Cost and Funding

Nothing in Demostatics has a price attached to it — not the pipeline, not the product, not the licences. This page puts the first numbers on paper so the plan can be argued about in money as well as in months.

Every figure on this page is an estimate unless it says otherwise

Only four numbers here were read out of a file: the desktop client’s dev allowance limits, the 5,000-row export limit, the roadmap’s “9 to 18 months”, and the pitch deck’s ten slides. Everything else is an ESTIMATE with the reasoning shown next to it, and every one of them needs a real quote before a decision rests on it. Getting those quotes is days of work.

The state of play

No document, spreadsheet, config file, code comment or README in any of the five repositories attaches a cost to anything. There is no COGS model, no infrastructure sizing, no per-source price, no inference budget, no cloud estimate, no split between fixed and variable cost, no burn, no runway, no financial model, and no funding ask. Business Model discusses four revenue lines and never mentions cost once. A grep across all five repositories for currency amounts — $, €, ₺, USD, EUR, TRY, “per month”, “/mo”, “per seat”, “annual” — returns zero monetary figures; every $ hit is a PHP variable or Rust syntax. The one repository that would run and pay for all of this, demostatics-backend, is an 80-byte README and a .git directory. The pitch deck has ten slides and no numbers on any of them: no ask, no financials, no traction, no team, no market size.

Nothing is on fire, because nothing is running. The cost of this gap is not a bill today — it is that the architecture is about to be scoped against a source list nobody has priced, and that an investor or customer who asks “what does it cost you to produce a report” gets silence, which reads as the founder not understanding his own product.

Two scenarios, roughly 100x apart

The nine declared sources in Data Pipeline have two honest readings, and they describe two different companies.

Scenario A — the cheap path

Free and open sources at modest scale. All figures ESTIMATE.

Driver	Assumption	Estimated monthly
Ingest compute	One always-on VM	$150–250
Database	Managed Postgres/Timescale	$100–300
Object storage	Raw archive growing ~2 TB/mo at ~$0.02/GB	~$40 now, ~$500 by year two
Imagery	Sentinel-2/Landsat free from Copernicus/USGS; processing only	$100–500
LLM classification	~20k documents/day at ~1.2k input + 200 output tokens each ≈ 720M input tokens/mo, small/cheap model	$150–400
Per-provider API fees	Free and openly redistributable sources only	$0
Total		~$500–$5,000/month

The single biggest swing factor is model choice. Nothing else on that table moves the total by an order of magnitude; the model does.

Scenario B — the founding document taken literally

Google Maps + Yandex + Baidu + social media + commercial satellite + continuous LLM, at “worldwide” scale. All figures ESTIMATE.

Driver	Assumption	Estimated monthly
Mapping and traffic	1,000 corridors sampled every 15 min ≈ 2.9M Google Routes calls/mo at roughly $5/1,000 requests	~$14,000 — and “global road networks” is trivially 10x that
Social media	X/Twitter API at the Pro tier	~$5,000, tens of thousands at Enterprise
Commercial satellite	Tasking, or a broad-AOI Planet-style subscription	Five to six figures per year minimum
LLM inference	~500k documents/day	~$5,400 on a cheap model, ~$54,000 at frontier pricing
Self-hosted alternative	4x A100 fleet	$4,300–7,200
Ingest, database, storage	Scenario A’s rows at “worldwide” volume	Larger; not separately estimated
Total		~$50,000–$250,000/month, i.e. $600k–$3M/year before a single salary

The gap is the strategic question

Scenario A is fundable by one person out of pocket. Scenario B needs institutional venture capital before it produces a single row of data — and would need it raised on a deck that currently has no numbers in it. Nobody had noticed that the two readings of the same source list differ this much, which means the choice between them has never been made deliberately.

Three things make the gap smaller than it looks:

• The expensive part of “satellite” is probably not needed. Flight paths come from ADS-B, maritime from AIS, road networks from OpenStreetMap and mapping APIs. Land-use and ecological change is the one genuine imagery case, and Sentinel-2 and Landsat serve it at zero data cost. Commercial high-resolution tasking — the five- and six-figure annual contracts — is not required by any stated capability.
• Google Maps, Yandex and Baidu may be unusable at any price for what this business does with data, which is store it and resell it. That is a terms question, not a pricing question, and it needs checking before an integration is written.
• A defensible minimum viable data set exists entirely on free, redistributable sources. It is written out on The Minimum Viable Data Set.

Pick a scenario before writing ingestion code

The failure mode is building toward the founding document’s source list, signing up for Google Maps Platform, connecting a news firehose, and discovering the cost structure by receiving a bill in month one — at which point the pipeline is architected around sources that have to be ripped out.

Fixed versus variable

Almost every cost above is fixed. Satellite processing, news ingestion, LLM classification and storage run at the same cost whether there are 0 subscribers or 1,000. The genuinely per-subscriber variable costs are small: egress of already-computed rows, and cloud AI-assistant tokens.

That makes Demostatics a high-fixed-cost, near-zero-marginal-cost business, so the whole thing reduces to one question — how many seats amortize the pipeline — and that number has never been computed.

