I started by modeling storage. That was the wrong center of gravity for a SaaS product that delivers terabytes: capacity matters, but every recipient download can create another full egress charge. Cloud list prices can establish an illustrative cost floor. They cannot tell me what the customer's deadline, control, or reduced operational risk is worth—and they do not determine TeraAirlift's final price.
TeraAirlift billing remains under evaluation. The examples below are a founder's pricing process, not a price announcement, quote, audited cost of goods sold, revenue disclosure, or claim about current margins.
First lesson: separate every cost meter
The initial model treated one terabyte as a storage problem. On August 10, 2026, the Azure Retail Prices API listed East US pay-as-you-go Hot LRS capacity at $0.0208 per binary GB-month. Under a seven-day retention assumption, one decimal TB—about 931.32 GiB—would cost roughly $4.52 in capacity.
The same date's Azure bandwidth page listed internet egress from North America/Europe over the Microsoft premium global network at $0.087 per decimal GB for the first paid 10 TB band, after the first 100 GB per month free. If that free allowance is already consumed elsewhere, sending one decimal TB once creates $87 of egress.
That changed the model. A delivery product needs separate inputs for:
- storage capacity by class, region, and redundancy;
- write and read operations;
- retrieval and early-deletion charges;
- internet, inter-region, CDN, and private-path transfer;
- recipient count and redownload behavior;
- failed or repeated work;
- metadata, monitoring, support, and engineering; and
- billing, refunds, disputes, tax handling, and fraud.
Cheap capacity does not rescue a mismatched lifecycle. Azure recommends minimum retention periods of 30 days for Cool, 90 for Cold, and 180 for Archive, with early-deletion charges. Archive is offline. Those tiers may suit long-lived, rarely accessed data; they are not automatically economical for short-lived delivery staging.
Build a floor, then stress it
Here is one illustrative floor using August 10, 2026 USD list prices: Azure East US, pay as you go, Hot LRS, seven-day retention, premium global-network routing from North America, one complete recipient download, no CDN, no tax, and the 100 GB allowance already used.
| Component | Illustrative cost per decimal TB delivered | |---|---:| | Internet egress: 1,000 GB × $0.087 | $87.00 | | Capacity: 931.32 GiB × $0.0208 × 7/30 | $4.52 | | Upload writes: about 119,210 requests using 8 MiB blocks | about $0.60 | | Download reads: about 119,210 requests at the same request size | about $0.05 | | Cloud subtotal | about $92 |
The request count uses a decimal 1 TB payload: 1,000,000,000,000 bytes divided into 8 MiB (8 × 2²⁰-byte) blocks requires about 119,210 requests after rounding up. The displayed components sum to about $92.17 before rounding the subtotal to about $92. This is author arithmetic from public list inputs. It excludes support, engineering, monitoring, metadata, retries, re-downloads, free usage, payment costs, discounts, refunds, and taxes. It is neither TeraAirlift's actual architecture bill nor its audited economics.
More importantly, the floor moves. Two recipient downloads roughly double the egress component while capacity stays unchanged. The first paid Azure premium-network band displayed on August 10, 2026 was $0.181/GB from South America, compared with $0.087/GB from North America/Europe. Larger monthly bands step down. Region, route, recipients, volume, retention, and contract can all change the result.
Second lesson: cost is not value
Cost-plus pricing answers, “What price avoids losing money under these assumptions?” It does not answer, “What outcome is worth paying for?”
A production team may value deadline visibility, controlled recipients, transfer history, integrity verification, and an operator who can see which job needs attention. Another buyer may care mostly about predictable procurement. A third may need occasional burst capacity and reject a large recurring commitment.
Those are hypotheses to test through demos and buyer conversations. Infrastructure cost creates a boundary; customer value, alternatives, procurement, and willingness to pay shape the offer. A competitor's published rate does not reveal its infrastructure, contract, support burden, utilization, or margin.
MASV, for example, published a pay-as-you-go offer on August 10, 2026 with 15 GB per month included and then $0.25 USD/GB when data exits its platform. It also listed five days of storage with an upload and $0.07 USD/GB-month for extended storage. Those are vendor list terms, subject to account, subscription, pre-purchase, enterprise, currency, tax, and negotiated changes—not evidence of MASV's margin and not a TeraAirlift benchmark.
Third lesson: plan shape allocates risk
Each pricing structure puts usage variance somewhere:
- Flat subscription: predictable for the buyer, but the vendor absorbs heavy-use risk.
- Pure metering: aligns revenue with transfer volume, but creates invoice uncertainty.
- Included allowance plus overage: gives a budget baseline while preserving a cost boundary.
- Prepaid credits: creates a committed pool for project-based usage, with questions about expiry and accounting.
- Base fee plus usage: separates access or operational value from variable delivery cost.
- Negotiated volume: can fit large, steady workloads but adds sales and forecasting complexity.
Choosing one is not only a pricing-page decision. Metering needs a defined event, unit, deduplication policy, late-event behavior, customer-visible records, dispute handling, thresholds, credits, and invoice reconciliation. “Per TB” is ambiguous until the contract says decimal TB or TiB and whether uploads, downloads, retries, and multiple recipients count.
What I am validating
The open questions matter more than a polished number:
- Can buyers forecast monthly delivered volume, or is work project-spiky?
- Which unit best matches value without confusing the invoice?
- How much predictability does procurement require?
- How should legitimate redownloads and multiple recipients be represented?
- Which costs vary by geography and which can be pooled?
- What usage visibility should exist before an overage?
- Which features belong to the service relationship rather than the usage meter?
Current site wording is deliberately narrower: plans start with a 10 TB included allotment, packages can be any size, and volume above that allotment is overage. That is not a final currency price, overage rate, meter definition, cadence, or SKU design. Details are confirmed on a demo.
Before modeling economics, size the actual delivery path with the transfer-time calculator and compare one-off physical delivery using the shipping-versus-transfer calculator.
Sources
- Azure Bandwidth pricing — dated USD list bands, allowance, routing, and regional sensitivity; accessed August 10, 2026.
- Azure Retail Prices API — Storage in East US — illustrative capacity and operation inputs; accessed August 10, 2026.
- Azure Blob Storage pricing — tier, binary-unit, retention, retrieval, and early-deletion context; accessed August 10, 2026.
- MASV pricing — published usage and temporary-storage terms; accessed August 10, 2026.
- Stripe usage-based billing — meters, usage recording, and credit concepts; accessed August 10, 2026.
