Hyperbolic growth playbook

Write a unique title and description for every site page: /inference ("Flux Inference API: open models from $0.06 per 1M tokens"), /marketplace ("Rent H100, H200 and B200 GPUs from $3.19/hr"), /about and /blog. Add /pricing as a short page that summarises both price lists and links to each. Add Product and FAQ structured data to both pages, and check Search Console a week later for the queries each page starts to win.

Search engines and AI assistants pick a page by its title and description. Pages that say exactly what they price get shown for price searches, which are the highest-intent searches a GPU cloud gets.

[ ] / : Open-Access GPU & AI Cloud (keep)
[ ] /inference : Flux Inference API, open models from $0.06 per 1M tokens
[ ] /marketplace : Rent H100, H200 and B200 GPUs from $3.19/hr
[ ] /pricing : GPU and inference pricing (new page)
[ ] /about and /blog : their own titles
[ ] Product + FAQ structured data on both pricing pages
[ ] Search Console check after 7 days

• Prices change weekly

  Write "from" prices in titles and update them with the weekly refresh.

• Short-term ranking wobble

  Change titles once, then leave them for a month.

• Your websiteThe /inference page lists 10 models with per-token prices (for example DeepSeek V4 Flash 0731 at $0.06 input and $0.18 output per 1M tokens, Kimi K3 at $3.00 and $15.00) and explains Flux, the OpenAI-compatible inference API. Its HTML title and meta description are the homepage's word for word ("Hyperbolic — Open-Access GPU & AI Cloud"; "Get on-demand H100, H200 & B200 GPUs and reserve dedicated multi-node clusters…"), and it has no structured data. Checked October 22, 2026.Source
• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source
• Your websiteRecent launches on the blog: Flux inference ("Next-Generation Inference Powered by Multi-Cloud Orchestration", October 21, 2026), Managed Kubernetes for GPU workloads "Now in Beta" (October 13, 2026), Forge (June 10, 2026), on-demand reservations up to 30 days (February 2026) and dedicated model hosting (January 2026).Source

Render the headline as plain text and animate it with CSS only after paint; reserve space for logo strips, testimonials and the cookie banner; defer HubSpot chat and tag-manager scripts until after interaction; lazy-load the testimonial images. Re-test in PageSpeed and watch the 28-day field data.

A developer who taps "Get Started" and waits a second, or misses a button that moved, is the sign-up the marketing hire is paid to win.

• Deferring tags loses attribution

  Keep the page-view tag; defer only chat and heavy tags.

• PageSpeed InsightsHomepage, real Chrome visitors over 28 days: fails Core Web Vitals on interaction (INP 1,226 ms; good is under 200 ms) and layout shift (CLS 0.27; good is under 0.1); load (LCP 2.2 s) passes. Lab test: mobile performance 41/100 with a 17.5 s LCP, desktop 55/100; the LCP element is the animated, letter-by-letter "for Global Compute" headline. Accessibility 90, SEO 100.Source
• Your websiteSite stack: Next.js on Vercel, HubSpot (forms and chat) and Google Tag Manager; "Schedule a Call" goes to a Calendly sales-demo page; sign-up and log-in live on app.hyperbolic.ai. No newsletter sign-up on the site.Source
• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Interview Higgsfield, write a one-page case study (capacity ramp, timeline, the switch to reserved, savings), add a chart, and link it from the GPU page next to the Reserved tier. Use it in sales follow-ups and as the first post of a refreshed case-study section.

Buyers of reserved capacity want proof that the price and the ramp are real; a named customer with numbers does that better than any claim.

• Customer will not share numbers

  Publish percentages only, as in the quote.

• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Ask three questions in the booking form (GPU type and count, on-demand or reserved, start date), push each booking into HubSpot with the source page, and match it to the console account when one exists. Send the prep note to the rep before every call.

Reps who know the workload before the call quote faster, and clean records let marketing see which pages create pipeline.

• Longer form lowers bookings

  Keep it to three optional questions.

