Normal Computing growth playbook

Rewrite the title, H1 and meta description of the six main pages around what each sells (for example "Normal Computing | Thermodynamic AI inference hardware" and "Normal EDA | AI verification for complex IP and SoCs"), turn on the Webflow sitemap, add canonicals and Organization and Product structured data, and publish one page that names and explains Normal Inference Systems. Then submit the sitemap in Search Console and check what search and AI assistants return for "Normal Computing inference" and "Normal EDA".

Titles and H1s are the strongest signals search and AI answers use to understand a page. They are a few hours of Webflow work, and every story the new PR lead places will send searchers to pages that say what Normal does.

[ ] Title tag names the company and the product
[ ] One H1 per page
[ ] A distinct meta description per page
[ ] og:title matches the title
[ ] Canonical tag
[ ] Organization and Product structured data
[ ] /sitemap.xml live and submitted
[ ] "Normal Inference Systems" page published

• Names change before launch

  Keep page names editable and update them with each announcement.

• New copy overstates results

  Use only claims that Engineering has signed off as measured.

• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteSearch basics (checked October 22, 2026): the homepage title tag and link-preview title are both just "Home"; the main pages are titled "Research", "Blog", "ASICs" and "EDA" with no brand; none of the homepage, About, Research, Blog, ASICs or EDA pages has an H1 heading; About, Research, Blog, ASICs and EDA all share one meta description about the founders ("founded by former members of the Google Brain Team, Palantir, and X"); /sitemap.xml returns 404; there are no canonical tags or structured data. The name "Normal Inference Systems" used in the job posts appears on none of these pages. Blog posts do carry full titles such as "AI Inference Needs New Hardware - Normal Computing".Source
• Your websiteThe Normal EDA page (/solutions/eda) describes "the unified AI verification stack" with test plan and testbench generation, traceability to the spec, a seven-question FAQ (SOC 2 prioritised; integrates with VCS and Xcelium) and "Book a demo". The site menu lists only "Solutions - ASICS", and the homepage has no link to /solutions/eda, although its EDA section says "Normal EDA · In production".Source
• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source

Rename the menu item "Solutions" with both ASICs and EDA beneath it, add a "See Normal EDA" button next to "Read the research" on the homepage, and give the EDA page its own demo form that asks for the verification problem (IP type, simulator, team size).

Verification teams who arrive from a talk or article find the product and the demo in one click, which shortens the path for the enterprise sellers.

• More unqualified requests

  Qualifying questions on the form.

• Your websiteThe Normal EDA page (/solutions/eda) describes "the unified AI verification stack" with test plan and testbench generation, traceability to the spec, a seven-question FAQ (SOC 2 prioritised; integrates with VCS and Xcelium) and "Book a demo". The site menu lists only "Solutions - ASICS", and the homepage has no link to /solutions/eda, although its EDA section says "Normal EDA · In production".Source
• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• Your websiteContact paths on the homepage: "Let's talk" and "Partner with Us" open one Typeform, a second "Partner with Us" button opens another, "Talk to our engineers" a third and "Express interest" a fourth, on three different Typeform accounts; the footer email link goes to an address at normalcomputing.ai.Source

Replace them with one form whose first question is the reason (EDA demo, reference system or pilot, research partnership, press, careers), then route each answer to its owner with a response target. Keep the labels on the buttons, but pre-select the reason. Point the footer email to a normalcomputing.com address.

One form gives one view of demand by product and lets the PR lead see press requests the same day.

• Lost answers during the switch

  Keep the old forms live for a month and forward their answers.

• Your websiteContact paths on the homepage: "Let's talk" and "Partner with Us" open one Typeform, a second "Partner with Us" button opens another, "Talk to our engineers" a third and "Express interest" a fourth, on three different Typeform accounts; the footer email link goes to an address at normalcomputing.ai.Source
• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source

Pick one public claim per product, label each number as measured, simulated or target with a footnote to its source (paper, benchmark or CN101 results), and keep a claims register that every page, deck and press release uses.

Consistent, sourced numbers are what technical buyers and editors trust, and they protect the PR program from the first awkward question.

• Smaller number sounds weaker

  Pair it with the path to the larger target.

