--- title: Volantis wants to pipe AI memory over light, at 10,000... description: Volantis, a startup backed by Sam Altman, Jeff Dean, and John Doerr, claims optical waveguides built into chip packaging can break the memory bandwidth... canonical: https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Volantis wants to pipe AI memory over light, at 10,000 tokens/sec | daily.dev og:description: Volantis, a startup backed by Sam Altman, Jeff Dean, and John Doerr, claims optical waveguides built into chip packaging can break the memory bandwidth... og:url: https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat og:image: https://api.daily.dev/og/posts/jMzGR2WaT.png og:image:alt: Volantis wants to pipe AI memory over light, at 10,000 tokens/sec og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Volantis wants to pipe AI memory over light, at 10,000 tokens/sec **[Trends](https://daily.dev/sources/trends)** · 3 min read · 1 upvotes · 0 comments ## Summary Volantis, a startup backed by Sam Altman, Jeff Dean, and John Doerr, claims optical waveguides built into chip packaging can break the memory bandwidth bottleneck limiting AI inference speed. The company targets up to 10,000 tokens per second per user on models above 10T parameters, claiming sub-picojoule-per-bit energy, under 5ns latency, and over 200TB/s bandwidth by linking more than 220 memory chips optically instead of with copper. Funding is reported as either an $88M Series A or $97M total raised. All performance figures are the company's own unverified claims, with commentators framing faster inference as a potential unlock for coding agents that could turn hour-long runs into minutes. ## Content The loudest argument for Volantis isn't the tech. It's the cap table. @kloss_xyz: "Jeff Dean, Sholto Douglas, Dwarkesh and Naveen Rao just angeled into the same optics chip startup. These are the people who know their shit. Nobody this close to frontier AI angels a BS hype round." His thesis: memory is the next AI bottleneck, and GPUs "spend most of their time waiting on it." ## The setup Volantis announced an $88M Series A co-led by Lachy Groom and Abstract, with John Doerr, Susa Ventures, VXI Capital and Triatomic Capital also in. @rohanpaul_ai puts total funding at $97M and lists Sam Altman among the backers. The company builds optical links inside the chip package so accelerators can reach big pools of faster, cheaper memory. ## The pitch is speed for coding agents @omarsar0 calls faster inference "one of the next big unlocks for coding agents." The target is up to 10,000 tokens per second per user on models above 10T parameters. "That's crazy!" he writes, adding that agents taking hours today could finish in minutes. Rohan Paul's version is sharper: 30 minutes down to 30 seconds. @testingcatalog notes the company expects memory bandwidth and capacity per chip to rise several times over current designs. ## Why optics The physics argument is simple. A single HBM stack is about 11mm long, longer than copper can reach, which is why Nvidia's best GPUs top out at 8 stacks. Volantis says light can wire more than 220 memory chips into one shared pool. Its claimed numbers: - well under 1 picojoule per bit - under 5ns latency - over 200TB/s total bandwidth The waveguides sit inside the package rather than running over fiber. Each is said to be over 2500x smaller than a fiber. Capacity and speed scale together, instead of forcing the SRAM-versus-HBM tradeoff. ## The geopolitical angle Another writeup flags a supply-chain bet. Volantis uses gallium arsenide lasers instead of indium phosphide, where China supplies about 70%. Founders are Tapa Ghosh and Roy Meade. First customer deliveries are planned for 2027, the same year Europe opens a EUR 153M indium phosphide pilot line in Eindhoven. Lightmatter, Ayar Labs, Lumilens and copper-signalling Kandou are chasing other routes. So the bet is on two things: that memory is the real wall, and that these investors know where the wall is. The numbers are still the company's own. Silicon is due in 2027. ## Questions this post answers ### what is Volantis claiming about using optics to speed up AI inference Volantis claims optical waveguides built directly into chip packaging can link more than 220 memory chips into a shared pool, targeting up to 10,000 tokens per second per user on models above 10 trillion parameters. The company cites under 1 picojoule per bit, under 5ns latency, and over 200TB/s total bandwidth, replacing copper interconnects that are limited by HBM stack length. _daily.dev surfaces emerging hardware bets like this for engineers tracking what could unlock faster coding agents._ ### how much funding has Volantis raised for its optical memory chip technology Reports vary: one figure cited is $97M raised in total, while the company's own announcement describes an $88M Series A round. The likely explanation is that $88M is the latest round and $97M represents cumulative funding, though that connection is an inference rather than a confirmed