• HiringGTM (Operations) (posted September 3, 2026): "Today one person holds together everything that makes a deal actually close: contract tracking, quote prep, CRM hygiene, customer follow-up, support routing, invoicing coordination, meeting prep." The hire owns CRM data quality in HubSpot ("Deals, contacts, companies, stages, close dates") and "nothing sits for three days".Source
• Your websiteSite stack: Next.js on Vercel, HubSpot (forms and chat) and Google Tag Manager; "Schedule a Call" goes to a Calendly sales-demo page; sign-up and log-in live on app.hyperbolic.ai. No newsletter sign-up on the site.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Build a /supply page: what Hyperbolic offers suppliers (demand, Forge orchestration, billing), the hardware it wants, a simple utilisation-to-revenue estimate, and a short form (GPU type, count, location, interconnect) routed to the supply team in HubSpot.

Structured supplier leads are faster to diligence, and the page itself becomes a channel the supply team can share at events.

• Estimate seen as a promise

  Show ranges and state the assumptions.

• HiringCompute GTM, a Supply Specialist (remote, posted August 6, 2026): "grow and strengthen the supply side of our GPU marketplace": source GPU capacity "across NeoClouds, data center operators, cloud infrastructure providers", negotiate pricing, reserved capacity and SLAs, and "Represent Hyperbolic at industry events, conferences". A Head of Compute is also open.Source
• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Competitor researchRivals' own homepages: Lambda, "AI compute in the cloud" with on-demand instances, 1-Click Clusters and a Compare section; Runpod, "The AI Developer Cloud", on-demand GPUs "across 31 global regions" plus serverless endpoints; Together AI, "The AI Native Cloud" for open-model inference and training; Vast.ai, "Rent GPUs" with a hosting programme and earnings calculator for GPU suppliers; CoreWeave, "The Essential Cloud for AI".See the research Source
• HiringMember of Technical Staff - Inference (posted September 9, 2026): build inference on Forge, "our unified control plane, so customers can consume model tokens without managing GPUs and our NeoCloud partners get a full-stack path to their own token-factory offering"; deployment across clusters "distributed around the world, on heterogeneous hardware", then optimization, autoscaling and "customer inference debugging".Source

Define five stages (visit, sign-up, first credit, first GPU hour or first 1M tokens, repeat spend) and build them in Hex from console and billing data, split by source page and product. Review it every week with marketing and GTM.

A shared funnel shows where builders drop off and which pages bring paying users, not just sign-ups.

• Attribution gaps

  Pass UTM data from the site into sign-up.

• HiringSenior Data Engineer (posted October 5, 2026): "Hyperbolic's first dedicated Data Engineer", one half of a two-person data team with the data analyst; "Today our product engineers maintain the data stack on top of their day jobs"; "As we launch new products like inference, our data volume will grow fast"; stack: Redshift, Postgres, Hex, Polytomic.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source
• HiringGTM (Operations) (posted September 3, 2026): "Today one person holds together everything that makes a deal actually close: contract tracking, quote prep, CRM hygiene, customer follow-up, support routing, invoicing coordination, meeting prep." The hire owns CRM data quality in HubSpot ("Deals, contacts, companies, stages, close dates") and "nothing sits for three days".Source
• Your websiteRecent launches on the blog: Flux inference ("Next-Generation Inference Powered by Multi-Cloud Orchestration", October 21, 2026), Managed Kubernetes for GPU workloads "Now in Beta" (October 13, 2026), Forge (June 10, 2026), on-demand reservations up to 30 days (February 2026) and dedicated model hosting (January 2026).Source

Generate one page per model from the catalogue: prices (input, output, cached), context length, an OpenAI-SDK code sample with the Hyperbolic base URL, and a sign-up button that lands in the playground with that model selected. Update pages automatically when prices or models change.

Per-model pages match how developers search and give AI assistants a citable price for each model.

• Thin pages

  Add a benchmark note and a use case per model.