• Your websiteASICs page: "Reduce AI energy consumption by 1000x with thermodynamic computing" (the homepage says "a 10-100x gain"); "In June 2025, we taped out our first thermodynamic compute chip, CN101"; "Benchmarking will validate performance and guide the scaling of CN201 and CN301"; workloads listed from AI inference and diffusion models to Bayesian ML, molecular dynamics and sampling.Source
• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source

Add a pilot page: the workloads that qualify first (long-context decoding up to ~1M tokens), hardware and integration requirements, the evaluation steps with forward-deployed engineers, and an intake form that captures model, context length and current cost per token.

Clear criteria raise the quality of requests and let the deployment team rank them by impact, as the page promises.

• Too many requests for few systems

  Publish a waitlist and expected timing.

• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• Your websiteASICs page: "Reduce AI energy consumption by 1000x with thermodynamic computing" (the homepage says "a 10-100x gain"); "In June 2025, we taped out our first thermodynamic compute chip, CN101"; "Benchmarking will validate performance and guide the scaling of CN201 and CN301"; workloads listed from AI inference and diffusion models to Bayesian ML, molecular dynamics and sampling.Source
• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source

Show a poster image with a play button and load the Vimeo player only on click, defer non-critical scripts and the cookie banner, and preload the hero font. Re-run PageSpeed after each change.

A lighter page keeps the result for real visitors as traffic from campaigns grows, at little cost.

• Fewer video plays

  Track plays and keep the poster inviting.

• Google PageSpeedMobile lab test of the homepage: performance 80/100, largest content at 4.1 s, render-blocking requests costing about 2.9 s, 3.6 MB downloaded, of which Vimeo video is about 2.0 MB over 36 requests. Real visitors pass Core Web Vitals (field LCP 1.5 s, INP 120 ms, CLS 0).Source

Publish /press with the company boilerplate, a fact sheet (founded, funding, investors, offices, products, CN101 tape-out), approved images of the chip and team, spokespeople by topic (hardware, EDA, AI research), recent coverage and a press contact.

Reporters use the company that makes their job easy, and a press page gets cited in stories and AI answers.

• Facts go stale

  Date the fact sheet.

• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• Your websiteFunding post (March 25, 2026): "Normal Computing raises $50M from Samsung Catalyst to tackle soaring AI chip costs and power demands" (Fortune exclusive). The About page lists investors including Samsung Catalyst, Brevan Howard, ArcTern Ventures and Drive Capital, and offices in New York City, San Francisco, London, Copenhagen and Pangyo; the team page includes a Vice President, Global Strategic Growth & APAC.Source

Publish a glossary and six short explainers (what thermodynamic computing is, why noise helps, why attention in memory matters for long context, how a PCIe card fits a rack, how to measure intelligence per dollar per watt, CN101 to CN301). Use one diagram per piece and link each from the product pages.

The company that defines the terms gets quoted when the category is discussed, and clear explainers shorten sales cycles.

• Too technical for buyers

  Lead each piece with the business effect.

• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• Competitor researchRivals' own homepages: Extropic, "thermodynamic computing hardware that is radically more energy efficient than GPUs"; d-Matrix, "Memory-Centric Compute for AI Inference"; ChipAgents, "Agentic AI for semiconductor design and verification" with "customer results"; Cognichip, "physics-informed foundation models purpose-built for chip design".See the research Source
• Your websiteASICs page: "Reduce AI energy consumption by 1000x with thermodynamic computing" (the homepage says "a 10-100x gain"); "In June 2025, we taped out our first thermodynamic compute chip, CN101"; "Benchmarking will validate performance and guide the scaling of CN201 and CN301"; workloads listed from AI inference and diffusion models to Bayesian ML, molecular dynamics and sampling.Source

Publish a benchmarks page: methods, configurations and results for CN101 and Normal EDA, labelled measured or simulated, each with a notebook that reproduces the simulation side on thermox or public data.

Engineers trust what they can run, and reproducible results are hard for rivals to wave away.

• Revealing too much

  Publish only what customers and partners approve.

• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteASICs page: "Reduce AI energy consumption by 1000x with thermodynamic computing" (the homepage says "a 10-100x gain"); "In June 2025, we taped out our first thermodynamic compute chip, CN101"; "Benchmarking will validate performance and guide the scaling of CN201 and CN301"; workloads listed from AI inference and diffusion models to Bayesian ML, molecular dynamics and sampling.Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source

Pick six spokespeople by topic, list the chip design, verification and AI infrastructure conferences and podcasts their buyers follow, submit talks a cycle ahead, and turn each talk into a post, clips and a slide download.

