Evidence-Backed Financial Intelligence
What companies say about AI — and what the record shows.
Connect what management promised with what subsequent evidence establishes. Ojontic helps equity analysts, portfolio managers, and investment researchers trace corporate AI strategy across disclosures—keeping quantified results, unresolved questions, and visible uncertainty in view.
Infrastructure spending is rising. How much of it is for AI?
Amazon and Microsoft both connect infrastructure investment to AI in their FY2025 Form 10-K filings. Neither separately quantifies the AI portion of that spending.
FY2025 Form 10-K filings · Amazon and Microsoft: verified analyses · Status labels
Same lens.
Different disclosures.
Amazon and Microsoft, compared across selected findings from their FY2025 annual reports on Form 10-K. Click a finding to see the passage behind it.
pretending to rank.
Swipe the table sideways to compare each company
| Finding | AmazonAMZN · December year-endFY2025 10-K · Verified analysis | MicrosoftMSFT · June year-endFY2025 10-K · Verified analysis |
|---|
Research cutoff: FY2025 Form 10-K for each company — Amazon’s for the year ended 31 Dec 2025 (filed 6 Feb 2026), Microsoft’s for the year ended 30 Jun 2025 (filed 30 Jul 2025). Every finding shown comes from a verified analysis of that filing. FY2025 is not Microsoft’s latest annual report: its FY2026 Form 10-K (filed 29 Jul 2026) has been acquired but is not yet annotated or adjudicated, so it is not reflected here. Alphabet’s FY2025 figures were checked by hand against its Form 10-K but not independently adjudicated, so they are not part of this verified comparison; preliminary screenings of other companies are not shown. What the status labels mean
Amazon and Microsoft each quantify their stakes in an AI developer, but neither isolates AI’s share of infrastructure spending. Comparing them means knowing which measure is which.
The spending is visible.
Are the returns?
Ojontic connects what management committed, what was invested, what was subsequently reported and what outcomes were disclosed — showing what can be established, and what cannot yet be compared.
the same as a successful investment.
“…will invest an additional $2.7 billion by Q4 2025.”
$1.25B in 2023 · $2.75B and $1.3B in 2024 · $2.7B in 2025
“…including $2.7 billion we invested in 2025.”
Were the 2025 Anthropic convertible-note payments made under the $2.7 billion commitment due by Q4 2025, or separately from it?
- Link between the 2024 commitment and 2025 investmentUnresolved — an open question in our approved research
- Amount investedQuantified
- Economic returnNot assessed in this example
- AI share of infrastructure spendingNot quantified
How much of annual capital spending supports AI workloads, and which measures should investors use to judge the return?
Research example, specific to Amazon’s FY2023–FY2025 Form 10-K filings, each with a verified analysis. The FY2024 commitment and the 2025 investment are separately verified disclosures; the filings reviewed do not establish that one fulfilled the other, and we do not infer it from the matching amounts. FY2025 is Amazon’s latest verified annual analysis; later Amazon filings are not reflected. Status labels
Follow one company
over time.
An AI narrative becomes more useful when claims, investments and commitments can be read in sequence. Select an Amazon disclosure to inspect its source. Each entry reflects that year’s Form 10-K only; later filings do not revise it.
is more than three summaries.
Period-specific record, FY2023–FY2025. Every entry draws on the verified analysis of that year’s Form 10-K and is not revised by later filings. Status labels
A multi-year record shows not only what was said, but when it became specific — when an investment was first quantified, when a risk was first named, and when a dated commitment first appeared.
The same lens,
every reporting period.
Ojontic tracks what companies disclose about AI across six key areas: what is established, what has changed, what remains unquantified, and what is not disclosed. One version-controlled method makes comparisons across companies and periods meaningful.
Products & adoption
What launched, what was planned, and what customers actually used — when reported.
Investment & infrastructure
Capital spending, computing capacity, committed infrastructure and research priorities.
Partnerships & deals
Collaborations, stakes and acquisitions, and the reasons management gives for them.
People & operations
Talent, training, workforce effects and AI used in a company's own processes.
Risks & constraints
Competitive, regulatory, supply and misuse risks the company explicitly links to AI.
Strategy, governance & results
Where AI sits in the strategy, who oversees it, and how stated plans evolve.
The method stays the same. It never forces an answer where the record is silent.
A claim should survive a click.
Not more AI commentary — findings an analyst can question, check and rely on.
This is the full verification process. Not every published example has completed it yet; each one carries its research status. See status labels
Evidence before interpretation
Every positive finding carries a verbatim passage and its location. Claims that cannot be traced to the source are discarded, not polished.
The same definitions, every time
Every company and period is assessed against one written policy, so a change in the findings reflects the company, not the method.
Complete sections, not search results
We examine complete relevant disclosure sections rather than relying only on search results, reducing the risk of overlooking important evidence.
Absence has a scope
“Not disclosed” records what was reviewed and under which definition. It never means the activity did not take place.
Uncertainty stays visible
Ambiguous cases are re-examined independently and flagged for review rather than turned into false certainty.
Tested on documents it has not seen
Each version is checked against independently labelled documents, with success criteria fixed in advance.
A focused first universe.
Our initial universe is ten major companies across technology, financials, retail, industrials and healthcare. Their FY2022–FY2025 annual reports were screened with an earlier, single-model version of our method; those screenings are preliminary and are not published as findings.
FY2022 — FY2025
Independently verified research to date: Amazon and Microsoft annual reports for FY2023–FY2025, plus Microsoft’s Q1 FY2026 Form 10-Q (filed 29 Oct 2025). Microsoft’s later FY2026 filings, including its FY2026 Form 10-K, have been acquired and await review. Alphabet’s FY2025 comparison evidence is hand-reviewed. Other companies and periods are preliminary single-model screenings, and the universe has not been refreshed through 2026.
What the research status labels mean
- Verified
- Independently annotated, adjudicated, validated and approved.
- Hand-reviewed
- Each displayed fact checked by hand against the primary-source SEC filing. Not yet through independent annotation and adjudication.
- Preliminary
- Experimental or single-model screening, not independently adjudicated. Not published as findings.
- In review / in progress
- Processing or review not complete. For a filing: acquired, but annotation or adjudication is not finished. For a displayed finding: a figure or interpretation not yet covered by approved research.
“Verified analysis” applies to a whole filing: two independent annotators, adjudication, validation and reviewer decisions on every open question. Each displayed finding also carries its own status, because a verified analysis does not by itself approve every figure or interpretation shown on this page. A research status describes how far the analysis has been checked. Labels such as “Quantified” or “Not in 10-K” describe what the filing discloses. Every example is specific to the reporting period shown; later filings do not overwrite earlier conclusions.
For equity analysts, portfolio managers and investment researchers
What did they commit to—and what happened next?
Request a complimentary AI strategy evidence comparison for up to three companies. We'll examine what each company disclosed, how its commitments evolved across annual reports, and what subsequent filings establish or leave unanswered.
- Each brief draws on the latest disclosures we have reviewed for each company and states its source cutoff. Newer filings not yet reviewed are named, not summarised.
- An accepted request results in a concise, source-linked research brief.
- Requests outside current coverage are reviewed for feasibility before acceptance.