ProgrammableMarketing/farmgpu.com (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 10
  • Big bets · 3
  • Wildcards · 3
  • How we can help
Generated Nov 4, 2026
  1. Programmable
  2. Playbooks
  3. FarmGPU
FarmGPU logo

growth playbook · Nov 4, 2026

farmgpu.com (opens in a new tab)

FarmGPU growth playbook

Diagnosis

FarmGPU has rare proof for a 12-person cloud: #1 storage benchmarks, MLPerf records, an OCP leadership role and endorsements from SemiAnalysis and Runpod, and its new headline sells on price. The website does not yet carry buyers from that promise to a start: "Deploy now" opens a multi-year contract form, "Rent now" lands on Runpod's homepage, "Calculate savings" opens an investor model, and the H100 price appears three ways.

This playbook fixes those doors first, then adds a quickstart, a benchmark trial, case studies, comparison pages and a home for the operator business, and builds toward sovereign compute and an inference storage tier.

Growth Score

6/10

Biggest Opportunity

Turn price and benchmark leadership into self-serve starts by making the on-demand path, the savings number and the unit price clear on every page.

Biggest Risk

Runpod, Lambda, Nebius and the hyperscalers compete on price and self-serve, and depending on Runpod for on-demand means FarmGPU cannot see or own most of its top-of-funnel demand.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Make "Deploy now" deploy: it opens a 1-5 year contract form, while the page above it promises no commitments and no sales calls

    Change the "Rent now" links from runpod.io to a tracked link to FarmGPU's cluster listing.

    High impact First result in 2 weeks
  2. 2

    Give buyers a savings calculator; the "Calculate savings" button opens an investor deal desk that shows a -9% IRR

    Repoint the "Calculate savings" link to the pricing table until the buyer calculator is live.

    High impact First result in 1 month
  3. 3

    One H100 price, clearly per GPU: the site says ~$2.99, ~$3.29 and "$3.29 per hour" for an 8-GPU cluster

    Decide the current H100 per-GPU price and update the homepage.

    High impact First result in 1 week

02

Your idea portfolio

22 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 8 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 2Big bets · 11Fill-ins · 9Reconsider · 0
Effort →

Showing 22 of 22 ideas

This week 6

Quick wins with a clear first action.

  • 01

    The homepage sells on "per-second billing, sub-5-minute deployment, no commitments" and "no sales calls". But the main "Deploy now" button opens a form whose Term field offers only 1, 2, 3 or 5 years, asks for storage in petabytes, and has no on-demand option. The pricing page's four "Rent now" buttons go to Runpod's generic homepage, where a visitor has to find FarmGPU capacity alone. The developer who came for an H100 by the second hits a procurement form or loses the trail. Growth is the funnel, and this is its first door.

    High impactLow effort First result in 2 weeks Marketing Manager

    First action

    Change the "Rent now" links from runpod.io to a tracked link to FarmGPU's cluster listing.

    How to measure it

    Tracked on-demand starts

    Now
    –
    Target
    100 a month
    By
    60 days

    Steps

    1. 1

      Map the paths

      List every Deploy, Rent and Book button and where each lands today.

      Marketing Manager 1 day

    2. 2

      Agree the link with Runpod

      A referral or deep link to FarmGPU Instant Clusters.

      CBO and Co-Founder 1 week

    3. 3

      Build the on-demand page

      Three steps, per-GPU price, a short video.

      Marketing Manager 3 days

    4. 4

      Fix the form

      Add an on-demand term and make storage optional.

      Marketing Manager 1 hour

    5. 5

      Relabel the buttons

      "Start on demand" and "Reserve capacity" across all pages.

      Marketing Manager 1 day

    6. 6

      Tag the clicks

      Google Analytics events on both paths.

      Marketing Manager 1 day

    7. 7

      Review weekly

      Starts, forms and demos by path.

      Global Sales Manager weekly

    First result
    2 weeks
    Cost
    Staff time only
    Owner
    Marketing Manager

    Smallest test

    Point one "Rent now" button at a tracked FarmGPU cluster link for a week and count clicks.

    Why it works

    Split the button into two clear paths. "Start on demand" goes to a FarmGPU page that walks through launching a FarmGPU Instant Cluster on Runpod (a tracked link to the exact cluster page, a 3-step guide, the per-GPU price) so FarmGPU sees every click. "Reserve capacity" keeps today's form, with "On-demand / month-to-month" added to the term choices. Measure both paths separately.

    Buyers at the top of the funnel want to try first and talk later; the promise on the page should be one click away. Tracked hand-offs to Runpod also tell FarmGPU how much on-demand demand its marketing creates, which it cannot see today.

    Ready to use

    ChecklistTwo-door deploy checklist

    [ ] "Start on demand" page with per-GPU prices
    [ ] Tracked link to FarmGPU capacity on Runpod
    [ ] "Reserve capacity" form with an on-demand term option
    [ ] Storage field optional
    [ ] Same two buttons on home, pricing, platform, performance
    [ ] Analytics event per button
    [ ] Weekly report: starts, forms, demos

    Risks and how to handle them

    • Runpod cannot pass conversion data back

      Use a dedicated referral code and compare usage monthly.

    • Large buyers skip the form

      Keep "Reserve capacity" beside the on-demand path on every page.

    Evidence

    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
    • Your websiteThe "Deploy now" page ("Break ground.") is a form: name, work email, company, end user yes/no, GPU type (H100 to Vera Rubin), workload, backend network, storage type, "Storage Capacity (PB)", location, and a Term field whose only choices are "1 year", "2 years", "3 years" and "5 years". There is no on-demand choice and no link to start right away.Source
    • Your websitePricing page: the table says H100 SXM "Per GPU/hr ~$3.29" vs hyperscalers "~$7-13"; the cards read "H100 Cluster $3.29 per hour" (8x H100), "H200 Cluster $4.59 per hour", "B200 Cluster $6.79 per hour" and "B300 Cluster $7.89 per hour", each above an "8x" spec list. Every card's "Rent now" button links to www.runpod.io/, the partner's generic homepage. Reserved 1-5 year rates are quoted on request.Source
    • Your docsdocs.farmgpu.com (Mintlify) is company documentation: snapshot, thesis, hardware and software stack, data center specs, security, legal, team, culture, brand. Its llms.txt index lists no quickstart, API reference or "launch your first cluster" guide; Instant Clusters are launched "from the RunPod dashboard with a few clicks". Two citations on that page link to the blog with "?utm_source=chatgpt.com" appended.Source
  • 02

    The homepage card "70% Less Than Hyperscalers / Calculate savings" sends AI teams to calculator.farmgpu.com, a deal desk for pricing a 512-GPU offtake with debt share and cost of equity. Its default view reads "Project IRR -9.0%" and "Payback Never". A buyer expecting to see their savings sees a losing investment case instead.

    High impactMedium effort First result in 1 month Marketing Manager

    First action

    Repoint the "Calculate savings" link to the pricing table until the buyer calculator is live.

    How to measure it

    Calculator-to-lead rate

    Now
    –
    Target
    10%
    By
    60 days

    Steps

    1. 1

      Repoint the link

      Homepage card to /pricing for now.

      Marketing Manager 1 hour

    2. 2

      Agree the inputs

      GPU type, count, hours, current provider, egress.

      Global Sales Manager 2 days

    3. 3

      Set comparison prices

      Public list prices with dates.

      Marketing Manager 2 days

    4. 4

      Build it

      Webflow page or a small app on the calculator host.

      Head of Engineering 1 week

    5. 5

      Add the hand-off

      Email the result and offer reserved pricing.