Talks by engineers reach verification teams and infrastructure buyers directly and give the sales team reasons to follow up.

• Engineers lose build time

  Limit each to two talks a year.

• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source
• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source

Plan one joint story per partner (a DRAMBench update with results, a Samsung Catalyst piece on AI chip power costs, an ARIA programme progress note), with shared quotes and both partners' channels.

Partners double the reach and add third-party credibility to a new architecture.

• Slow partner approvals

  Start reviews early.

• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• Your websiteFunding post (March 25, 2026): "Normal Computing raises $50M from Samsung Catalyst to tackle soaring AI chip costs and power demands" (Fortune exclusive). The About page lists investors including Samsung Catalyst, Brevan Howard, ArcTern Ventures and Drive Capital, and offices in New York City, San Francisco, London, Copenhagen and Pangyo; the team page includes a Vice President, Global Strategic Growth & APAC.Source

Write one anonymised case study: the IP type, team size, simulator, before-and-after time to coverage, and what engineers still did by hand. Get the customer's approval to describe them by segment.

Verification leads buy on peer proof, and a method makes the 2x believable.

• Customer declines naming

  Use segment and size only.

• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• Your websiteThe Normal EDA page (/solutions/eda) describes "the unified AI verification stack" with test plan and testbench generation, traceability to the spec, a seven-question FAQ (SOC 2 prioritised; integrates with VCS and Xcelium) and "Book a demo". The site menu lists only "Solutions - ASICS", and the homepage has no link to /solutions/eda, although its EDA section says "Normal EDA · In production".Source
• Competitor researchRivals' own homepages: Extropic, "thermodynamic computing hardware that is radically more energy efficient than GPUs"; d-Matrix, "Memory-Centric Compute for AI Inference"; ChipAgents, "Agentic AI for semiconductor design and verification" with "customer results"; Cognichip, "physics-informed foundation models purpose-built for chip design".See the research Source

Edit an external version of the weekly brief each fortnight: three industry developments in AI inference, energy and chip design, with Normal's short view on each. Send it to the newsletter and post it on LinkedIn.

The work is already being done; publishing it builds an audience of engineers and buyers between announcements.

• Commenting on rivals

  Stick to public facts and Normal's own view.

• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• Public profilesThe GitHub organisation linked from the site shows a single public repository (its profile, 3 stars, last push January 2026); a Discord invite is in the footer next to a newsletter sign-up; the Verilog simulator post drew 3 points on Hacker News.Source

Collect the top 30 questions from sales and deployment calls, answer each on its own short page with FAQ structured data, and link the answers from the product pages and sales follow-ups.

Buyers ask search and AI assistants these questions before they ask a salesperson.

• Sharing roadmap details

  Answer only what is released.

• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteThe Normal EDA page (/solutions/eda) describes "the unified AI verification stack" with test plan and testbench generation, traceability to the spec, a seven-question FAQ (SOC 2 prioritised; integrates with VCS and Xcelium) and "Book a demo". The site menu lists only "Solutions - ASICS", and the homepage has no link to /solutions/eda, although its EDA section says "Normal EDA · In production".Source
• Your websiteSearch basics (checked October 22, 2026): the homepage title tag and link-preview title are both just "Home"; the main pages are titled "Research", "Blog", "ASICs" and "EDA" with no brand; none of the homepage, About, Research, Blog, ASICs or EDA pages has an H1 heading; About, Research, Blog, ASICs and EDA all share one meta description about the founders ("founded by former members of the Google Brain Team, Palantir, and X"); /sitemap.xml returns 404; there are no canonical tags or structured data. The name "Normal Inference Systems" used in the job posts appears on none of these pages. Blog posts do carry full titles such as "AI Inference Needs New Hardware - Normal Computing".Source

Publish one "what we are building" post per team (RTL, memory, compilers, physical design), linking to that team's roles, and share it through the team members' own networks.

Recruiting content raises reply rates to outreach and also tells buyers the team is real.

• Revealing unreleased work

  Review each post with Engineering.