The arithmetic is trivial once COGS exists. ESTIMATE, on the cheap path: ~$2,000/mo COGS plus one modest founder salary, call it $5,000/mo all-in. Then $200/seat/month needs 25 paying seats, or $2,000/seat/year needs 30 seats. That is a plausible target for a specialist product. On the literal-founding-document path at ~$100,000/mo, the same margin needs 500 seats at $200/mo, or 40 seats at $30,000/year — Bloomberg-adjacent pricing, sold by a one-person company with no track record. The first is a business. The second is not.

Note also what the roadmap’s headline estimate leaves out: “9 to 18 months for one full-stack engineer” is expressed purely in time, and 9–18 months of runway at any salary is a number that has never appeared anywhere. See Roadmap.

The only allowance model that exists meters the wrong things

The one place in any repository with concrete allowance numbers is a mock server, and it inverts the cost structure. From demostatics-pc_application/examples/mock-stream-server/src/session.rs:

Tier	data (MB)	compute (credits)	ai (tokens)
Free	1,000	50	—
Individual	10,000	500	—
Pro	100,000	5,000	200,000
Enterprise	0 = unlimited	0 = unlimited	200,000

data is egress, which is cheap. compute is work performed by ds-compute, a rayon pool running on the customer’s own PC — Demostatics does not pay for it and cannot verify it, and billing a customer for their own laptop’s electricity is hard to hold in a renewal conversation. ai tokens are the one genuinely expensive per-user resource, and the AI meter is attached only when the AiAssistant feature is granted.

What the model does not meter at all is the thing that costs almost everything: source breadth and refresh frequency. Those figures are labelled “dev limits”, but they are the only allowance numbers that exist anywhere, so they become the real ones by default unless someone replaces them.

Nothing caps the spend

There is no rate limit, per-account quota, spend cap, budget alarm or circuit breaker anywhere. GET /api/v1/reports and GET /api/v1/database-items carry no throttle middleware; the export endpoint returns up to EXPORT_LIMIT (default 5,000) rows per call with no cap on the number of calls and no record that an export happened. On the desktop side, crates/ds-ai/src/http.rs defines a CloudProvider that POSTs to /v1/ai/chat on a backend that does not exist, with no retry policy, no quota and no metering behind it.

For a pre-revenue firm a runaway bill is a solvency event

An LLM proxy has unbounded cost per request — image attachments, long context and agentic tool loops are all in the designed request shape — and it is exposed to a desktop client the firm does not control. Whatever spend cap is chosen, it has to exist before the first paid API key does. See Risks.

What licensing costs, beyond time

Regulatory Posture treats licensing entirely as a clock — months to years, and no amount of engineering shortens it. That is true and it is half the picture. Authorisation also has a price, and the cost items below exist in essentially every regime:

Item	Nature	Applies to
Minimum regulatory capital	Held and maintained, not spent	Lines 2 and 3
Compliance officer	A salary, not a checkbox	Lines 2 and 3
Custodian	Ongoing fee	Line 3
Fund administrator	Ongoing fee	Line 3
Auditor	Annual	Line 3, and the entity generally
Professional indemnity insurance	Annual premium	Lines 2 and 3
Counsel	Hourly, front-loaded	All lines

No threshold appears on this page on purpose

Minimum capital requirements, which licence carries which threshold, and whether advisory and discretionary management differ all vary by jurisdiction and by licence type. The jurisdiction is UNDECIDED. Counsel gives the numbers; nothing in this repository should be used as a substitute. Record the answers on Regulatory Posture once you have them.

The structural point survives without any threshold: line 3 cannot be started with zero capital regardless of how long you wait. Waiting out the clock does not produce the money.

And line 4, proprietary trading, is not a revenue line in the sense that matters here. It consumes capital, produces volatile returns rather than recurring revenue, and cannot be sold to anyone. No document anywhere says where the trading book’s money comes from.

The cost of being sellable, not just licensable

The four named buyer segments — investors and analysts, risk professionals, research institutions, policy and strategy teams — are the four with the heaviest procurement. Most will require a security questionnaire, a DPA, an SLA and, from financial institutions, usually SOC 2 Type II or ISO 27001. ESTIMATE: roughly $20k–50k for a first SOC 2 including tooling and audit, on a 6–12 month timeline. That figure needs quoting, but its existence is not in doubt, and none of those artifacts exists today. See Security and Privacy Posture.

The cheapest capital available is not mentioned anywhere

No document mentions TÜBİTAK, KOSGEB, technopark/Teknokent residency or Law 4691 R&D incentives. This is specific rather than generic advice: the founder is at Gebze Technical University, which operates a technopark, and the company is a software-and-R&D business with an ML/LLM pipeline at its centre — close to the exact profile those programmes fund. Technology Development Zone status also exempts qualifying software revenue from corporate tax and R&D personnel from income tax, which for a subscription business is a margin effect rather than a rounding error.

