AI Agent Hosting Boom Driving the Decision\n\nModal’s framing of this announcement around AI agents rather than general cloud compute is itself a signal of where infrastructure demand is concentrating heading into 2027. Autonomous agents, systems that can take a goal, break it into steps, and execute code or browse the web to complete them, place different demands on a CPU than a typical web server or database workload. They spin up and tear down sandboxed environments rapidly, often running many short-lived, isolated processes concurrently rather than a handful of long-running ones.\n\nThat workload shape is exactly where Modal reported its roughly 2x throughput gain with Vera. If that result holds up under broader, independently verified testing, it would suggest Nvidia’s Arm-based CPU architecture has a genuine structural advantage for agent-hosting specifically, separate from Nvidia’s unrelated and already-dominant position in the GPUs that actually run the AI models those agents call. Other recent agent-infrastructure moves across the industry, including expanded agent support inside major cloud platforms, point to the same underlying trend: demand for agent-hosting capacity is growing faster than demand for traditional compute, and vendors across the stack are racing to optimize for it.\n\n## Historical Context: Nvidia’s Grace-to-Vera CPU Strategy\n\nNvidia’s CPU ambitions did not start with Vera. The company’s Grace CPU, introduced years earlier as an Arm-based chip meant to pair with Nvidia’s Hopper and Blackwell GPU generations, laid the architectural groundwork for this current push. Grace was largely positioned as a companion chip bundled into Nvidia’s own GPU superchip designs, like Grace Hopper and Grace Blackwell, rather than sold widely as an independent server CPU competing directly for general-purpose workloads.\n\nVera represents a continuation and broadening of that strategy, built specifically for the newer Vera Rubin GPU platform but, according to Modal’s own description, also sold as a standalone CPU server that does not require an Nvidia GPU attached. That standalone availability is the detail that makes the Modal story different from a typical Grace Hopper superchip sale. It means a customer like Modal, which already owns plenty of Intel and AMD CPU capacity and was not necessarily shopping for more Nvidia GPUs at that exact moment, could still choose Vera purely on CPU merits.\n\nThat mirrors a pattern seen across Nvidia’s broader memory and compute roadmap. Our coverage of [Micron’s HBM4 memory roadmap](https://shattered.io/micron-hbm4-memory-roadmap-2026/) showed how closely Nvidia’s silicon partners are timing their own product cycles to Nvidia’s GPU and CPU release schedule, with HBM4 promising roughly 2.6x bandwidth gains timed to arrive alongside next-generation Nvidia platforms like Vera Rubin. The CPU and memory ecosystem around Nvidia’s AI hardware is being built in lockstep, not as isolated product launches.\n\n## Competitive Landscape: Where Intel and AMD Stand Now\n\nIntel and AMD are not standing still while Nvidia tests the CPU waters. Intel has continued investing in its Xeon server line, and AMD has pushed its EPYC line aggressively on both price and core count, which is part of why AMD has posted record CPU market share gains against Intel in recent reporting. Our earlier piece on [Intel’s stock dropping 2.6% as Nvidia’s RTX Spark drew attention away from Xeon 6](https://shattered.io/intel-drops-2-6-percent-nvidia-rtx-spark-eclipses-xeon-6-2026/) illustrates how sensitive investor sentiment around Intel has become to any sign that Nvidia is encroaching on territory Intel considers its own, even when the actual revenue impact of a single deal like Modal’s is small in absolute terms.\n\nThe honest read on competitive positioning right now is that Intel and AMD still control the overwhelming majority of data center CPU shipments worldwide, and nothing in the Modal announcement changes that near-term reality. What has changed is the narrative: Nvidia now has a named, fast-growing AI infrastructure customer willing to publicly say it tested Vera against its incumbent x86 fleet and liked what it saw enough to deploy it in production, even as part of a mixed setup rather than a full replacement.\n\n## What Analysts and Outlets Are Saying\n\nReuters’ original wire report, syndicated through financial outlets including Fidelity’s news desk, described the deal plainly: Nvidia has won central processor chip business with another fast-growing AI startup that until now relied entirely on chips from Intel and AMD, calling it an early sign that Nvidia’s push into a market long controlled by those two rivals has traction. The Economic Times picked up the same wire report under the headline “Nvidia wins fast-growing AI startup Modal for new central processors,” underscoring how quickly the story spread across financial and tech media on the day of the announcement.