• HiringThe Ashby board lists 37 open roles (checked October 22, 2026), 30 of them posted in October: silicon engineering (RTL Lead, Memory Architect, Physical Design Lead, Head of Silicon Product Engineering), compilers and runtime, four recruiting roles (Senior Hardware Recruiter, Senior Technical Recruiter, Recruiting Coordinator, Talent Operations Specialist), AI and thermodynamic hardware residencies, and a Foundry Sourcing & Supplier Manager to manage "our transition from prototype silicon to scaled manufacturing".Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• Your websiteFunding post (March 25, 2026): "Normal Computing raises $50M from Samsung Catalyst to tackle soaring AI chip costs and power demands" (Fortune exclusive). The About page lists investors including Samsung Catalyst, Brevan Howard, ArcTern Ventures and Drive Capital, and offices in New York City, San Francisco, London, Copenhagen and Pangyo; the team page includes a Vice President, Global Strategic Growth & APAC.Source

Pin or link the open-source projects from the organisation profile (wherever they now live), add a short README on what Normal maintains, and link each project from its blog post and the Research page.

GitHub is where technical audiences check credibility, and the work already exists.

• Unmaintained projects

  Mark their status clearly.

• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• Public profilesThe GitHub organisation linked from the site shows a single public repository (its profile, 3 stars, last push January 2026); a Discord invite is in the footer next to a newsletter sign-up; the Verilog simulator post drew 3 points on Hacker News.Source

Build a calculator: model size, context length, tokens per day and power price in; current cost and energy on GPUs versus a Normal reference system (labelled as target) out, with the method shown and a link to the pilot form.

A calculator turns an abstract architecture into a budget conversation and qualifies pilot requests.

• Projected numbers mistaken as measured

  Label outputs as targets.

• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteASICs page: "Reduce AI energy consumption by 1000x with thermodynamic computing" (the homepage says "a 10-100x gain"); "In June 2025, we taped out our first thermodynamic compute chip, CN101"; "Benchmarking will validate performance and guide the scaling of CN201 and CN301"; workloads listed from AI inference and diffusion models to Bayesian ML, molecular dynamics and sampling.Source

Host a few reference systems and offer time-boxed remote evaluations for qualified teams: run their long-context model, measure tokens per second and energy, and hand over a report. Convert the best results into on-premises pilots.

Lower friction brings more evaluations and better data for CN201 and CN301 decisions.

• Customer data security

  Isolated environments and clear terms.

• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source
• Your websiteASICs page: "Reduce AI energy consumption by 1000x with thermodynamic computing" (the homepage says "a 10-100x gain"); "In June 2025, we taped out our first thermodynamic compute chip, CN101"; "Benchmarking will validate performance and guide the scaling of CN201 and CN301"; workloads listed from AI inference and diffusion models to Bayesian ML, molecular dynamics and sampling.Source
• HiringThe Ashby board lists 37 open roles (checked October 22, 2026), 30 of them posted in October: silicon engineering (RTL Lead, Memory Architect, Physical Design Lead, Head of Silicon Product Engineering), compilers and runtime, four recruiting roles (Senior Hardware Recruiter, Senior Technical Recruiter, Recruiting Coordinator, Talent Operations Specialist), AI and thermodynamic hardware residencies, and a Foundry Sourcing & Supplier Manager to manage "our transition from prototype silicon to scaled manufacturing".Source

Publish a yearly report on cost and energy per unit of AI work: public data on data-centre power, GPU efficiency trends and long-context growth, plus Normal's measured results, with charts free to reuse.

Original data earns coverage, links and AI citations, and positions Normal as the reference on inference efficiency.

• Seen as advocacy

  Show sources and methods for every chart.

• HiringMedia & PR Lead (New York, posted October 20, 2026): "develop the public story of Normal Inference Systems" and bring "an unfamiliar computing architecture into the public conversation about AI infrastructure and energy" through "earned media, alternative media and speaking"; "Develop spokespeople" among "executives, engineers and researchers"; "Lead partner PR"; "Write media kits, briefings, Q&A and contributed pieces"; audiences include "technical media and broader technology, business and policy audiences".Source
• HiringTechnical Content Lead (New York, posted October 20, 2026, $145K-$205K): "explain the science and engineering behind Normal Inference Systems" and "help establish language for an unfamiliar architecture"; "Make answers findable. Turn recurring customer questions into useful, searchable technical explanations"; "Distinguish measured results from targets and projections"; build "diagrams, plots and, where useful, reproducible examples or notebooks"; send "weekly reports on industry trends and milestones"; reports to the Head of Communications and is embedded with Engineering.Source
• Your websiteHomepage: "A new paradigm of AI hardware to power human progress."; "In Production with the world's largest semiconductor companies"; "2x Faster verification on the most complex IP and SoCs"; "Reference Systems available, prioritizing long-context agentic reasoning"; "$85M+ funding raised. Strategic round led by Samsung Catalyst"; Normal ASICs "built toward a 10-100x gain in AI inference per dollar, per watt"; "PCIe accelerator cards in standard air-cooled servers"; attention "for up to ~1M context length at rack scale"; "Pilot deployments open. We prioritize by application and impact".Source