• Your websiteThe /inference page lists 10 models with per-token prices (for example DeepSeek V4 Flash 0731 at $0.06 input and $0.18 output per 1M tokens, Kimi K3 at $3.00 and $15.00) and explains Flux, the OpenAI-compatible inference API. Its HTML title and meta description are the homepage's word for word ("Hyperbolic — Open-Access GPU & AI Cloud"; "Get on-demand H100, H200 & B200 GPUs and reserve dedicated multi-node clusters…"), and it has no structured data. Checked October 22, 2026.Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringMember of Technical Staff - Inference (posted September 9, 2026): build inference on Forge, "our unified control plane, so customers can consume model tokens without managing GPUs and our NeoCloud partners get a full-stack path to their own token-factory offering"; deployment across clusters "distributed around the world, on heterogeneous hardware", then optimization, autoscaling and "customer inference debugging".Source
• Competitor researchRivals' own homepages: Lambda, "AI compute in the cloud" with on-demand instances, 1-Click Clusters and a Compare section; Runpod, "The AI Developer Cloud", on-demand GPUs "across 31 global regions" plus serverless endpoints; Together AI, "The AI Native Cloud" for open-model inference and training; Vast.ai, "Rent GPUs" with a hosting programme and earnings calculator for GPU suppliers; CoreWeave, "The Essential Cloud for AI".See the research Source

Publish a public index page and chart of weekly starting and median prices per GPU type (on-demand and reserved), with a short note on what moved. Offer it as an email and an embeddable chart.

A recurring data series earns links, press quotes and AI citations, and every visit lands on a page one click from renting.

• Suppliers object to price exposure

  Publish aggregates only, never one supplier's rate.

• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringCompute GTM, a Supply Specialist (remote, posted August 6, 2026): "grow and strengthen the supply side of our GPU marketplace": source GPU capacity "across NeoClouds, data center operators, cloud infrastructure providers", negotiate pricing, reserved capacity and SLAs, and "Represent Hyperbolic at industry events, conferences". A Head of Compute is also open.Source
• Competitor researchRivals' own homepages: Lambda, "AI compute in the cloud" with on-demand instances, 1-Click Clusters and a Compare section; Runpod, "The AI Developer Cloud", on-demand GPUs "across 31 global regions" plus serverless endpoints; Together AI, "The AI Native Cloud" for open-model inference and training; Vast.ai, "Rent GPUs" with a hosting programme and earnings calculator for GPU suppliers; CoreWeave, "The Essential Cloud for AI".See the research Source

Write one page per rival: GPU types, starting prices, reserved terms, quotas, inference catalogue, Kubernetes, billing for failed instances. Date every table, cite sources, and end with a free test path.

Comparison searches are late-stage and high-intent, and fair tables are what AI assistants quote.

• Prices go stale

  Date the tables and refresh monthly.

• Competitor researchRivals' own homepages: Lambda, "AI compute in the cloud" with on-demand instances, 1-Click Clusters and a Compare section; Runpod, "The AI Developer Cloud", on-demand GPUs "across 31 global regions" plus serverless endpoints; Together AI, "The AI Native Cloud" for open-model inference and training; Vast.ai, "Rent GPUs" with a hosting programme and earnings calculator for GPU suppliers; CoreWeave, "The Essential Cloud for AI".See the research Source
• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Flag accounts with steady on-demand spend for three or more weeks, send an in-console and email offer with an instant reserved quote for their actual usage, and route replies to GTM.

Moving steady users to reserved locks in revenue and lowers their bill before a rival offers it.

• Cannibalises on-demand margin

  Offer only above a usage threshold that supply can back.

• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• HiringGTM (Operations) (posted September 3, 2026): "Today one person holds together everything that makes a deal actually close: contract tracking, quote prep, CRM hygiene, customer follow-up, support routing, invoicing coordination, meeting prep." The hire owns CRM data quality in HubSpot ("Deals, contacts, companies, stages, close dates") and "nothing sits for three days".Source
• HiringSenior Data Engineer (posted October 5, 2026): "Hyperbolic's first dedicated Data Engineer", one half of a two-person data team with the data analyst; "Today our product engineers maintain the data stack on top of their day jobs"; "As we launch new products like inference, our data volume will grow fast"; stack: Redshift, Postgres, Hex, Polytomic.Source

For each runbook, publish a cleaned-up public guide in the docs ("Fix NFS mount timeouts on a Hyperbolic GPU instance"), link it from the ticket reply, and track repeat tickets per topic.