      Marketing Manager 2 days

    6. 6

      Re-home the deal desk

      Under a GPU investor page.

      CBO and Co-Founder 1 day

    7. 7

      Measure

      Uses, leads and reserved quotes.

      Marketing Manager monthly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Marketing Manager

    Smallest test

    Repoint the link today and watch whether pricing-page leads rise.

    Why it works

    Build a one-screen buyer calculator: GPU type, GPUs, hours a month, current provider (AWS, GCP, Azure, Runpod, Lambda) and egress, showing FarmGPU on-demand and reserved costs and the yearly saving. Link it from the homepage card and pricing. Move the deal desk under a "For GPU investors" page with a sensible default case.

    A savings number in the buyer's own terms is the strongest form fill trigger for a price-led headline, and the investor tool then reaches the audience it was built for.

    Risks and how to handle them

    • Competitor prices change

      Show the date and source of each price.

    Evidence

    • Your websiteThe homepage "Lowest TCO / 70% Less Than Hyperscalers / Calculate savings" card links to calculator.farmgpu.com, which opens a "Deal Desk" to "Price an offtake for 512 x H100 SXM": server price, debt share, cost of equity, colocation per kW. Its default view shows "Project IRR -9.0%", "Equity multiple 0.44x" and "Payback Never". It is a GPU-investor model, not a buyer savings calculator.Source
    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
    • Your websitePricing page: the table says H100 SXM "Per GPU/hr ~$3.29" vs hyperscalers "~$7-13"; the cards read "H100 Cluster $3.29 per hour" (8x H100), "H200 Cluster $4.59 per hour", "B200 Cluster $6.79 per hour" and "B300 Cluster $7.89 per hour", each above an "8x" spec list. Every card's "Rent now" button links to www.runpod.io/, the partner's generic homepage. Reserved 1-5 year rates are quoted on request.Source
    • Your docsCompany snapshot: Farm GPU Inc., founded 2024, Rancho Cordova, CA; data centers in Rancho Cordova and Santa Clara; "~12 (FTE + Strategic Contractors)". Four offerings: on-demand compute; managed Slurm/Kubernetes clusters; "Turnkey GPU Investment: Bring your own capex, have FarmGPU manage end to end ... to monetization on our partner platforms"; and an "AI Data Center Blueprint" for traditional facility operators. Products page: managed infrastructure for "Data center operators with unused GPU capacity" and "Investors or consortiums deploying GPU capital".Source
  • 03

    The homepage shows the H100 at "~$2.99", the pricing table "~$3.29" per GPU-hour, and the pricing cards "H100 Cluster $3.29 per hour" over an "8x H100" spec list, which reads as $3.29 for eight GPUs. Runpod lists H100 SXM at $3.99 per GPU-hour. Price-led buyers compare unit prices to the cent; three numbers and an unclear unit make the cheapest offer look uncertain.

    High impactLow effort First result in 1 week Marketing Manager

    First action

    Decide the current H100 per-GPU price and update the homepage.

    How to measure it

    Pricing-page conversions

    Now
    –
    Target
    +15%
    By
    60 days

    Steps

    1. 1

      Confirm prices

      Per-GPU on-demand rates for all four GPUs.

      CRO and Co-Founder 1 day

    2. 2

      Relabel cards

      "per GPU-hour" and the node price.

      Marketing Manager 2 hours

    3. 3

      Fix the homepage

      Same H100 number.

      Marketing Manager 1 hour

    4. 4

      Add a comparison line

      Dated rival list prices.

      Marketing Manager 2 hours

    5. 5

      Check the docs

      Prices and units in docs and blog.

      Marketing Manager 1 day

    6. 6

      Set an owner

      Monthly price review.

      CRO and Co-Founder monthly

    First result
    1 week
    Cost
    $0
    Owner
    Marketing Manager

    Smallest test

    Fix the card labels today.

    Why it works

    Pick one current price per GPU, label every card "per GPU-hour" with the 8-GPU node price beneath, use the same number on the homepage, and add a dated "vs Runpod, Lambda, AWS" line.

    Clear unit pricing is what lets a price claim win; ambiguity sends buyers back to the rival with a simpler table.

    Risks and how to handle them

    • Prices move often

      Keep prices in one Webflow CMS field used everywhere.

    Evidence

    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
    • Your websitePricing page: the table says H100 SXM "Per GPU/hr ~$3.29" vs hyperscalers "~$7-13"; the cards read "H100 Cluster $3.29 per hour" (8x H100), "H200 Cluster $4.59 per hour", "B200 Cluster $6.79 per hour" and "B300 Cluster $7.89 per hour", each above an "8x" spec list. Every card's "Rent now" button links to www.runpod.io/, the partner's generic homepage. Reserved 1-5 year rates are quoted on request.Source
    • Competitor researchRivals' own pages: Runpod lists H100 SXM at "$3.99 /hr", H200 at "$5.29 /hr" and B200 at "$7.99 /hr" per GPU on demand; Lambda lists 1-Click Clusters of "16 to 2,000+" B200 or H100 GPUs, B200 at "$9.86" per GPU-hour for 2 weeks to 1 year; Nebius promises "From zero to clusters in minutes, with built-in repeatability and self-service"; Crusoe calls itself "The AI factory company".Source
  • 04

    Every page ends "© {{year}} Copyright." with the template tag unfilled. For a company asking teams to trust it with production training runs and confidential workloads, a visible template error on every page is a small but real credibility cost.

    Medium impactLow effort First result in 1 day Marketing Manager

    First action

    Edit the footer symbol in Webflow.

    How to measure it

    Unfilled placeholders

    Now
    –
    Target
    0
    By
    1 week

    Steps

    1. 1

      Fix the footer

      Webflow footer symbol.

      Marketing Manager 15 minutes

    2. 2

      Search for tags

      Look for "{{" across pages.

      Marketing Manager 30 minutes

    3. 3

      Check forms

      Success and error messages.

      Marketing Manager 15 minutes

    4. 4

      Check meta

      Titles and descriptions.

      Marketing Manager 15 minutes

    5. 5

      Publish

      Republish the site.

      Marketing Manager 5 minutes

    6. 6

      Add to checklist

      Pre-publish review.

      Marketing Manager 5 minutes

    First result
    1 day
    Cost
    $0
    Owner
    Marketing Manager

    Smallest test

    Fix it today.

    Why it works

    Replace the tag with a static year or a small script, and run a quick check of the site for other unrendered tags and placeholder text.

    Infrastructure buyers read details as signs of operational care; it is a one-hour fix.

    Risks and how to handle them

    • Script-based year fails

      Use a static year and update it each January.

    Evidence

    • Your websiteEvery page footer reads "© {{year}} Copyright. All Rights Reserved." with the template placeholder unfilled (checked November 4, 2026 on the homepage, pricing, company, platform and performance pages).Source
    • Your websiteCompany page: "The Neocloud vision" - "a network of local, auditable facilities so you know where your workloads run"; Sovereign ("even we cannot see your workloads") and Sustainable pillars; founders from Intel data center SSD strategy, IT asset disposition and cloud ventures. Main navigation: Platform, Infrastructure, Performance, Pricing, Company, Docs, Blog, Products; no page for GPU owners, investors or data center operators.Source
  • 05

    The main site has no sitemap (the URL returns "Page Not Found"), and the blog and docs live on separate hosts. Search engines and AI crawlers find FarmGPU's pages only by following links, which slows indexing of new pricing and benchmark pages.

    Medium impactLow effort First result in 2 weeks Marketing Manager

    First action

    Enable the Webflow sitemap setting.