Offer a hosted sandbox on a public standard: users pick a section of a public spec and see the generated test plan with links to the source text, then book a demo for their own designs.

Seeing the output on a known spec is the strongest demo, and it creates leads without sales time.

• Licensing of the spec

  Use only openly published standards.

• Your websiteThe Normal EDA page (/solutions/eda) describes "the unified AI verification stack" with test plan and testbench generation, traceability to the spec, a seven-question FAQ (SOC 2 prioritised; integrates with VCS and Xcelium) and "Book a demo". The site menu lists only "Solutions - ASICS", and the homepage has no link to /solutions/eda, although its EDA section says "Normal EDA · In production".Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• HiringProduct Manager (New York, posted October 14, 2026): helps build "the AI systems and autoresearch platform our ASIC team needs to tapeout unconventional, physics-based architecture"; "use coding agents to prototype and land fixes yourself"; candidates should already use "Claude Code, Codex, Cursor, or something similar".Source

Turn each into a talk, a short video and a thread for developer communities, with code and lessons. Time posts for weekday mornings US time and invite engineers to the Discord.

Agentic engineering is a topic developers seek out, and it shows how Normal builds.

• Off-topic for buyers

  Treat it as recruiting and brand.

• HiringProduct Manager (New York, posted October 14, 2026): helps build "the AI systems and autoresearch platform our ASIC team needs to tapeout unconventional, physics-based architecture"; "use coding agents to prototype and land fixes yourself"; candidates should already use "Claude Code, Codex, Cursor, or something similar".Source
• Your websiteWriting and Research pages: recent posts "CN101 - A Digital Thermodynamic Computer for Generative AI" (August 10, 2026), "AI Inference Needs New Hardware" (August 4), "Thermodynamic Programming: What Happens When 10 Agents Share a Branch?" (July 9), DRAMBench "developed by Normal Computing and Fraunhofer IESE" (April 23) and "Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days" (March 2); press posts from Semiconductor Engineering, Tom's Hardware and EE Times Silicon 100, and "Normal Computing Selected for ARIA's £50M Scaling Compute Programme" (October 2024); open-source work "includes Posteriors and Outlines, with contributions to DSPy"; thermox, "The First Thermodynamic Computing Simulator".Source
• Public profilesThe GitHub organisation linked from the site shows a single public repository (its profile, 3 stars, last push January 2026); a Discord invite is in the footer next to a newsletter sign-up; the Verilog simulator post drew 3 points on Hacker News.Source

Publish Korean versions of the homepage, Normal EDA page and press fact sheet, and a Korean contact path to the APAC team. Use them in partner and event outreach in Korea.

Korean semiconductor teams can read and share the story internally, and the pages support APAC sales.

• Translations fall behind

  Translate only stable pages.

• Your websiteFunding post (March 25, 2026): "Normal Computing raises $50M from Samsung Catalyst to tackle soaring AI chip costs and power demands" (Fortune exclusive). The About page lists investors including Samsung Catalyst, Brevan Howard, ArcTern Ventures and Drive Capital, and offices in New York City, San Francisco, London, Copenhagen and Pangyo; the team page includes a Vice President, Global Strategic Growth & APAC.Source
• Your websiteThe Normal EDA page (/solutions/eda) describes "the unified AI verification stack" with test plan and testbench generation, traceability to the spec, a seven-question FAQ (SOC 2 prioritised; integrates with VCS and Xcelium) and "Book a demo". The site menu lists only "Solutions - ASICS", and the homepage has no link to /solutions/eda, although its EDA section says "Normal EDA · In production".Source