Public guides cut tickets, bring in developers searching errors, and show the platform is well supported.

• Exposing internal setup

  Review every guide before publishing.

• HiringTechnical Support Engineer (remote, posted August 28, 2026): owns tickets "against our SLA", runs and writes runbooks ("When you solve something new, write the runbook so the next person does not escalate it") and keeps "customer-facing docs and internal knowledge base current. Most repeat tickets are a documentation gap." Typical issues: SSH keys, NFS mounts, quotas, security groups, drivers, billing.Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Invite 15 teams as design partners with credits, a shared channel and a weekly call; publish two partner stories and a "from kubeconfig to training in minutes" video when the beta closes.

Partners shape the product and become the launch proof.

• Partners churn after credits

  Agree a reserved path before the credits end.

• Your websiteRecent launches on the blog: Flux inference ("Next-Generation Inference Powered by Multi-Cloud Orchestration", October 21, 2026), Managed Kubernetes for GPU workloads "Now in Beta" (October 13, 2026), Forge (June 10, 2026), on-demand reservations up to 30 days (February 2026) and dedicated model hosting (January 2026).Source
• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Give each programme a page in the main navigation, put the referral link in the console and invoices, and send partners a quarterly statement. Recruit ten partners from ML consultancies and university labs.

Referrals from developers and consultants convert better than paid ads in developer infrastructure, and the programmes already exist.

• Low-quality referrals

  Pay on spend, not sign-ups.

• Your websiteTwo growth programmes exist only as blog posts: a referral programme ("Share Hyperbolic. Get Paid.", May 2025) and the Hyperbolic Partner Program, a "1% commission fee for successful deals in GPU reservations" (August 2025). Neither is linked from the homepage navigation or footer.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source
• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source

Turn six testimonials into short case studies by product line (on-demand, reserved, inference, private cloud), each with the workload, timeline, numbers and a quote. Add a case-study section to the navigation and filter by use case.

Enterprise and platform buyers look for a peer with the same workload before booking a call.

• Slow approvals

  Start with customers who already gave quotes.

• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Set a target to serve major open models within 48 hours of release, keep a public "models we're adding" list, and let developers sign up for an alert by model. Announce each launch on X, the blog and the model page.

Being first with a new model wins the developers who try it that week, and many stay.

• Capacity for spikes

  Cap free-tier rate limits in launch week.

• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Your websiteThe /inference page lists 10 models with per-token prices (for example DeepSeek V4 Flash 0731 at $0.06 input and $0.18 output per 1M tokens, Kimi K3 at $3.00 and $15.00) and explains Flux, the OpenAI-compatible inference API. Its HTML title and meta description are the homepage's word for word ("Hyperbolic — Open-Access GPU & AI Cloud"; "Get on-demand H100, H200 & B200 GPUs and reserve dedicated multi-node clusters…"), and it has no structured data. Checked October 22, 2026.Source
• HiringMember of Technical Staff - Inference (posted September 9, 2026): build inference on Forge, "our unified control plane, so customers can consume model tokens without managing GPUs and our NeoCloud partners get a full-stack path to their own token-factory offering"; deployment across clusters "distributed around the world, on heterogeneous hardware", then optimization, autoscaling and "customer inference debugging".Source
• Your websiteRecent launches on the blog: Flux inference ("Next-Generation Inference Powered by Multi-Cloud Orchestration", October 21, 2026), Managed Kubernetes for GPU workloads "Now in Beta" (October 13, 2026), Forge (June 10, 2026), on-demand reservations up to 30 days (February 2026) and dedicated model hosting (January 2026).Source

Build a calculator: model size, tokens or dataset size, target time; it returns GPU type, count, hours and the on-demand and reserved cost on Hyperbolic, with a button to launch or request a quote with the answer filled in.

A planning tool is shared, bookmarked and cited, and hands the sales team a sized lead.

• Estimates seen as quotes

  Show ranges and assumptions.

• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• HiringGTM (Operations) (posted September 3, 2026): "Today one person holds together everything that makes a deal actually close: contract tracking, quote prep, CRM hygiene, customer follow-up, support routing, invoicing coordination, meeting prep." The hire owns CRM data quality in HubSpot ("Deals, contacts, companies, stages, close dates") and "nothing sits for three days".Source

Package it: white-label inference on the partner's own hardware, revenue share or licence, onboarding in weeks. Sign three launch partners, then publish a product page and a partner story.

It deepens supply relationships, earns on hardware Hyperbolic does not own, and gives neoclouds a reason to choose Hyperbolic over selling bare capacity.

• Partners compete with Hyperbolic's own inference

  Split by region or customer segment in the contract.

• HiringMember of Technical Staff - Inference (posted September 9, 2026): build inference on Forge, "our unified control plane, so customers can consume model tokens without managing GPUs and our NeoCloud partners get a full-stack path to their own token-factory offering"; deployment across clusters "distributed around the world, on heterogeneous hardware", then optimization, autoscaling and "customer inference debugging".Source
• HiringCompute GTM, a Supply Specialist (remote, posted August 6, 2026): "grow and strengthen the supply side of our GPU marketplace": source GPU capacity "across NeoClouds, data center operators, cloud infrastructure providers", negotiate pricing, reserved capacity and SLAs, and "Represent Hyperbolic at industry events, conferences". A Head of Compute is also open.Source
• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Competitor researchRivals' own homepages: Lambda, "AI compute in the cloud" with on-demand instances, 1-Click Clusters and a Compare section; Runpod, "The AI Developer Cloud", on-demand GPUs "across 31 global regions" plus serverless endpoints; Together AI, "The AI Native Cloud" for open-model inference and training; Vast.ai, "Rent GPUs" with a hosting programme and earnings calculator for GPU suppliers; CoreWeave, "The Essential Cloud for AI".See the research Source

Publish monthly aggregate figures: launch success rate, time to ready, failure rate and refunds, by GPU type. Show a reliability tier next to each offer in the console.

Proof of reliability removes the main objection to a marketplace and turns the no-charge policy into a measurable promise.

• Bad months are public

  Publish trends and the fixes made.

• Your websiteThe GPU rental page (/marketplace) shows H100 SXM from $3.19, H200 from $3.99 and B200 from $5.99 per GPU hour, "refreshed weekly based on the best available rates from suppliers", three tiers (On-Demand, Reserved under one year, Private Cloud over one year) and "Get notified within 3 minutes if an instance fails. No charges for failed instances". It also carries the homepage's title and description word for word, as do /about and /blog.Source
• HiringTechnical Support Engineer (remote, posted August 28, 2026): owns tickets "against our SLA", runs and writes runbooks ("When you solve something new, write the runbook so the next person does not escalate it") and keeps "customer-facing docs and internal knowledge base current. Most repeat tickets are a documentation gap." Typical issues: SSH keys, NFS mounts, quotas, security groups, drivers, billing.Source
• HiringCompute GTM, a Supply Specialist (remote, posted August 6, 2026): "grow and strengthen the supply side of our GPU marketplace": source GPU capacity "across NeoClouds, data center operators, cloud infrastructure providers", negotiate pricing, reserved capacity and SLAs, and "Represent Hyperbolic at industry events, conferences". A Head of Compute is also open.Source
• HiringSenior Data Engineer (posted October 5, 2026): "Hyperbolic's first dedicated Data Engineer", one half of a two-person data team with the data analyst; "Today our product engineers maintain the data stack on top of their day jobs"; "As we launch new products like inference, our data volume will grow fast"; stack: Redshift, Postgres, Hex, Polytomic.Source

Offer quarterly compute grants to university labs (application, review, credits), ask for acknowledgement in papers and a short write-up, and publish a yearly list of papers run on Hyperbolic.

Grants build the "open access" brand with the people who choose tools for the next generation of startups.

• Credits without return

  Require write-ups and cap credits per lab.

• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Your websiteFunding: a $12M round led by Variant and Polychain Capital closed in December 2024, bringing total funding to $20M "to Advance GPU Marketplace and Serverless Inference". Advisors include Databricks co-founder Reynold Xin and Prof. Yi Ma.Source
• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Publish one deep technical post a month from the team (KV-cache orchestration, heterogeneous clusters, failure detection), signed by the engineer, with a link to open roles and a free-credit link.