\n\nIndustry newsletter [AI Weekly characterized the news](https://aiweekly.co/alerts/modal-adds-nvidia-vera-cpu-to-host-ai-agents-claims-2x-throughput) in similarly measured terms, noting that Modal added Nvidia’s Vera CPU to the mix of chips it uses to host AI agents after early tests showed it running up to twice the throughput of comparable x86 hardware under heavy concurrent load, while being careful to frame that figure as an early test result rather than an industry-wide benchmark. That consistent framing across independent outlets, an early positive signal rather than a finished verdict, is worth holding onto as more detailed, third-party benchmarks of Vera against Intel and AMD’s latest server chips become available in the months ahead.\n\n## What Comes Next: Predictions for Nvidia’s CPU Push\n\nSeveral things seem likely to play out over the coming months based on the pattern this deal fits into. First, expect Nvidia to publicize additional CPU design wins with other AI-native infrastructure startups over the next two to three quarters, since Modal is unlikely to be the only cloud provider testing Vera against its existing Intel and AMD fleet, and Nvidia has clear incentive to keep building a list of credible reference customers.\n\nSecond, independent benchmarking firms and rival cloud providers will likely push to verify or challenge Modal’s roughly 2x throughput claim once more Vera hardware reaches customers outside Nvidia’s initial rollout partners, since a figure that large, if confirmed broadly, would represent a meaningful architectural edge rather than a marginal improvement.\n\nThird, Intel and AMD will almost certainly respond with their own agent-optimized server CPU messaging, framing their existing or upcoming x86 chips as better suited to concurrent, short-lived agent workloads, rather than ceding that narrative to Nvidia uncontested.\n\nFourth, expect continued scrutiny of exactly how much of Nvidia’s CPU business depends on being bundled with its own GPU platforms versus genuinely standing on its own, since the Modal deal is notable precisely because it demonstrates standalone CPU demand rather than a bundled GPU-plus-CPU rack sale.\n\nFifth, the mixed-fleet approach Modal describes, routing specific workloads to whichever chip performs best, is likely to become the default strategy across AI infrastructure providers more broadly, rather than any single company standardizing entirely on one CPU vendor, since the economics of chip procurement increasingly reward flexibility over uniformity.\n\n## Timeline of Nvidia’s CPU Market Push\n\nThe table below places the Modal announcement in the context of Nvidia’s broader CPU strategy as reported across recent coverage, helping readers see how this latest deal fits a pattern rather than standing as an isolated event.\n\n| Milestone | What Happened | Significance | \n|---|---|---|\n| Nvidia Grace CPU era | Arm-based CPU paired with Hopper and Blackwell GPUs in bundled superchip designs | Established Nvidia’s Arm-based CPU architecture inside its own GPU racks | \n| Nvidia Vera Rubin platform | Next-generation Nvidia GPU platform paired with the new Vera CPU | Vera designed as both a GPU companion chip and a standalone server CPU | \n| HPE Vera CPU server launch | HPE packaged Nvidia’s Vera CPU into its own server line | First major OEM to productize Vera outside Nvidia’s own hardware | \n| Modal deployment, October 19, 2026 | Modal adds Vera to its mixed Intel/AMD/Nvidia fleet for AI agent hosting | First widely reported independent AI startup choosing Vera on its own | \n| Next phase (ongoing) | Additional customer wins and independent benchmarking expected | Will determine whether Vera becomes a durable third CPU option | \n\n## Why Enterprise Buyers Should Pay Attention\n\nFor engineering teams evaluating infrastructure vendors, the Modal-Nvidia deal is a useful signal even for organizations with no direct plans to buy Vera hardware themselves. It suggests that CPU choice for AI-specific workloads, particularly agent hosting, is becoming an active decision rather than a default one. Teams that have simply assumed x86 is the only sensible choice for their compute layer may want to start asking their cloud providers whether Arm-based alternatives like Vera are on the roadmap, and under what workload conditions they might actually outperform the status quo.