    How to measure it

    Indexed pages

    Now
    –
    Target
    all public pages
    By
    30 days

    Steps

    1. 1

      Enable sitemap

      Webflow SEO settings.

      Marketing Manager 15 minutes

    2. 2

      Update robots.txt

      Sitemap lines for each host.

      Marketing Manager 15 minutes

    3. 3

      Verify hosts

      Search Console and Bing.

      Marketing Manager 1 hour

    4. 4

      Submit sitemaps

      Main, blog and docs.

      Marketing Manager 30 minutes

    5. 5

      Check coverage

      Indexed vs excluded.

      Marketing Manager 2 weeks

    6. 6

      Monitor

      Monthly.

      Marketing Manager monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Marketing Manager

    Smallest test

    Enable the sitemap and submit it today.

    Why it works

    Turn on Webflow's automatic sitemap, add the sitemap to robots.txt, verify all three hosts in Google Search Console and Bing Webmaster Tools, and submit each sitemap.

    Faster, complete indexing matters most for pages that change often, such as pricing and benchmarks.

    Risks and how to handle them

    • Thin pages indexed

      Exclude form-only pages.

    Evidence

    • Your websitehttps://farmgpu.com/sitemap.xml returns HTTP 404 ("Page Not Found | FarmGPU"), checked November 4, 2026. The site is on Webflow; the blog (blog.farmgpu.com) and docs (docs.farmgpu.com) are separate hosts.Source
    • Your docsdocs.farmgpu.com (Mintlify) is company documentation: snapshot, thesis, hardware and software stack, data center specs, security, legal, team, culture, brand. Its llms.txt index lists no quickstart, API reference or "launch your first cluster" guide; Instant Clusters are launched "from the RunPod dashboard with a few clicks". Two citations on that page link to the blog with "?utm_source=chatgpt.com" appended.Source
    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
  • 06

    The performance page lists "99.9% Production SLA". The SLA document commits 99.7% by default for bare metal (99.9% only on the paid Enterprise tier) and no SLA for on-demand, while the compliance page cites a 99.99% facility target. Enterprise buyers and procurement read the SLA; a headline number that the contract does not back slows deals.

    Medium impactLow effort First result in 1 week Global Sales Manager

    First action

    Change the performance page row to "Up to 99.9% (Enterprise SLA)".

    How to measure it

    Quote-to-contract rate

    Now
    –
    Target
    +10%
    By
    2 quarters

    Steps

    1. 1

      Agree wording

      Tiers and coverage.

      CRO and Co-Founder 1 day

    2. 2

      Update performance page

      Row and footnote.

      Marketing Manager 1 hour

    3. 3

      Add an SLA box

      On pricing.

      Marketing Manager 2 hours

    4. 4

      Link the SLA

      From both pages.

      Marketing Manager 15 minutes

    5. 5

      Brief sales

      One answer for buyers.

      Global Sales Manager 1 day

    6. 6

      Review

      With each SLA change.

      CRO and Co-Founder ongoing

    First result
    1 week
    Cost
    $0
    Owner
    Global Sales Manager

    Smallest test

    Edit the performance page row today.

    Why it works

    Show the real tiers on the pricing and performance pages (Standard 99.0%, Premium 99.7%, Enterprise 99.9%, facility 99.99%), say which products they cover, and link the SLA.

    Precise, honest reliability numbers build trust with the engineers and procurement teams who check them.

    Risks and how to handle them

    • Lower headline number

      Lead with the 99.99% facility design and the Enterprise tier.

    Evidence

    • Your websitePerformance page: "#1 in Clustermax 2.0 Storage, Torch Import benchmark", "Fastest single node / client for ML Perf 2.0 Storage Training Workloads", torch import "<1 second" vs "tens of seconds", cold start "< 60 SEC"; Availability "99.9% Production SLA". It ends: "Spin up a cluster, run the tests yourself, and compare against your current infrastructure." with a "Request access" button.Source
    • Your docsService Level Agreement: covers Reserved Instance and Bare Metal Services only; "On-Demand Services and Spot Instance Services are provided without service level guarantees". Committed monthly uptime: Standard 99.0%, Premium 99.7%, Enterprise 99.9%; Bare Metal defaults to Premium (99.7%).Source
    • Your docsCompliance page: "SOC 2 Type II (In Progress)", "Status: In progress", "Target completion: Q2 2026" (still shown on November 4, 2026); data centers ISO/IEC 27001 "certified or aligned"; Tier III+ with a "Facility availability target: 99.99%".Source

This quarter 10

The core plays for the next three months.

  • 07

    The compliance page says SOC 2 Type II is "In progress" with a Q2 2026 target. That date has passed. Enterprise and confidential-computing buyers check this first, and an expired target reads worse than a later honest one.

    High impactMedium effort First result in 1 quarter CRO and Co Founder

    First action

    Update the SOC 2 line with today's status.

    How to measure it

    Sales cycle length

    Now
    –
    Target
    -20%
    By
    2 quarters

    Steps

    1. 1

      Confirm status

      With the auditor.

      CFO 1 day

    2. 2

      Update docs

      Status and date.

      CRO and Co-Founder 1 hour

    3. 3

      Build trust page

      SLA, certifications, confidential computing.

      Marketing Manager 1 week

    4. 4

      Add report request

      NDA form.

      CRO and Co-Founder 2 days

    5. 5

      Link it

      Footer, pricing, demo emails.

      Marketing Manager 1 day

    6. 6

      Keep current

      Quarterly review.

      CRO and Co-Founder quarterly

    First result
    1 quarter
    Cost
    Audit costs already planned
    Owner
    CRO and Co Founder

    Smallest test

    Fix the date line this week.

    Why it works

    Publish the current status with a new date (or the report, if issued), add a simple trust page with the SLA, data center certifications, confidential computing and a request form for the report, and link it from the footer and pricing.

    Security questionnaires are the main delay in infrastructure deals; a current trust page answers most of them up front.

    Risks and how to handle them

    • Audit slips again

      Publish a status without a fixed date.

    Evidence

    • Your docsCompliance page: "SOC 2 Type II (In Progress)", "Status: In progress", "Target completion: Q2 2026" (still shown on November 4, 2026); data centers ISO/IEC 27001 "certified or aligned"; Tier III+ with a "Facility availability target: 99.99%".Source
    • Your websiteCompany page: "The Neocloud vision" - "a network of local, auditable facilities so you know where your workloads run"; Sovereign ("even we cannot see your workloads") and Sustainable pillars; founders from Intel data center SSD strategy, IT asset disposition and cloud ventures. Main navigation: Platform, Infrastructure, Performance, Pricing, Company, Docs, Blog, Products; no page for GPU owners, investors or data center operators.Source
    • Your docsService Level Agreement: covers Reserved Instance and Bare Metal Services only; "On-Demand Services and Spot Instance Services are provided without service level guarantees". Committed monthly uptime: Standard 99.0%, Premium 99.7%, Enterprise 99.9%; Bare Metal defaults to Premium (99.7%).Source
  • 08

    docs.farmgpu.com covers company snapshot, thesis, security and legal, but has no quickstart, API reference or first-run guide. Instant Clusters are launched from the Runpod dashboard, so a developer has to work out the path between two companies.

    High impactMedium effort First result in 1 month Head of Engineering

    First action

    Record the steps of one engineer launching a cluster from scratch.

    How to measure it

    Quickstart completions

    Now
    –
    Target
    200 a quarter
    By
    2 quarters

    Steps

    1. 1

      Record a run

      From signup to first job.

