Data Infra › AI Infra & Compute Economics

Data Infra · 04

AI Infra & Compute Economics

The economics of the AI buildout as 11 clean panels: LLM prices, GPU rents, token demand, audited hyperscaler capex. Every row carries its source URL and a quality tag, so your analysts cite the primary source, not us.

Every row passes the Bargo Verification Layer

Accurate · Reliable · Source‑backed

11
hand-curated panels
2019→
audited hyperscaler capex from SEC filings
2023→
LLM price history, every cut
100%
of rows carry a source URL and quality tag

Eleven panels · eleven endpoints

Every panel is its own API.

Call one panel or join them all. Every endpoint returns dated rows that carry source_url and quality_tag (company-disclosed / official / press / estimate), so what you pull is what you can cite.

Pricing & demand
GET/v1/ai-econ/llm-prices

LLM API price history

Per-model pricing per million tokens for the major labs, every price cut since 2023.

dated series · on every price change
GET/v1/ai-econ/token-usage

Token usage disclosures

Public token volumes and usage: monthly tokens, throughput, MAU/WAU/seat counts.

dated series · on disclosure
GET/v1/ai-econ/gpu-spot

GPU spot pricing

Market rental rates for AI accelerators (H100 and others) over time.

dated series · continuous
GET/v1/ai-econ/inference-costs

Inference economics

Cost-to-serve estimates per model tier and token demand mix.

dated series · on model & price shifts
Buildout & hardware
GET/v1/ai-econ/hyperscaler-capex

Hyperscaler capex panel

Audited quarterly capex from SEC filings since 2019, segment revenue, guidance, neocloud backlogs.

quarterly · 2019→ · company-disclosed
GET/v1/ai-econ/hardware

Hardware specs & shipments

Accelerator datasheet constants plus shipment anchors from Omdia, TrendForce and companies.

dated series · on report
GET/v1/ai-econ/power

Data-center power model

Power requirements and constraints behind the AI buildout.

model · on disclosure
GET/v1/ai-econ/cloud-tco

Cloud TCO model

Total-cost-of-ownership comparisons for AI compute.

model · on pricing change
Frontier & disruption
GET/v1/ai-econ/capabilities

AI capabilities panel

Frontier benchmarks with saturation, training compute and cost, open-source lag, cost decline at fixed capability.

dated series · on model release
GET/v1/ai-econ/saas-disruption

SaaS disruption evidence

Documented cases of seat-based pricing moving to consumption pricing under AI pressure.

cases · per documented case
GET/v1/ai-econ/private-ai-revenues

Private AI company revenues

Press-reported revenue run-rates for private AI labs and infrastructure companies.

dated series · tagged estimate

The data, not a description of it

A row you can check yourself

Every panel row cites the primary source it came from, with an honest quality tag: company-disclosed, official, reputable press, or estimate.

■ llm_price_change · illustrative format
provider / modelmajor AI lab · frontier tier effective_date2026-07-15 price_per_1M_tokens$3.00 → $1.90 change-37% · computed source_urlthe provider’s own pricing page, archived at capture source_qualitycompany-disclosed verifyopen the source URL; the figure is printed there
Field names simplified for display. Full data dictionary ships with every evaluation snapshot.

Bargo Verification Layer · Accurate · Reliable · Source‑backed

Watch one price cut survive the three gates

A lab cuts its API price. Here is what happens before that number enters the panel:

Gate 1 · The data

The figure exists at the source

Every row must trace to a primary source, and the figure must actually appear there. Pricing pages are archived at capture, because they change and vanish.

✓ $1.90/1M printed on the provider’s pricing page · archived
a number from a tweet thread · enters only as “estimate”, clearly tagged, or not at all
Gate 2 · The method

Quality is a field, not a vibe

Company-disclosed, official, reputable press, estimate: the tag rides on every row, so audited capex is never silently blended with rumor.

✓ audited 10-Q capex tagged company-disclosed · weighted accordingly
✗ press estimate presented as disclosed fact · never; the tag travels with the row
Gate 3 · The answer

Trends are computed, not narrated

Cost-decline curves, $/token trajectories and capex growth are arithmetic on the panel, never an impression of where things are heading.

✓ -37% computed from the two archived prices
“prices are falling roughly 40%” · from memory, never served

Only then does the row join the panel, carrying its source, its tag and its archive. Facts scattered across hundreds of disclosures, assembled into something you can regress. See the full Verification Layer →

Why it compounds: these panels are assembled from hundreds of primary disclosures since 2019, many of which no longer exist at their original URLs. The archive of sources captured at the moment of disclosure cannot be rebuilt after the fact.

Access

The same panels, three ways

Ask it anything

Bargo Agent

AI-economy questions with receipts: prices, capex, capacity and who captures the margin.

> How fast are inference prices falling, and who keeps the margin?
For AI agents & terminals

Bargo MCP / CLIs

Wire it into Claude, Cursor or your own agents, or pull straight into pandas.

$ bargo panels pull llm-prices \ --since 2023 --format parquet
Programmatic

Bargo APIs

All 11 panels as clean endpoints, source URL and quality tag on every row.

GET /v1/ai-econ/llm-prices GET /v1/ai-econ/hyperscaler-capex

Who runs on it

Built for the AI trade’s input side

Portfolio managers

The capex cycle, positioned

Audited capex, guidance and backlogs in one panel: where the buildout is accelerating and where it is not.

Analysts

Cite the primary source

Every number in your note traces to a disclosure, not to a data vendor’s black box.

Quants

Panels born joinable

Clean dated series that key to tickers where issuers are listed, ready for cross-asset work.

Coverage & delivery

The facts sheet

Coverage11 panels spanning prices, usage, hardware, capex, power, TCO, capabilities and disruption. Global where the supply chain is global, anchored to US-listed issuers.
CadenceRows added on disclosure events: earnings, price changes, shipment reports.
HistoryCapex from 2019; LLM pricing from 2023, the full commercial era; other panels from 2023-2024 onward.
FormatsJSON over API, CSV / Parquet snapshots, digest-stamped.
IdentifiersTickers and CIKs wherever the row concerns a listed issuer.
VerificationEvery row carries its source and its checks. Corrections are versioned, snapshots digest-stamped. How the Verification Layer works →
We would rather show you the data than describe it.
Request an evaluation snapshot