\n\nIt is also a reminder that infrastructure decisions at AI-native companies are increasingly workload-driven rather than vendor-loyalty-driven. Modal’s own description of its strategy, assigning jobs to whichever hardware runs them best rather than standardizing on one chip family, reflects a level of operational sophistication that smaller or more traditional IT shops may not yet have the tooling or expertise to replicate, but it is clearly where the leading edge of AI infrastructure management is heading.\n\n## The Bigger Picture for Nvidia’s Business\n\nNvidia’s GPU business remains, by an enormous margin, the company’s primary revenue driver, and nothing about the Modal CPU deal changes that fundamental fact. But Nvidia has been explicit for several years that it sees the CPU as a strategic complement to its GPU dominance, not a side project. Grace and now Vera exist, in part, to prevent rivals from ever building a compelling enough CPU-plus-accelerator package that could pull customers away from Nvidia’s GPUs on the strength of better CPU integration alone.\n\nSeen that way, winning Modal’s CPU business is less about the CPU revenue itself, which is almost certainly a tiny fraction of what Nvidia earns from GPU sales to the same customer, and more about deepening Modal’s overall dependence on Nvidia’s broader hardware and software stack. Every additional Nvidia chip inside a customer’s data center, CPU included, makes it marginally harder for that customer to ever fully migrate away from Nvidia’s ecosystem, a dynamic that has served the company well throughout its GPU-driven rise.\n\n## Frequently Asked Questions\n\n**What exactly did Modal announce about Nvidia?**\n\nModal said it will begin using Nvidia’s new Vera central processor to help host AI agents, running it alongside its existing Intel and AMD processors in a mixed hardware fleet rather than replacing those suppliers outright.\n\n**Is Nvidia acquiring or buying Modal?**\n\nNo. Despite some headlines using the word “wins,” the confirmed reporting describes a CPU supplier relationship, meaning Modal is purchasing and deploying Nvidia’s Vera chips as a customer. There is no confirmed acquisition or ownership change involved.\n\n**What is Nvidia’s Vera CPU?**\n\nVera is Nvidia’s newer Arm-compatible central processor, built primarily for the company’s upcoming Vera Rubin GPU platform but also available as a standalone CPU server that does not require an attached Nvidia GPU.\n\n**How much faster is Vera than Intel or AMD chips, according to Modal?**\n\nModal reported that early internal tests showed Vera roughly doubling the amount of AI agent work completed compared with its prior x86 hardware under heavy concurrent load. That figure comes from Modal’s own early testing and has not yet been independently verified by third-party benchmarks.\n\n**Is Modal dropping Intel and AMD entirely?**\n\nNo. Modal’s chief executive, Erik Bernhardsson, said the company is not abandoning its existing suppliers. Modal plans to run Nvidia processors alongside Intel and AMD hardware, assigning each workload to whichever chip performs best for that specific task.\n\n**Why does this matter for Nvidia’s stock and the broader chip market?**\n\nThe deal is being read as an early sign that Nvidia’s push into the server CPU market, long dominated by Intel and AMD, is gaining real traction with independent customers rather than only appearing in Nvidia’s own bundled GPU hardware. It comes amid a period of heightened volatility across chip stocks generally.\n\n**What is Modal’s business?**\n\nModal is a cloud-computing startup that rents out computing power to AI developers, including the infrastructure used to host AI agents, the autonomous software systems capable of carrying out multi-step tasks with limited human supervision.\n\n**Will other AI infrastructure companies follow Modal’s lead?**\n\nNo contract or specific commitment from other companies has been confirmed, but industry watchers expect Nvidia to pursue and publicize additional similar deals with other fast-growing AI infrastructure providers as it continues building out Vera’s customer base.", "url": "https://wpnews.pro/news/nvidia-wins-modal-cpu-deal-claims-2x-ai-agent-speed-2026", "canonical_source": "https://shattered.io/nvidia-vera-cpu-modal-ai-agent-deal-2026/", "published_at": "2026-10-19 21:31:02+00:00", "updated_at": "2026-10-19 21:47:14.936150+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips", "ai-agents"], "entities": ["Nvidia", "Modal", "Vera", "Erik Bernhardsson", "Intel", "AMD", "Grace", "Vera Rubin"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/nvidia-wins-modal-cpu-deal-claims-2x-ai-agent-speed-2026", "markdown": "https://wpnews.pro/news/nvidia-wins-modal-cpu-deal-claims-2x-ai-agent-speed-2026.md", "text": "https://wpnews.pro/news/nvidia-wins-modal-cpu-deal-claims-2x-ai-agent-speed-2026.txt", "jsonld": "https://wpnews.pro/news/nvidia-wins-modal-cpu-deal-claims-2x-ai-agent-speed-2026.jsonld"}}