      Head of Engineering 1 day

    2. 2

      Write the guide

      Ten minutes, copy-paste commands.

      Head of Engineering 1 week

    3. 3

      Add screenshots

      Runpod and FarmGPU views.

      Marketing Manager 2 days

    4. 4

      Add benchmarks

      Torch import and storage tests.

      Head of Engineering 1 week

    5. 5

      Link everywhere

      Deploy page, pricing, blog.

      Marketing Manager 1 day

    6. 6

      Add reserved guides

      Slurm and Kubernetes.

      Head of Engineering 1 month

    7. 7

      Track

      Page views to starts.

      Marketing Manager monthly

    First result
    1 month
    Cost
    Engineering time
    Owner
    Head of Engineering

    Smallest test

    Publish the first guide as a blog post.

    Why it works

    Add "Launch your first FarmGPU cluster in 10 minutes": choose FarmGPU capacity on Runpod, pick a template, run a short PyTorch job, check the sub-second torch import and the storage speed. Add guides for Slurm and Kubernetes on reserved clusters.

    Developers judge a cloud by its first hour; a quickstart turns benchmark claims into a lived experience.

    Risks and how to handle them

    • Runpod UI changes

      Review the guide monthly.

    Evidence

    • Your docsdocs.farmgpu.com (Mintlify) is company documentation: snapshot, thesis, hardware and software stack, data center specs, security, legal, team, culture, brand. Its llms.txt index lists no quickstart, API reference or "launch your first cluster" guide; Instant Clusters are launched "from the RunPod dashboard with a few clicks". Two citations on that page link to the blog with "?utm_source=chatgpt.com" appended.Source
    • Your websitePerformance page: "#1 in Clustermax 2.0 Storage, Torch Import benchmark", "Fastest single node / client for ML Perf 2.0 Storage Training Workloads", torch import "<1 second" vs "tens of seconds", cold start "< 60 SEC"; Availability "99.9% Production SLA". It ends: "Spin up a cluster, run the tests yourself, and compare against your current infrastructure." with a "Request access" button.Source
    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
  • 09

    The performance page invites buyers to "Spin up a cluster, run the tests yourself" but the button is "Request access" with no details. FarmGPU's strongest proof is #1 storage results; letting a prospect reproduce them on their own workload is the most persuasive sales step.

    High impactMedium effort First result in 1 quarter Global Sales Manager

    First action

    Write the benchmark scripts as a public repository.

    How to measure it

    Trial-to-quote rate

    Now
    –
    Target
    30%
    By
    2 quarters

    Steps

    1. 1

      Build the kit

      Scripts and expected results.

      Head of Engineering 2 weeks

    2. 2

      Set credit rules

      Who qualifies and how much.

      CRO and Co-Founder 1 week

    3. 3

      Trial page

      Replace "Request access".

      Marketing Manager 1 week

    4. 4

      Follow-up

      Results call and quote.

      Global Sales Manager ongoing

    5. 5

      Publish anonymized results

      Customer vs prior cloud.

      Marketing Manager quarterly

    6. 6

      Measure

      Trials, quotes, contracts.

      Global Sales Manager monthly

    First result
    1 quarter
    Cost
    Credits for trials
    Owner
    Global Sales Manager

    Smallest test

    Run one trial with a current prospect.

    Why it works

    Package a benchmark kit: scripts for torch import, MLPerf-style storage and NCCL tests, a few hours of credits for qualified teams, and a results page comparing their current cloud. Sales follows up with a reserved quote based on their numbers.

    Buyers trust their own measurements, and a trial converts the benchmark lead into a pricing conversation.

    Risks and how to handle them

    • Credit abuse

      Require a work email and a call.

    Evidence

    • Your websitePerformance page: "#1 in Clustermax 2.0 Storage, Torch Import benchmark", "Fastest single node / client for ML Perf 2.0 Storage Training Workloads", torch import "<1 second" vs "tens of seconds", cold start "< 60 SEC"; Availability "99.9% Production SLA". It ends: "Spin up a cluster, run the tests yourself, and compare against your current infrastructure." with a "Request access" button.Source
    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
    • Your websiteHomepage testimonials from the CEO of SemiAnalysis ("FarmGPU is going to take over the world."), the CEO of game studio Other Ocean (dedicated nodes "it just works"), Runpod's CTO ("a key strategic partner for RunPod") and Solidigm's head of products and marketing (AI Central Lab). The site has no case studies or customer results pages.Source
  • 10

    FarmGPU's headline is price, but its only comparison is a generic "hyperscalers" column. Buyers shortlist neoclouds: Lambda sells 1-Click Clusters, Nebius "zero to clusters in minutes", Crusoe "the AI factory company".

    High impactMedium effort First result in 1 quarter Marketing Manager

    First action

    Pick the two rivals most often in deals.

    How to measure it

    Comparison-page leads

    Now
    –
    Target
    20 a month
    By
    2 quarters

    Steps

    1. 1

      Pick rivals

      From recent deals.

      Global Sales Manager 1 day

    2. 2

      Collect facts

      Public prices and specs.

      Marketing Manager 1 week

    3. 3

      Write pages

      Fair tables.

      Marketing Manager 2 weeks

    4. 4

      Review

      Claims and trademarks.

      CRO and Co-Founder 3 days

    5. 5

      Publish

      With a savings calculator link.

      Marketing Manager 1 day

    6. 6

      Refresh

      Monthly price check.

      Marketing Manager monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Marketing Manager

    Smallest test

    Publish one vs-hyperscaler page first.

    Why it works

    Write fair, dated comparison pages: per-GPU prices, billing increment, egress, storage throughput, minimum terms, SLA, confidential computing. Cite sources and update monthly.

    Comparison searches come from buyers close to a decision, and they are what AI assistants quote.

    Risks and how to handle them

    • Partner sensitivity with Runpod

      Frame Runpod as the on-demand route, compare only reserved terms.

    Evidence

    • Competitor researchRivals' own pages: Runpod lists H100 SXM at "$3.99 /hr", H200 at "$5.29 /hr" and B200 at "$7.99 /hr" per GPU on demand; Lambda lists 1-Click Clusters of "16 to 2,000+" B200 or H100 GPUs, B200 at "$9.86" per GPU-hour for 2 weeks to 1 year; Nebius promises "From zero to clusters in minutes, with built-in repeatability and self-service"; Crusoe calls itself "The AI factory company".Source
    • Your websitePricing page: the table says H100 SXM "Per GPU/hr ~$3.29" vs hyperscalers "~$7-13"; the cards read "H100 Cluster $3.29 per hour" (8x H100), "H200 Cluster $4.59 per hour", "B200 Cluster $6.79 per hour" and "B300 Cluster $7.89 per hour", each above an "8x" spec list. Every card's "Rent now" button links to www.runpod.io/, the partner's generic homepage. Reserved 1-5 year rates are quoted on request.Source
    • Your websitePerformance page: "#1 in Clustermax 2.0 Storage, Torch Import benchmark", "Fastest single node / client for ML Perf 2.0 Storage Training Workloads", torch import "<1 second" vs "tens of seconds", cold start "< 60 SEC"; Availability "99.9% Production SLA". It ends: "Spin up a cluster, run the tests yourself, and compare against your current infrastructure." with a "Request access" button.Source
  • 11

    The homepage has praise from SemiAnalysis, game studio Other Ocean, Runpod and Solidigm, but there are no case studies with numbers. Buyers want "who like me saved what".

    High impactMedium effort First result in 1 quarter Marketing Manager

    First action

    Ask Other Ocean for a 30-minute interview.