fact. Investors include Sam Altman, Jeff Dean, and John Doerr. _those weighing hardware startup bets can follow funding details like these on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 3 discussions and 2 comments across x (as of 2026-10-02). **TL;DR:** There's very limited discussion available, but the few replies seen frame optical memory interconnects as a potentially major piece of the AI stack if memory truly is the bottleneck, while one commenter argues speed isn't the real fix and pitches a different approach to memory. **Sentiment:** 45% positive · 45% mixed · 10% skeptical **The case for** - If memory bandwidth is indeed the bottleneck, an optics layer could become a critical part of the AI stack. **The pushback** - One commenter argues that making memory faster doesn't address the deeper need for memory that 'remembers better,' implying raw speed may be the wrong focus. **By community** - x (mixed): A couple of substantive replies show cautious interest in optics as a stack-defining technology alongside skepticism that speed alone solves memory problems. **Highlights** > @kloss_xyz That seems like they’re just making memory faster. But doesn’t have to be faster if it remembers better. Everybody’s fixing the current system’s problems. I’m building a new kind of memory (among other things), not just improving the current state. Guna be needing testers soon > — [BeerMagician on x](https://x.com/BeerMagician/status/2105877914499822078) > @kloss_xyz If memory is the bottleneck, the optics layer could become a huge piece of the AI stack > — [banishahr66 on x](https://x.com/banishahr66/status/2105920560140288246) **Source threads** - [x](https://x.com/svpino/status/2105985768397385867) · 0 points · 0 comments - [x](https://x.com/kloss_xyz/status/2105833978032955552) · 1 points · 2 comments - [x](https://x.com/testingcatalog/status/2105980651379782074) · 0 points · 0 comments ## Similar posts on daily.dev - [Enlightra Raises $15M to Advance Laser Interconnects for AI and Quantum Systems](https://daily.dev/posts/enlightra-raises-15m-to-advance-laser-interconnects-for-ai-and-quantum-systems-ogty3or33) · The Quantum Insider · 0 upvotes · 0 comments - [CScale exits stealth with $145m to stop laser failures stalling AI](https://daily.dev/posts/cscale-exits-stealth-with-145m-to-stop-laser-failures-stalling-ai-nqdtdqmug) · The Next Web · 0 upvotes · 0 comments - [SpaceX vets raise $50M Series A for data center links](https://daily.dev/posts/spacex-vets-raise-50m-series-a-for-data-center-links-weq7j4iow) · TechCrunch · 0 upvotes · 0 comments - [AI chip startups soak up $1.1B in VC funding this week](https://daily.dev/posts/ai-chip-startups-soak-up-1-1b-in-vc-funding-this-week-sbcfggett) · The Register · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#ai-inference](https://daily.dev/tags/ai-inference) [View this post on daily.dev](https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Volantis wants to pipe AI memory over light, at 10,000 tokens/sec","url":"https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat"},"datePublished":"2026-10-02T01:22:27.265Z","dateModified":"2026-10-02T16:22:24.505Z","description":"Volantis, a startup backed by Sam Altman, Jeff Dean, and John Doerr, claims optical waveguides built into chip packaging can break the memory bandwidth...","image":"https://pbs.twimg.com/media/HTlmk7AagAAt-JN.png","thumbnailUrl":"https://pbs.twimg.com/media/HTlmk7AagAAt-JN.png","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,ai-inference","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Volantis wants to pipe AI memory over light, at 10,000 tokens/sec"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/volantis-wants-to-pipe-ai-memory-over-light-at-10-000-tokens-sec-jmzgr2wat#faq","mainEntity":[{"@type":"Question","name":"what is Volantis claiming about using optics to speed up AI inference","acceptedAnswer":{"@type":"Answer","text":"Volantis claims optical waveguides built directly into chip packaging can link more than 220 memory chips into a shared pool, targeting up to 10,000 tokens per second per user on models above 10 trillion parameters. The company cites under 1 picojoule per bit, under 5ns latency, and over 200TB/s total bandwidth, replacing copper interconnects that are limited by HBM stack length. daily.dev surfaces emerging hardware bets like this for engineers tracking what could unlock faster coding agents."}},{"@type":"Question","name":"how much funding has Volantis raised for its optical memory chip technology","acceptedAnswer":{"@type":"Answer","text":"Reports vary: one figure cited is $97M raised in total, while the company's own announcement describes an $88M Series A round. The likely explanation is that $88M is the latest round and $97M represents cumulative funding, though that connection is an inference rather than a confirmed fact. Investors include Sam Altman, Jeff Dean, and John Doerr. those weighing hardware startup bets can follow funding details like these on daily.dev."}}]} ```