QnTuM AI
QnTuM AI

What companies said, and when they said it.

> CORPUS ........ 101,083 DOCUMENTS · 120 TICKERS ......... [OK]
> QUOTE CHECK ... 675,779 / 689,944 VERBATIM (98.0%) ...... [OK]
> LINK GRAPH .... 2.2M EDGES · POINT-IN-TIME ............ [OK]
> INFERENCE ..... OWNED HARDWARE · LISBON · ON-PREM .... [ACTIVE]

A structured, source-attributed record of every valuable statement on an earnings call: what was claimed, by whom, on what date, with the figure the speaker actually gave — tagged to the technology it concerns, linked to the statements that corroborate or contradict it, and traceable to the exact characters of the exact paragraph it came from.

Semiconductors, AI hardware and the supply chain around them. Sold as a managed platform for funds, a quarterly report, and a weekly letter — all from one pipeline, run on hardware we own.

NVDA FY2027 Q2 call 2026-08-26 Jensen Huang, CEO
For Grace Blackwell, we are about $25 billion per gigawatt. For Vera Rubin, it is about $40 billion per gigawatt
Verbatim · Exceptional band n/324870 ↗   8-K ↗
Documents
101,083
Datapoints
689,944
Verified verbatim
98.0%
Adjudicated shortlist
44,329
Companies
120
Measured from the running system as at 2026-09-01.
Evidence, not a demo

One number, three companies, seven days.

Between 26 August and 2 September 2026, three managements priced the same quantity — what a gigawatt of AI compute is worth — from three sides of the trade. Nobody searched for this. The link graph connected the rows.

NVDA 2026-08-26 Jensen Huang
For Grace Blackwell, we are about $25 billion per gigawatt. For Vera Rubin, it is about $40 billion per gigawatt
The vendorn/324870 ↗
AVGO 2026-09-02 Hock Tan
what we are seeing per gigawatt is in the range of less than $30 billion, $20 billion-$30 billion per gigawatt. And we expect that to be very sustaining in that level
The other vendorn/690545 ↗
IREN 2026-08-27 Daniel Roberts
Recent three-year contracts are pricing in excess of $20 million per megawatt of IT load, paying back the compute investment in around 2 years.
The buyern/690998 ↗
> n/324870 follows n/30607 (2025-08-27) · n/23589 (2025-11-19) · n/15910 (2026-02-25) · n/8119 (2026-05-20)
> n/324870 corroborated_by n/690545 [AVGO] · n/690998 [IREN] ........ [INDEPENDENT]
> verification links open on the source paragraph for evaluation clients and subscribers
Two rules, both contractual

Everything above rests on two constraints written into the agreement.

Rule 1

A quote is verbatim, or it is flagged.

Every quotation is checked as a literal substring of the source document before it can reach a view. A row that fails is marked, never silently corrected, and never served as evidence.

675,779 of 689,944 pass · 98.0% · the remainder carry the flag
Rule 2

Nothing consults a row dated after the checkpoint it is answering.

Corroboration looks back, never forward. A statement is linked only to what was already on the record when it was made — which is what makes a backtest honest and a screen usable in real time.

Point-in-time by construction
Method

Six stages, each one measured.

The pipeline runs unattended, continuously, on hardware we own. Every stage leaves a number behind, and the numbers are the specification.

Ingest
Calls and news

Transcripts and a news feed, kept in a form where every later step can point back to the paragraph.

101,083 docs
Extract
Statements

Every specific, differentiated datapoint — the figure a CFO gives once and never files.

689,944 rows
Verify
Substring check

The quote is matched character-for-character against the source. Not a paraphrase.

98.0% verbatim
Adjudicate
Analyst rubric

Scored against an analyst's definition of what improves a bottom-up model; three bands.

44,329 shortlisted
Link
Before & around

Tied to the same company's earlier statements and to other issuers' calls — corroborate, contradict, narrow.

2.2M edges
Re-read
Headline check

The one-line headline is re-read against its own passage. Unsupported rows are excluded, reason shown.

reason shown on every exclusion
Location
Lisbon, PT
Architecture
On Prem / Private · EEA
Ground truth
4 yrs of annotated analyst work

Each client's own material sits in an isolated workspace, enforced in code, alongside the shared public corpus. Extraction is measured against human judgement, not asserted: a companion narrative-inflection classifier runs at 94.3% statement-level precision against analyst verdicts.

One production run, sold three ways

The same corpus, at three depths.

Funds with semiconductor and AI-hardware exposure; corporate strategy and IR teams inside the covered supply chain; individual analysts. Choose by how close to the rows you need to be.

Meet us in Lisbon

Bring a ticker. We will open the record on it.

Twenty minutes, one of your names, no slides. Clicking a quote and landing on the paragraph it came from is the whole pitch.

17Sep 2026

Startup Summit Lisbon

Unicorn Factory Lisboa, Beato Innovation District · 17–18 September.

9–12Nov 2026

Web Summit

MEO Arena & FIL, Parque das Nações · 9–12 November.

Contact

Researched, built and run in Lisbon.

Everything on this site is research, not advice. It records what company management said, on what date, with the figures they gave. It is not a recommendation to buy, sell or hold any security; no output is a price target or a rating, and no position in any company named here is held by QNTUM AI. Every figure attributed to a company is a verbatim quotation from a public earnings call or filing and carries a link to the passage it came from. QNTUM AI, Unipessoal Lda. is not authorised or regulated by the CMVM or any other financial-services regulator, does not manage assets, and does not execute.