    How to measure it

    Case study views before demo

    Now
    –
    Target
    30% of demos
    By
    2 quarters

    Steps

    1. 1

      Pick three

      Different customer types.

      Global Sales Manager 1 week

    2. 2

      Interview

      30 minutes each.

      Marketing Manager 2 weeks

    3. 3

      Get numbers

      Cost and speed before and after.

      Head of Engineering 2 weeks

    4. 4

      Write

      One page each.

      Marketing Manager 2 weeks

    5. 5

      Approve

      Customer sign-off.

      Global Sales Manager 2 weeks

    6. 6

      Publish and share

      Site, LinkedIn, sales emails.

      Marketing Manager 1 week

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Marketing Manager

    Smallest test

    Write the first story from the existing testimonial.

    Why it works

    Write three short stories with measured results (cost per training run, time to first step, storage throughput) for a startup, a studio and a partner lab. Add them to pricing and the deploy page.

    Specific results from similar teams turn a price claim into a believable one.

    Risks and how to handle them

    • Customers decline to share costs

      Use percentages.

    Evidence

    • Your websiteHomepage testimonials from the CEO of SemiAnalysis ("FarmGPU is going to take over the world."), the CEO of game studio Other Ocean (dedicated nodes "it just works"), Runpod's CTO ("a key strategic partner for RunPod") and Solidigm's head of products and marketing (AI Central Lab). The site has no case studies or customer results pages.Source
    • Your websitePerformance page: "#1 in Clustermax 2.0 Storage, Torch Import benchmark", "Fastest single node / client for ML Perf 2.0 Storage Training Workloads", torch import "<1 second" vs "tens of seconds", cold start "< 60 SEC"; Availability "99.9% Production SLA". It ends: "Spin up a cluster, run the tests yourself, and compare against your current infrastructure." with a "Request access" button.Source
    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
  • 12

    FarmGPU also runs GPUs for others: "Turnkey GPU Investment" (bring your capex, FarmGPU manages and monetizes it) and an "AI Data Center Blueprint" for facility operators. These are described only in the docs; the site navigation speaks to AI teams alone, and the investor calculator sits behind a buyer button.

    High impactMedium effort First result in 1 quarter CBO and Co Founder

    First action

    Draft the page from the docs text.

    How to measure it

    Qualified supply-side calls

    Now
    –
    Target
    5 a quarter
    By
    2 quarters

    Steps

    1. 1

      Draft the page

      From docs and deal desk.

      CBO and Co-Founder 1 week

    2. 2

      Build it

      Webflow page and nav link.

      Marketing Manager 3 days

    3. 3

      Set a default case

      A realistic deal desk example.

      CFO 1 week

    4. 4

      List targets

      Operators and GPU funds.

      CBO and Co-Founder 2 weeks

    5. 5

      Outreach

      Personal emails and events.

      CBO and Co-Founder ongoing

    6. 6

      Track

      Calls and deals.

      CBO and Co-Founder monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    CBO and Co Founder

    Smallest test

    Publish the page with one example deal.

    Why it works

    Add a "For GPU owners and operators" page: what FarmGPU manages, the monetization route through partner platforms, the deal desk calculator, and a call with the CBO. Target data center operators, family offices and GPU funds.

    Supply-side deals are large and few; one page and a calculator give that pipeline a home.

    Risks and how to handle them

    • Mixed audiences confuse buyers

      Keep it in its own nav item.

    Evidence

    • Your docsCompany snapshot: Farm GPU Inc., founded 2024, Rancho Cordova, CA; data centers in Rancho Cordova and Santa Clara; "~12 (FTE + Strategic Contractors)". Four offerings: on-demand compute; managed Slurm/Kubernetes clusters; "Turnkey GPU Investment: Bring your own capex, have FarmGPU manage end to end ... to monetization on our partner platforms"; and an "AI Data Center Blueprint" for traditional facility operators. Products page: managed infrastructure for "Data center operators with unused GPU capacity" and "Investors or consortiums deploying GPU capital".Source
    • Your websiteThe homepage "Lowest TCO / 70% Less Than Hyperscalers / Calculate savings" card links to calculator.farmgpu.com, which opens a "Deal Desk" to "Price an offtake for 512 x H100 SXM": server price, debt share, cost of equity, colocation per kW. Its default view shows "Project IRR -9.0%", "Equity multiple 0.44x" and "Payback Never". It is a GPU-investor model, not a buyer savings calculator.Source
    • Your websiteCompany page: "The Neocloud vision" - "a network of local, auditable facilities so you know where your workloads run"; Sovereign ("even we cannot see your workloads") and Sustainable pillars; founders from Intel data center SSD strategy, IT asset disposition and cloud ventures. Main navigation: Platform, Infrastructure, Performance, Pricing, Company, Docs, Blog, Products; no page for GPU owners, investors or data center operators.Source
  • 13

    The blog publishes rare material (MLPerf Storage v3.0, KV cache results, a 17-day B200 network build) and has a Subscribe link, but nothing on the main site collects emails, and posts are not repackaged.

    Medium impactLow effort First result in 1 quarter Marketing Manager

    First action

    Add a sign-up block to the homepage.

    How to measure it

    Subscribers

    Now
    –
    Target
    1,000
    By
    2 quarters

    Steps

    1. 1

      Add sign-up

      Main site pages.

      Marketing Manager 1 day

    2. 2

      Connect to blog list

      One list.

      Marketing Manager 1 day

    3. 3

      Plan issues

      Three months ahead.

      Marketing Manager 1 week

    4. 4

      Send monthly

      Short and technical.

      Marketing Manager monthly

    5. 5

      Repurpose

      LinkedIn and X threads.

      Marketing Manager weekly

    6. 6

      Measure

      Opens and demos.

      Marketing Manager monthly

    First result
    1 quarter
    Cost
    Email tool costs
    Owner
    Marketing Manager

    Smallest test

    Send one issue to existing contacts.

    Why it works

    Add a newsletter sign-up to the main site, pricing and deploy pages; send a monthly note with one benchmark, one build lesson and one price update; cut each post into LinkedIn and X threads.

    Infrastructure buyers evaluate for months; email keeps FarmGPU in the shortlist.

    Risks and how to handle them

    • Low opens

      Lead each issue with a number.

    Evidence

    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
  • 14

    farmgpu.com mail runs on Google Workspace with SPF "~all" and no DMARC record. Without DMARC, newsletters, sales sequences and trial emails are more likely to land in spam, and the domain can be spoofed.

    Medium impactLow effort First result in 2 months Marketing Manager

    First action

    Publish a DMARC record with p=none.

    How to measure it

    DMARC policy

    Now
    –
    Target
    p=reject
    By
    2 months

    Steps

    1. 1

      Enable DKIM

      Google Workspace.

      Head of Engineering 1 hour

    2. 2

      Publish DMARC

      p=none with reports.

      Head of Engineering 30 minutes

    3. 3

      Review reports

      Two weeks of senders.

      Head of Engineering 2 weeks

    4. 4

      Fix senders

      Form and email tools.

      Marketing Manager 1 week

    5. 5

      Tighten

      Quarantine then reject.

      Head of Engineering 1 month

    6. 6

      Monitor

      Monthly reports.

      Head of Engineering monthly

    First result
    2 months
    Cost
    $0
    Owner
    Marketing Manager

    Smallest test

    Publish p=none today.

    Why it works

    Turn on DKIM in Google Workspace, publish DMARC at "p=none" with reporting, fix any failing senders, then move to quarantine and reject.

    Deliverability is the base of every email channel the other ideas add.

