Two components, both visible. Your licence never exceeds $200,000 a year, however large your fleet grows. Once it reaches that ceiling it stops — every additional camera carries only a $1/month support charge. Compute is charged at what the analytics actually cost to run, and locked for your contract term.
No base platform fee. No usage charges. No surprises on the invoice.
Compute falls when you commit, because ours does — we buy reserved GPU capacity for the length of your contract and pass the saving straight through. The licence never changes.
Identical analytics and support. The only question is who buys the compute.
We run the GPUs, the storage and the image history. Nothing to rack, nothing to patch, no cloud bill of your own. Your compute rate is fixed for the whole term.
Every use case, every capability, running on your servers or at the edge. You supply the GPU nodes and storage — we size them for you below. Licence caps apply the same way.
Built from use cases deployed across live airport terminals — and equally applicable to transit hubs, logistics yards, stadiums, hospitals and large retail. Every site is a different combination. Enter cameras against the jobs you need done — pricing and infrastructure follow from the mix. Licence is charged per camera up to $200,000 a year, then it stops. Past that ceiling every further camera costs $1/month in support and nothing more.
Getting live is a one-time integration project — networking, VPN, VMS connection, stream onboarding and accuracy validation. It is priced as a project, not per camera, because that is what it actually is.
Flat fee, whatever the fleet size. Covers network integration, secure stream onboarding, model tuning and validation against a manual baseline before go-live.
Ten cameras, your real footage, your real use cases. We stand it up, tune it and validate the numbers against your manual counts. No obligation, nothing to install.
Standard is included. Upgrade if you need faster response, phone cover or a named team — priced as a percentage of your subscription, so it scales with your fleet instead of a flat fee that punishes small sites.
The things competitors charge extra for.
Everything a procurement team asks before signing.
Because they behave differently and you deserve to see that. The licence is software — it costs us the same whether you run 10 cameras or 1,000, so it caps as you scale. Compute is real GPU time that grows with every camera and every frame, so it doesn't cap. Splitting them means you can see exactly what you're paying for, and it's why we can offer a cap at all — most vendors can't, because their single blended rate has to cover compute forever.
Only where the compute genuinely differs. Counting people, timing a queue and measuring dwell all run the same single-model pipeline at about 5 frames per second, so they sit in one category at one price — we do not invent tiers to create upsell. Safety alerting runs at 10–15 frames per second with evidence capture. Asset ID adds a second recognition model on 4K crops. Those two cost materially more to run, so they are priced separately. That is the whole rationale.
Because on the licence option you buy and run the GPU servers, and a price comparison that ignored that would be dishonest. The figure is our estimate of what those nodes and their storage cost you per year — we do not bill it and never see it. It is there so you can compare the two options on total cost rather than on our invoice alone. On Eye4 Hosted that line is zero, because the compute is inside the price you already pay us.
The licence stops growing. Support inside the ceiling is already included and stays included. What continues past it is a small support charge: $1 per camera per month, whichever option you take. It is billed in blocks of 100 cameras, because support is hired in people rather than fractions. One rate, no asterisks.
Because ours does. We buy reserved GPU capacity for the length of your contract, which costs substantially less than on-demand, and we pass that straight through. It also means your compute rate is locked for the whole term — it cannot rise, even if cloud prices do. The annual adjustment applies to the licence only.
Yes — your licence is capped at $200,000 a year, permanently, not as a promotion. On a three-year term that ceiling is reached at roughly 1,100 cameras; every camera beyond it carries only the $1/month support charge, no further licence at all. Compute is not capped, because the GPUs genuinely have to be there for every camera — but it is the component we charge closest to cost, and it drops when you commit to a term.
As a one-time project, not per camera. Getting live is network integration, VPN, VMS connection, secure stream onboarding and validating accuracy against your manual counts — that work does not multiply by camera count, so we do not bill it that way. It is a flat $12,000, half price on an annual term and waived entirely at three years.
Minimum is 10 cameras. Most customers start with one site or zone, prove the numbers against a manual baseline, then expand. Licence caps apply automatically across the whole fleet as you grow — you never have to renegotiate to get them.
The two components earn very differently. Licence is software margin; compute is close to pass-through.
| Category | Cams | Wt | Lic billed | Lic COGS | Lic GM | Compute chg | Compute COGS | Compute GM | Total GM |
|---|
Blended hosted rate against the market, USD/camera/month equivalents.
| Vendor | Effective rate | vs Eye4 | Note |
|---|
Which option to steer toward, and what support actually costs on each.
| Metric | Hosted | Licence |
|---|
Licence rises with CPI from year 2; compute is locked for the term; cost rises with input inflation.
| Period | Revenue | COGS | Gross margin | GM % |
|---|
| Cost line | Per camera | Monthly total | Share |
|---|
Identical either way — the only question is who pays for it.
Indicative planning figures. Verify cloud rates and measure setup effort before committing to price changes.
Price has exactly two components. The licence is flat across every use case and caps as the fleet grows. Compute scales with the use-case weight and never caps, because GPU cost is linear in cameras forever. That asymmetry is the whole design — it is what lets us publish a cap at all without the deal going underwater at scale.
hosted_cam(cat) = licence + compute_base × weight(cat) × term_compute_mult
licence_cam = licence (flat, all use cases)
licence_billed = min(cams, 120) per use case — hard cap
∧ min(total, 600) account-wide — hard cap
compute_billed = cams never capped
CLU_total = Σ (cameras_cat × weight_cat)
nodes = ceil(CLU_total ÷ (node_CLU × utilisation))
GPU_cam = node_cost ÷ (node_CLU × utilisation) × weight_cat
revenue_yr(k) = licence × (1+CPI)^(k−1) + compute # compute locked for term
COGS_yr(k) = COGS_cam × cameras × 12 × (1 + cost_inflation)^(k−1)
All 27 use cases in the portfolio map to exactly one of these seven.
On the term ladder. The compute discount is a genuine pass-through of reserved-capacity savings, not a concession — which is why it is defensible in negotiation and why the rate can be contractually locked for the term. The licence is not discounted by term at all; it is discounted by scale, through the cap.
On margin split. Licence carries software-grade margin (75%+). Compute carries 39–45% at the stated base, and falls toward zero if we price it at true cost. Steer discounting to compute and protect the licence — the reverse destroys the business.
On hardware. These figures assume L4-class GPUs under a reserved commitment. L4 delivers roughly 2.2× a T4's throughput in the same 70 W envelope, cutting cost per CLU by more than half. Switch the GPU selector to T4 on-demand to see compute margin go negative at a $24 base.