Modal, a cloud-computing startup that rents out computing power to AI developers, said on 19 October 2026 that it plans to start using Nvidia’s new Vera central processor for hosting AI agents, Reuters reported. Reuters says Modal has until now run such work only on Intel and AMD chips.
In its own post the same day, Modal said that 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.”
Business Pill · Latency vs Throughput
A short explainer on latency versus throughput. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, Modal’s case for Vera rests on contention rather than raw speed: its own post says Vera was not far ahead one task at a time, and pulled ahead when hundreds of workloads ran at once.
Modal’s engineering blog says that last week Nvidia “hand-delivered Modal’s first Vera CPU server to our SF office.” It describes Vera as “NVIDIA’s new Arm-compatible CPU, built for the Vera Rubin GPU platform but also available as a standalone CPU server.”
Supporting Arm “would be a huge change for Modal, coming from a long history exclusively running x86 machines,” the post says. Reuters reports that the move “will require some deep software changes on Modal’s side.”
The post ends: “We’re very excited to roll out support for Vera CPUs to power Modal Sandboxes, and work closely with NVIDIA and cloud providers to scale up our fleet.” Neither Modal nor Reuters gives a date, a server count or a price.
Nvidia’s CPU push in its own numbers, with Modal’s result: up to 2x agent throughput versus comparable x86 hardware in Modal’s early tests (19 October 2026); 1.8x faster on agentic tasks on the spec benchmark, five times the bandwidth per watt, more than $5 billion of trailing Grace CPU revenue, about $20 billion of server CPU demand and CPU revenue expected to more than double in fiscal 2028, all from Nvidia’s 26 August 2026 earnings call.
What Was Posted on X
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.
For Modal Sandboxes, that could mean running more agents in parallel without sacrificing latency.
Modal (@modal) posted at 14:03 UTC on 19 October 2026, two minutes after its blog post: “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.” This piece’s headline figure comes from that post and the blog it links to; it is Modal’s own benchmark.
Today we’re announcing that Modal has crossed $1B in annualized run rate, as well as raised a $750M Series D at a $15.75B valuation, led by @Accel.
Since our Series C, usage has tripled and weโve grown more than 15x year-over-year.
We are deeply grateful to the developers and companies building on Modal. There’s still so much more to build, and we can’t think of a better time to be building software.
Modal’s chief executive Erik Bernhardsson posted on 14 October 2026, five days earlier, that Modal “has crossed $1B in annualized run rate” and raised a $750M Series D at a $15.75B valuation led by Accel. This piece cites his run-rate, usage and growth figures from that post.
How Modal Tested Vera
Modal says it built a benchmark called Agentless Agent Bench: “a collection of 50 Terminal-Bench oracle solutions and verifiers, stripped of all network calls and prewarmed in concurrent Docker containers.” The aim, per the post, was to isolate the CPU work agents and reinforcement-learning runs actually do, such as compilation, file grepping and git operations.
Run one task at a time on an idle machine, the results were close, Modal says. The post compares Vera with Intel GNR, AWS Graviton 4, AWS Graviton 5 and AMD Zen 5 chips, and says Vera “often performed better than the other CPUs, but not always, and not by a particularly large margin.”
Under load the picture changed, Modal says. Its early 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.”
Modal’s explanation is memory bandwidth: Vera’s “high memory bandwidth allows for significantly more container workloads to pack per core, without CPU throttling.”
Why the CPU Matters for Agents
Modal’s post argues that agent sandboxes are spiky. A sandbox “could spend most of its uptime relatively inactive, waiting on an LLM, with sudden and unpredictable CPU surges 10x above baseline.”
The post names two uses. For background agents, it cites Ramp Inspect and agents running on Niteshift, and says faster agents mean “lower CPU time, driving down COGS.” For reinforcement learning, it says “there are GPUs sitting idle while sandboxed processes run,” so “Any time saved on CPU is time saved on GPU.”
Intel and AMD Stay in the Fleet
Reuters reports that Chief Executive Erik Bernhardsson “was careful to stress it is not abandoning its existing suppliers.” The company said it will run the Nvidia chips alongside its Intel and AMD offerings in a mixed fleet, “putting each job on whatever hardware runs it best,” according to Reuters.
Bernhardsson said the tests were run against the Intel and AMD chips already on the market, Reuters reports. Reuters adds that both rivals are developing newer models. “We’ll test them when they’re available,” he said, adding that a mixed setup means the company “can always deploy whatever hardware performs best.”