    Risks and how to handle them

    • Legitimate mail blocked

      Move to reject only after clean reports.

    Evidence

    • DNS and security checkfarmgpu.com uses Google Workspace mail with SPF "~all" and no DMARC record. Mozilla Observatory grades the site C- (45/100) with no Content-Security-Policy header. The company is not registered in SAM.gov (the US federal supplier register).Source
    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
  • 15

    Real Chrome visitors already pass Core Web Vitals (largest paint 659 ms). In Google's phone lab test the homepage scores 70 with a 4.6 s largest paint, mostly 2.3 s of render-blocking requests and 447 KB of unused JavaScript. Mobile visitors from social links get the slower experience.

    Low impactLow effort First result in 2 weeks Marketing Manager

    First action

    Load reCAPTCHA only on the form pages.

    How to measure it

    Lab performance score

    Now
    –
    Target
    90+
    By
    1 month

    Steps

    1. 1

      Move reCAPTCHA

      Form pages only.

      Marketing Manager 1 hour

    2. 2

      Defer scripts

      Async analytics.

      Marketing Manager 1 hour

    3. 3

      Compress images

      WebP and sizes.

      Marketing Manager 2 hours

    4. 4

      Retest

      PageSpeed on phone.

      Marketing Manager 15 minutes

    5. 5

      Check forms

      Still submit.

      Marketing Manager 30 minutes

    6. 6

      Monitor

      Monthly.

      Marketing Manager monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Marketing Manager

    Smallest test

    Move reCAPTCHA off the homepage.

    Why it works

    Defer the reCAPTCHA script to the form pages only, load jQuery and analytics asynchronously, compress the large images, and retest.

    A fast page backs up a speed-led brand, and it is a small job.

    Risks and how to handle them

    • Forms break

      Test each form after changes.

    Evidence

    • Google PageSpeedHomepage, phone test: performance 70/100 (desktop 99); largest content paint 4.6 s in the lab (target 2.5 s), render-blocking requests about 2.3 s, 447 KB of unused JavaScript, 361 KB of image savings. Real Chrome visitors pass: largest paint 659 ms, interaction 30 ms, layout shift 0 at the 75th percentile.Source
  • 16

    The October blog post says confidential VMs with Intel TDX and TractorOS "launch on FarmGPU Cloud by the end of 2026". There is no product page or waitlist to capture interest from teams building agents with sensitive data.

    High impactMedium effort First result in 1 quarter CEO and Co Founder

    First action

    Add a waitlist form to the blog post.

    How to measure it

    Waitlist

    Now
    –
    Target
    200
    By
    1 quarter

    Steps

    1. 1

      Product page

      Use cases and specs.

      Marketing Manager 1 week

    2. 2

      Waitlist

      Form and email list.

      Marketing Manager 1 day

    3. 3

      Design partners

      Five teams.

      CEO and Co-Founder 1 month

    4. 4

      Launch content

      Partner stories.

      Marketing Manager 1 month

    5. 5

      Intel co-marketing

      Joint post.

      CBO and Co-Founder 1 month

    6. 6

      Measure

      Sign-ups to usage.

      Marketing Manager monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    CEO and Co Founder

    Smallest test

    Add the waitlist form this week.

    Why it works

    Publish a page for confidential agents with a waitlist, recruit five design partners, and launch with their stories.

    A launch with partners and a list converts far better than an announcement into silence.

    Risks and how to handle them

    • Launch slips

      Say "early access" rather than a date.

    Evidence

    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
    • Your websiteCompany page: "The Neocloud vision" - "a network of local, auditable facilities so you know where your workloads run"; Sovereign ("even we cannot see your workloads") and Sustainable pillars; founders from Intel data center SSD strategy, IT asset disposition and cloud ventures. Main navigation: Platform, Infrastructure, Performance, Pricing, Company, Docs, Blog, Products; no page for GPU owners, investors or data center operators.Source
    • Your docsCompliance page: "SOC 2 Type II (In Progress)", "Status: In progress", "Target completion: Q2 2026" (still shown on November 4, 2026); data centers ISO/IEC 27001 "certified or aligned"; Tier III+ with a "Facility availability target: 99.99%".Source

Big bets 3

Larger plays that can change the trajectory.

  • 17

    FarmGPU's vision is "a network of local, auditable facilities" where "even we cannot see your workloads", and it runs in Rancho Cordova and Santa Clara. It is not registered in SAM.gov, which public universities, labs and agencies often need before they can buy.

    High impactHigh effort First result in 3 quarters CBO and Co Founder

    First action

    Start the SAM.gov registration.

    How to measure it

    Public-sector pilots

    Now
    –
    Target
    2
    By
    4 quarters

    Steps

    1. 1

      Register

      SAM.gov and UEI.

      CFO 1 month

    2. 2

      Write the page

      Sovereign compute.

      Marketing Manager 2 weeks

    3. 3

      List targets

      Universities and agencies in California.

      CBO and Co-Founder 1 month

    4. 4

      Find vehicles

      Cooperative purchasing.

      CBO and Co-Founder 1 quarter

    5. 5

      Pilot

      Two contracts.

      CRO and Co-Founder 2 quarters

    6. 6

      Publish results

      Case studies.

      Marketing Manager 1 quarter

    First result
    3 quarters
    Cost
    Registration and proposal time
    Owner
    CBO and Co Founder

    Smallest test

    Complete the SAM.gov registration.

    Why it works

    Register in SAM.gov, write a sovereign compute page for public research and regulated industries, and pursue two pilot contracts with University of California campuses or state agencies through cooperative purchasing.

    Sovereign and confidential compute is a premium segment where a local, US-owned neocloud has an edge over hyperscalers.

    Risks and how to handle them

    • Long cycles

      Pursue research grants with short paths first.

    Evidence

    • Your websiteCompany page: "The Neocloud vision" - "a network of local, auditable facilities so you know where your workloads run"; Sovereign ("even we cannot see your workloads") and Sustainable pillars; founders from Intel data center SSD strategy, IT asset disposition and cloud ventures. Main navigation: Platform, Infrastructure, Performance, Pricing, Company, Docs, Blog, Products; no page for GPU owners, investors or data center operators.Source
    • DNS and security checkfarmgpu.com uses Google Workspace mail with SPF "~all" and no DMARC record. Mozilla Observatory grades the site C- (45/100) with no Content-Security-Policy header. The company is not registered in SAM.gov (the US federal supplier register).Source
    • Your docsCompany snapshot: Farm GPU Inc., founded 2024, Rancho Cordova, CA; data centers in Rancho Cordova and Santa Clara; "~12 (FTE + Strategic Contractors)". Four offerings: on-demand compute; managed Slurm/Kubernetes clusters; "Turnkey GPU Investment: Bring your own capex, have FarmGPU manage end to end ... to monetization on our partner platforms"; and an "AI Data Center Blueprint" for traditional facility operators. Products page: managed infrastructure for "Data center operators with unused GPU capacity" and "Investors or consortiums deploying GPU capital".Source
  • 18

    FarmGPU's results are storage-first: #1 in ClusterMax 2.0 storage, MLPerf Storage v3.0 records, and 100x-280x KV cache acceleration shown at GTC 2026. Inference is where demand is moving, and long-context KV caching is a storage problem.

    High impactHigh effort First result in 3 quarters Head of Engineering

    First action

    Write a one-page spec with Solidigm and the KV cache partners.

    How to measure it

    Inference cluster customers

    Now
    –
    Target
    5
    By
    4 quarters

    Steps

    1. 1

      Spec

      Tier and pricing.