The same post attracts candidates and developer users, which halves the cost of each.

• Engineers too busy

  Interview them and ghost-write drafts for review.

• HiringHead of Talent (posted October 5, 2026): growth "from ~30 to ~70-100 employees in the next 6-12 months", tooling to "run 60+ hires per year", and "Build Hyperbolic's employer brand". Sixteen roles are open, eight of them in engineering (careers board, October 22, 2026).Source
• HiringMember of Technical Staff - Inference (posted September 9, 2026): build inference on Forge, "our unified control plane, so customers can consume model tokens without managing GPUs and our NeoCloud partners get a full-stack path to their own token-factory offering"; deployment across clusters "distributed around the world, on heterogeneous hardware", then optimization, autoscaling and "customer inference debugging".Source
• Your websiteRecent launches on the blog: Flux inference ("Next-Generation Inference Powered by Multi-Cloud Orchestration", October 21, 2026), Managed Kubernetes for GPU workloads "Now in Beta" (October 13, 2026), Forge (June 10, 2026), on-demand reservations up to 30 days (February 2026) and dedicated model hosting (January 2026).Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source

Send a monthly email to all users and a sign-up form on the blog: launches, new models, price moves and one customer story. Track reactivated accounts.

Dormant users who come back cost nothing to acquire.

• Unsubscribes

  Keep it monthly and short.

• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• Your websiteRecent launches on the blog: Flux inference ("Next-Generation Inference Powered by Multi-Cloud Orchestration", October 21, 2026), Managed Kubernetes for GPU workloads "Now in Beta" (October 13, 2026), Forge (June 10, 2026), on-demand reservations up to 30 days (February 2026) and dedicated model hosting (January 2026).Source
• Your websiteHomepage: "Liquidity AI Infra for Global Compute"; "Trusted by 250,000+ builders"; four products (On-Demand GPU Compute, Inference, Managed Kubernetes, Private Cloud); audience blocks for startups, researchers, enterprise platform teams, neoclouds and developer platforms; the neocloud block's "Grow Your GPU Cloud" button is a mailto link to one person's inbox; "Zero Quota Limit for GPU rentals"; 14 testimonials, including Higgsfield, converted from on-demand to "a reserved contract at a rate that was 35% cheaper than on-demand and 20% cheaper than anything else we were seeing in the market".Source
• Your websiteSite stack: Next.js on Vercel, HubSpot (forms and chat) and Google Tag Manager; "Schedule a Call" goes to a Calendly sales-demo page; sign-up and log-in live on app.hyperbolic.ai. No newsletter sign-up on the site.Source

Redirect both to their live pages, remove the links, and add a check in the deploy process that draft pages are not public.

Clean analytics make the new marketing function's numbers trustworthy from day one.

• Someone uses the drafts

  Move drafts to a preview deployment.

• Your websiteSitemap (checked October 22, 2026): eight site pages (home, about, inference, marketplace, blog, privacy, privacy FAQ, terms) and 99 blog posts. No pricing page (/pricing returns 404), no per-GPU or per-model pages, no comparison pages and no case-study section; the only case studies are two blog posts from 2025 (Reve AI, "300 Billion Tokens", and an LMSYS research write-up). Blog posts do have their own titles, e.g. "NVIDIA H200 Price in 2026" and "How Much Compute Do You Actually Need?". The monthly recap series stopped in May 2025.Source
• Your websiteSite stack: Next.js on Vercel, HubSpot (forms and chat) and Google Tag Manager; "Schedule a Call" goes to a Calendly sales-demo page; sign-up and log-in live on app.hyperbolic.ai. No newsletter sign-up on the site.Source
• HiringHead of Marketing (San Francisco, posted September 9, 2026): "brand marketing and PR strategies that resonate with developers and enterprise leaders"; a "highly quantitative mindset focused on metrics, conversion rates, and understanding how users sign up and engage with products"; "translate complex infrastructure into clear and credible narratives"; "building a marketing function from the ground up"; preferred: "build and manage a diverse ecosystem of partners, contributors, and community advocates" and "proactive channel growth tactics and creative experimentation".Source