Amounts and current terms need checking, and eligibility depends on the entity structure — which is being decided right now, for regulatory reasons, on Regulatory Posture Question 2. Decide the tax and grant dimension at the same moment. Get it wrong and the exemption needs a restructuring to recover, and equity gets given away for money that was available for free.

Revenue sequencing by line

No document states which line produces revenue first or funds the others. Working it through:

Line	Blocked on	Earliest realistic revenue	Nature
1 — Data and intelligence	A price, a billing rail, and a pipeline with rows in it	Soonest — months, not years	Recurring
2 — Investment advisory	A licence	After authorisation, months to years out	Fee-based
3 — Discretionary management	A licence, plus custodian, administrator, auditor, and an audited track record institutional allocators will accept	Three-plus years, even after a licence lands	Fee on AUM
4 — Proprietary trading	Capital	Never, in the recurring-revenue sense	Consumes capital

So for at least the next three years, 100% of realistic recurring revenue comes from line 1 — the line that today has no price, no billing, no ingestion and no customer. The founder decision of 2026-08-01 added regulatory burden, conflict-of-interest engineering, record-keeping obligations and capital requirements to a company whose only revenue engine is unbuilt.

This is not an argument against the four-line strategy. It is an argument that the strategy needs a stated sequence, and the sequence is almost certainly: line 1 funds everything, for years, alone.

Line 1 cannot earn until something can take money, and the named rail may not work

There is no billing code, no plans/subscriptions/invoices tables, no payment integration, no tax handling and no currency decision. The roadmap names Stripe. Stripe does not support businesses incorporated in Türkiye, and the only address in any project document is in Gebze, Kocaeli. The realistic options are a merchant of record (~5% but absorbs global VAT/sales-tax compliance), a domestic PSP (cheaper, leaves EU VAT/OSS and US nexus to you), or incorporating abroad. Check this before the entitlement system is designed around any one provider’s webhook model.

Competitor prices as anchors

No competitor is named anywhere in any repository. The incumbents’ prices matter because they anchor what the buyer already pays for adjacent tools — which is the only external reference a first price has.

Approximate list prices, needing verification

Every figure below is APPROXIMATE and unverified, gathered as an order-of-magnitude anchor. Real enterprise pricing is negotiated, bundled and rarely public. Verify before quoting any of it to anyone.

Vendor	Approximate list price	Note
Bloomberg Terminal	~$32,000/user/year	The reference point everyone knows
LSEG Workspace (ex-Refinitiv Eikon)	~$12,000–25,000/user/year
FactSet	~$12,000/user/year
S&P Capital IQ Pro	~$13,000–30,000/year
Recorded Future	commonly quoted $60k–150k/year	Enterprise risk intelligence
Dataminr	commonly quoted $60k–150k/year	Real-time event detection from worldwide news and social media — closest to what Demostatics describes

Three conclusions nobody has written down:

1. A one-person company cannot win on breadth against firms with hundreds of data engineers and exclusive licensed feeds. The only viable entry is narrow — one region or one sector, covered better and cheaper than a generalist covers it.
2. A self-serve $10–200/month prosumer tier exists in this market (MarineTraffic proves it), and it is a far more realistic first revenue than a $30k enterprise seat that requires procurement.
3. “We aggregate worldwide news” is not a product — GDELT gives structured worldwide news events away for free. The refinement layer has to be the product, and it has to be demonstrably better than free.

Worth naming the cautionary case: Orbital Insight built satellite-derived economic signals for investors, raised heavily, and was sold in 2024 having failed to scale the business. “Satellites plus ML produce investable signals” has been tried and is commercially hard.

What to do about it

In this order. The first three are days of work and they change the entire plan.

1. Get real quotes for the top three cost drivers — LLM inference at your assumed document volume, managed database, object storage. These are self-serve pricing pages and a calculator; no vendor conversation is required. That converts the Scenario A table above from estimate to fact.
2. Choose Scenario A or Scenario B, in writing. Everything downstream — architecture, price, funding, whether you need investors at all — follows from that one line.
3. Write the minimum viable data set and price it. It is drafted on The Minimum Viable Data Set; confirm the licence terms for each source before committing to any of them.
4. Compute the break-even seat count at two or three candidate prices, using the real quotes from step 1 plus whatever founder salary is honest. If it is 25 seats, you have a business; if it is 500, redesign the cost base rather than the sales plan.
5. Start the archiver. An ESTIMATE of under $100/month buys a cron job writing raw source files to object storage with a timestamp and a source tag. History cannot be bought back later, and it is what a quantitative buyer asks for first. This is the cheapest item on the page and the most time-sensitive.
6. Put a spend cap and a budget alarm on every paid dependency before the first API key is issued.
7. Ask counsel for the capital requirements, per licence, in the jurisdiction you are most likely to choose, at the same time as the entity-structure question. Ask about technopark residency in the same conversation.
8. Then, and only then, build the funding instrument — burn, runway, use of funds, and an ask. There is nothing to raise against until steps 1–4 exist, and the current deck cannot be used.

For where these sit against everything else, see What To Do Next and the full Gap Register.