      Head of Engineering 1 month

    2. 2

      Partners

      Storage and software.

      CBO and Co-Founder 1 month

    3. 3

      Benchmark

      Published method.

      Head of Engineering 1 quarter

    4. 4

      Beta

      Three startups.

      Global Sales Manager 1 quarter

    5. 5

      Launch

      Page and post.

      Marketing Manager 1 month

    6. 6

      Price

      From beta data.

      CRO and Co-Founder 1 month

    First result
    3 quarters
    Cost
    Engineering and hardware
    Owner
    Head of Engineering

    Smallest test

    Interview five inference teams about cache costs.

    Why it works

    Offer an inference cluster with a named KV-cache tier, a published benchmark against standard setups, and pricing per token-served improvement. Sell it to inference startups first.

    It turns a benchmark lead into a product rivals cannot easily copy.

    Risks and how to handle them

    • Niche demand

      Validate with five interviews first.

    Evidence

    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
    • Your websitePerformance page: "#1 in Clustermax 2.0 Storage, Torch Import benchmark", "Fastest single node / client for ML Perf 2.0 Storage Training Workloads", torch import "<1 second" vs "tens of seconds", cold start "< 60 SEC"; Availability "99.9% Production SLA". It ends: "Spin up a cluster, run the tests yourself, and compare against your current infrastructure." with a "Request access" button.Source
    • Competitor researchRivals' own pages: Runpod lists H100 SXM at "$3.99 /hr", H200 at "$5.29 /hr" and B200 at "$7.99 /hr" per GPU on demand; Lambda lists 1-Click Clusters of "16 to 2,000+" B200 or H100 GPUs, B200 at "$9.86" per GPU-hour for 2 weeks to 1 year; Nebius promises "From zero to clusters in minutes, with built-in repeatability and self-service"; Crusoe calls itself "The AI factory company".Source
  • 19

    FarmGPU chairs a new OCP workstream on scaling AI clusters at neoclouds and has written up building a 32-node B200 cluster. Operators and investors want that know-how.

    High impactMedium effort First result in 2 quarters CEO and Co Founder

    First action

    Outline the guide from existing posts.

    How to measure it

    Guide downloads

    Now
    –
    Target
    2,000
    By
    1 year

    Steps

    1. 1

      Outline

      From blog posts.

      CEO and Co-Founder 2 weeks

    2. 2

      Contributors

      OCP members.

      CEO and Co-Founder 1 month

    3. 3

      Write

      Chapters.

      Head of Engineering 1 quarter

    4. 4

      Publish

      Gated PDF and web.

      Marketing Manager 1 month

    5. 5

      Workshop

      Yearly event.

      CBO and Co-Founder 1 year

    6. 6

      Follow up

      Blueprint offers.

      CBO and Co-Founder ongoing

    First result
    2 quarters
    Cost
    Writing time
    Owner
    CEO and Co Founder

    Smallest test

    Publish chapter one as a post.

    Why it works

    Publish an open guide (network, storage, provisioning, reliability) with OCP contributors, host a yearly workshop, and offer the AI Data Center Blueprint service to readers.

    Being the author of the playbook makes FarmGPU the first call for operators building capacity.

    Risks and how to handle them

    • Helps competitors

      Keep the operations software proprietary.

    Evidence

    • Your blogBlog (with a Subscribe link): "Agents Need a Cloud They Can Verify" (October 23, 2026: TractorOS and confidential VMs on Intel Xeon 6 with Intel TDX "launch on FarmGPU Cloud by the end of 2026"); MLPerf Storage v3.0 results with Solidigm and WEKA (September 2026); FarmGPU chairing a new OCP workstream on scaling AI clusters at neoclouds (April 2026); 100x-280x KV cache acceleration with Lightbits and ScaleFlux at GTC 2026; #1 in ClusterMax 2.0 storage (March 2026); a 17-day crash course building a 32-node B200 cluster.Source
    • Your docsCompany snapshot: Farm GPU Inc., founded 2024, Rancho Cordova, CA; data centers in Rancho Cordova and Santa Clara; "~12 (FTE + Strategic Contractors)". Four offerings: on-demand compute; managed Slurm/Kubernetes clusters; "Turnkey GPU Investment: Bring your own capex, have FarmGPU manage end to end ... to monetization on our partner platforms"; and an "AI Data Center Blueprint" for traditional facility operators. Products page: managed infrastructure for "Data center operators with unused GPU capacity" and "Investors or consortiums deploying GPU capital".Source

Wildcards 3

Unexpected experiments, each with a cheap way to test it.

  • 20

    Developers now ask assistants for "cheapest H100 cloud". The docs have an llms.txt, but the main site, where prices live, does not, and prices differ between pages. Docs also link with "utm_source=chatgpt.com" tags, a sign the content was drafted with an assistant and not cleaned.

    Medium impactLow effort First result in 1 quarter Marketing Manager

    First action

    Write the llms.txt with the pricing facts.

    How to measure it

    Correct price mentions

    Now
    –
    Target
    5 of 10 prompts
    By
    1 quarter

    Steps

    1. 1

      Write llms.txt

      Prices and links.

      Marketing Manager 1 day

    2. 2

      Add structured data

      Offers on pricing.

      Marketing Manager 1 day

    3. 3

      Clean docs links

      Remove tags.

      Head of Engineering 1 hour

    4. 4

      Test prompts

      Ten buyer questions.

      Marketing Manager 1 day

    5. 5

      Repeat

      Monthly.

      Marketing Manager monthly

    6. 6

      Update

      With price changes.

      Marketing Manager ongoing

    First result
    1 quarter
    Cost
    $0
    Owner
    Marketing Manager

    Smallest test

    Run ten prompts and record answers.

    Why it works

    Add an llms.txt on farmgpu.com with current per-GPU prices, billing, SLA tiers and links; add pricing structured data; clean tracking tags from docs links; check monthly what assistants answer.

    Clear, consistent facts are what assistants repeat, and price is FarmGPU's headline.

    Risks and how to handle them

    • Assistants lag

      Keep facts consistent everywhere.

    Evidence

    • Your docsdocs.farmgpu.com (Mintlify) is company documentation: snapshot, thesis, hardware and software stack, data center specs, security, legal, team, culture, brand. Its llms.txt index lists no quickstart, API reference or "launch your first cluster" guide; Instant Clusters are launched "from the RunPod dashboard with a few clicks". Two citations on that page link to the blog with "?utm_source=chatgpt.com" appended.Source
    • Your websitePricing page: the table says H100 SXM "Per GPU/hr ~$3.29" vs hyperscalers "~$7-13"; the cards read "H100 Cluster $3.29 per hour" (8x H100), "H200 Cluster $4.59 per hour", "B200 Cluster $6.79 per hour" and "B300 Cluster $7.89 per hour", each above an "8x" spec list. Every card's "Rent now" button links to www.runpod.io/, the partner's generic homepage. Reserved 1-5 year rates are quoted on request.Source
    • Your websiteHomepage: "Stop overpaying for GPU compute. Up to 70% cheaper than hyperscalers on the same NVIDIA GPUs. Per-second billing, sub-5-minute deployment, no commitments." "Spin up dedicated GPU instances on-demand through RunPod. No quotas, no approval queues, no sales calls." The comparison strip shows an H100 hourly rate of "~$2.99" against "$7+". Main buttons: "Deploy now" and "Book a demo".Source
  • 21

    FarmGPU sells sovereign, confidential infrastructure, and security-minded buyers run tools such as Mozilla Observatory on vendors. The marketing site grades C- and sends no Content-Security-Policy.