Nvidia’s CPU Push
Reuters describes the win as “an early sign that its push into a market long controlled by those two rivals has traction.” It reports that Nvidia has told investors it expects to generate “at least $20 billion from Vera chip sales this fiscal year,” and that Nvidia has previously said OpenAI, Anthropic and Oracle also plan to use the chips, along with AI startups such as Perplexity.
On Nvidia’s 26 August earnings call, chief financial officer Colette Kress said Grace CPU revenue on a trailing 12-month basis was “exceeding $5 billion” and that Vera “completes agentic tasks 1.8x faster on the spec benchmark and provides five times the bandwidth per watt than any other data center CPU.”
Kress named lead partners already receiving shipments: “OCI, SpaceXAI, and starting this quarter, AWS.” She said: “We continue to see demand for approximately $20 billion in total server CPUs,” and that Nvidia’s preliminary expectation is for CPU revenue “to more than double in fiscal 2028.”
Modal by the numbers. The round size, lead investor and valuation are as Reuters reports them; the Series D label, run rate, usage and growth figures are Erik Bernhardsson’s own, from his 14 October 2026 post; the 50 tasks are from Modal’s 19 October blog.
Modal’s Numbers
Modal raised $750 million last week in a round led by Accel at a $15.75 billion valuation, Reuters reports. In his own post on 14 October, Bernhardsson wrote that Modal “has crossed $1B in annualized run rate” and that since its Series C, “usage has tripled and we’ve grown more than 15x year-over-year.”
The Structural Read
As we read it, the CPU is becoming a cost line in agent economics: Modal’s post says sandboxes spend much of their time waiting on an LLM and then surge, and that for reinforcement learning “Any time saved on CPU is time saved on GPU.”
As we read it, Modal is keeping its options open: Reuters reports a mixed fleet with Intel and AMD, and Bernhardsson says newer rival chips will be tested when they are available.
As we read it, Modal is one data point in a larger Nvidia bet: Kress put trailing Grace CPU revenue above $5 billion and named OCI, SpaceXAI and AWS as lead Vera partners.
Erik Bernhardsson, Modal chief executive, to Reuters, 19 October 2026
“We’ll test them when they’re available”
Three Implications
A SECOND CHIP ARCHITECTURE Modal says supporting Arm would be a huge change after a long history exclusively running x86 machines; Reuters reports it will require deep software changes on Modal’s side.
A BENCHMARK UNDER LOAD Modal’s early runs show up to double the task throughput on Vera versus an equivalent x86 GNR machine, measured with its own Agentless Agent Bench.
NO EXCLUSIVITY Reuters reports Modal will run the Nvidia chips alongside its Intel and AMD offerings in a mixed fleet, putting each job on whatever hardware runs it best.
The framework puts it this way: “the conversational interface was scaffolding. The harness is the building.”
As we read it, Modal sells part of that building, the sandboxes where agents run between model calls, and its post argues the CPU underneath sets how many agents can run in parallel. The framework was published in July 2026 and does not discuss Modal, Nvidia or Vera.
What Is Not Established
The 2x figure is Modal’s own early benchmark, run on the Vera node Modal had access to, against chips already on the market; Bernhardsson told Reuters that newer Intel and AMD parts will be tested when they are available. No source gives the number of Vera servers Modal will deploy, what it pays or when Sandboxes users can choose Vera. Nvidia’s own comment on Modal does not appear in either report.
The two Nvidia revenue figures are worded differently: Reuters writes of “at least $20 billion from Vera chip sales this fiscal year,” while Kress, on the call, spoke of “demand for approximately $20 billion in total server CPUs.”
Business Engineer Framework
The Agentic Harness War
A Business Engineer framework on why value is moving to the layer where AI agents are run and delegated work.
Modal said on 19 October that it plans to run AI agent sandboxes on Nvidia’s Vera CPU after its own early tests showed up to 2x the throughput of comparable x86 hardware under load. Reuters reports that it will keep Intel and AMD chips in a mixed fleet. Nvidia’s finance chief said in August that it sees demand for about $20 billion in server CPUs.
A note on sourcing. This piece is based on Reuters’ report of 19 October 2026 (read in full via a syndicated copy), Modal’s own blog post and X post of the same day, Erik Bernhardsson’s X post of 14 October and the transcript of Nvidia’s 26 August 2026 earnings call. The 2x throughput figure is Modal’s own early benchmark. Nvidia’s comment on Modal does not appear in these sources. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