    Medium impactLow effort First result in 2 weeks Head of Engineering

    First action

    Add the headers in Cloudflare in report-only mode.

    How to measure it

    Observatory grade

    Now
    –
    Target
    A
    By
    1 month

    Steps

    1. 1

      Inventory scripts

      Webflow, Google, reCAPTCHA.

      Head of Engineering 1 day

    2. 2

      Report-only CSP

      Cloudflare rule.

      Head of Engineering 1 hour

    3. 3

      Review reports

      One week.

      Head of Engineering 1 week

    4. 4

      Enforce

      Switch on.

      Head of Engineering 1 hour

    5. 5

      Other headers

      Referrer and permissions.

      Head of Engineering 1 hour

    6. 6

      Retest

      Observatory.

      Head of Engineering 15 minutes

    First result
    2 weeks
    Cost
    $0
    Owner
    Head of Engineering

    Smallest test

    Run the report-only header for a week.

    Why it works

    Add standard headers through Cloudflare (Content-Security-Policy, referrer and permissions policies), test that forms and analytics still work, and retest.

    It removes an easy objection from the most careful buyers at almost no cost.

    Risks and how to handle them

    • Broken widgets

      Use report-only first.

    Evidence

    • DNS and security checkfarmgpu.com uses Google Workspace mail with SPF "~all" and no DMARC record. Mozilla Observatory grades the site C- (45/100) with no Content-Security-Policy header. The company is not registered in SAM.gov (the US federal supplier register).Source
    • Your websiteCompany page: "The Neocloud vision" - "a network of local, auditable facilities so you know where your workloads run"; Sovereign ("even we cannot see your workloads") and Sustainable pillars; founders from Intel data center SSD strategy, IT asset disposition and cloud ventures. Main navigation: Platform, Infrastructure, Performance, Pricing, Company, Docs, Blog, Products; no page for GPU owners, investors or data center operators.Source
  • 22

    The headline changed from "Leaders In AI Infra" to "Stop overpaying for GPU compute", so price is now the brand. FarmGPU already compares against hyperscalers and models offtake prices in its deal desk.

    Medium impactLow effort First result in 1 quarter CEO and Co Founder

    First action

    Collect this month's public prices into a sheet.

    How to measure it

    Monthly visitors

    Now
    –
    Target
    3,000
    By
    2 quarters

    Steps

    1. 1

      Pick providers

      Ten clouds.

      Marketing Manager 1 day

    2. 2

      Collect prices

      Monthly.

      Marketing Manager monthly

    3. 3

      Publish

      Chart and method.

      Marketing Manager monthly

    4. 4

      Promote

      LinkedIn and press.

      CEO and Co-Founder monthly

    5. 5

      Add signup

      Index alerts.

      Marketing Manager 1 day

    6. 6

      Review

      Traffic and leads.

      Marketing Manager quarterly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    CEO and Co Founder

    Smallest test

    Publish one month's table.

    Why it works

    Publish a monthly index of public per-GPU prices (H100, H200, B200, B300) across major clouds and neoclouds, with FarmGPU's own rate, the method and a chart. Share on LinkedIn and with press.

    A recurring data asset earns links, press and repeat visits from both buyers and investors.

    Risks and how to handle them

    • Seen as biased

      Publish the method and include FarmGPU's own price change history.

    Evidence

    • Site historySince October 2025 the site moved from Framer to Webflow and the headline changed from "Leaders In AI Infra" to "Stop overpaying for GPU compute.": the positioning has shifted from infrastructure credibility to price.Source
    • Competitor researchRivals' own pages: Runpod lists H100 SXM at "$3.99 /hr", H200 at "$5.29 /hr" and B200 at "$7.99 /hr" per GPU on demand; Lambda lists 1-Click Clusters of "16 to 2,000+" B200 or H100 GPUs, B200 at "$9.86" per GPU-hour for 2 weeks to 1 year; Nebius promises "From zero to clusters in minutes, with built-in repeatability and self-service"; Crusoe calls itself "The AI factory company".Source
    • Your websiteThe homepage "Lowest TCO / 70% Less Than Hyperscalers / Calculate savings" card links to calculator.farmgpu.com, which opens a "Deal Desk" to "Price an offtake for 512 x H100 SXM": server price, debt share, cost of equity, colocation per kW. Its default view shows "Project IRR -9.0%", "Equity multiple 0.44x" and "Payback Never". It is a GPU-investor model, not a buyer savings calculator.Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

ProgrammableInfo

Turn the research into a list you can act on

These ideas start with finding the right companies. Info finds them and the right person at each, researches every one and drafts the outreach, then asks before it sends.

  • Build this list in Info for idea #14 (opens programmable.info in a new tab)

New accounts: 100,000 free credits with code BETA100K

Find the companies (opens programmable.info in a new tab)
More example playbooks and reviews
ProgrammableDigital

Show FarmGPU’s product working

Turn screenshots or the website into an animated product demo in its own brand, for the homepage, launch posts and outreach in this playbook.

Make a demo (opens programmable.digital in a new tab)

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • MangoDesk growth strategyBuilds production-grade, long-horizon RL environments and custom evals that measure and improve AI on real-world software and…22 growth ideas
  • Sitemate growth strategySoftware and AI agents for the built world: Storm (an AI agent that turns voice, photos and video into structured records), Dashpivot…23 growth ideas
  • Budli growth strategyAn Indian recommerce company that buys used phones, laptops, tablets and gadgets with instant quotes and free doorstep pickup, and sells…22 growth ideas
  • PragerU growth strategyA donor-funded 501(c)(3) nonprofit media organization that publishes free short educational videos, shows, email courses, children's…22 growth ideas
  • Thesis growth strategyDirect-to-consumer nootropic supplements: four goal-based formulas (Clarity, Motivation, Stress Reset, Neuroprotection) chosen through an…23 growth ideas
  • Mohawk General Store growth strategyA Silver Lake, Los Angeles boutique founded in 2008 with men's and women's shops, selling designer clothing, shoes, jewelry, home and…21 growth ideas
  • Kapitus growth strategySmall business financing, both as a lender and as a marketplace (KapitusPLUS): business loans, lines of credit, revenue-based financing…21 growth ideas
  • Rhombus Power growth strategyAI-powered predictive intelligence and decision support for defense and national security: Guardian for resourcing and strategic…22 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on November 4, 2026, and cites them throughout. FarmGPU did not commission, review or endorse it, and Programmable is not affiliated with FarmGPU. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how FarmGPU runs its business.

Something wrong, or do you represent FarmGPU? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 306 ideas

    Fix the deploy path, the savings button, the H100 price, the footer, the sitemap and the uptime wording.

  2. Days 31 to 6010 ideas

    SOC 2 status, quickstart, benchmark trial, comparisons, case studies, the operator page, newsletter, DMARC, phone speed and the confidential agents waitlist.

  3. Days 61 to 906 ideas

    Commit to public-sector compute, the KV-cache tier and the open neocloud playbook; test the assistant, security-header and price-index ideas.

The research behind the plan

2 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value dedicated nodes that "just work". The most cited problem: hyperscaler prices, quotas and lock-in.

  • Storage and data loading that keep GPUs busy
  • GPUs waiting on data
  • In their words: “neocloud”

Market

Competitive landscape

5 competitors compared: Runpod, Lambda, Nebius, Crusoe and 1 more.

  • Gap to exploit: Storage-first clusters for data-heavy training and long-context inference
  • Runpod: Shared marketplace capacity
  • Lambda: Higher cluster prices

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks first (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence