Quarterly · Q2 2026

STATE OF CLINICAL AI

Australia's clinical-AI quarter, from the public record.

DRAFT v0 · facts with citations · not for circulation until legal review

Built from 10 independently-verified public datasets (see Methodology). Every figure is traceable to a government source and carries an as-at date. This brief reports what the public record shows — it makes no judgement about any product's quality, safety, value-for-money, or clinical effect.


The bottom line

Across the public record we can currently see, Australia's health system has committed ~$1.3 billion to health-IT in four jurisdictions — ~$0.5 billion of it digital-health-specific — and a pattern emerges that no single government portal shows: the same handful of vendors recur across state lines. Oracle/Cerner is bought in NSW, QLD and VIC (~$120M); Orion Health and Alcidion likewise span three states. Meanwhile the fastest-moving clinical-AI category — ambient documentation scribes — leaves almost no procurement trace at all, because it is sold by subscription below disclosure thresholds. The contract record tells you what the public system bought; it cannot tell you what clinicians are using. That gap is the reason this Observatory also has a verified-clinician layer.


1. The headline — what Australia is funding, approving, buying

Across the public record we can currently see, in clinical-AI & digital-health:

LayerSignal (this quarter's snapshot)Source
Funded (research)~$316M AI/digital-health-centric across 119 grants (of 158 matched / $423.5M broad)MRFF #7
Approved (regulator)1,430 FDA AI/ML device clearances; 14 AU/ANZ-founded (Annalise ×8, Artrya, Volpara, See-Mode)FDA #3
Bought (procurement)~$1.32B of state+federal health-ICT identified; ~$0.5B of that is digital-health-specific (rest is infrastructure/platform)#1/#2/#11/#12
Trialled55 AU clinical-AI trials, ~40–50% genuine AI, ~121 named sitesCT.gov #9

Procurement, by jurisdiction (multi-year contract totals, not annual spend):

JurisdictionAll health-ICT identifiedDigital-health-specific coreNote
NSW~$814.6M~$328Mone flagship program (SDPR ~$198M) + enterprise imaging (Sectra $98M)
VIC~$404.3M~$122M (~$85.5M ex-COVID)federated, services-led; COVID test-trace ~$36M, EMPI/messaging ~$26M
QLD~$92.3M~$57MOracle/Cerner ieMR rollout dominant (2019–26 cumulative)
Federal~$4.2M~$3.5Mmost federal health-IT is hidden under generic ICT line items
Read this as a floor, not a ceiling. It is what leaves a public procurement trace in four jurisdictions. SA/WA/TAS/ACT/NT (beyond research grants), private-provider purchases, SaaS subscriptions and sub-threshold buys are not captured.

2. The cross-jurisdiction view — the same vendors, traced across states

This is what no single government portal shows: the same vendor appearing across jurisdictions and lifecycle stages. You can explore the same picture interactively on the deployment radar.

Vendors bought by ≥2 states:

VendorStatesA$ (identified)What
Oracle / CernerNSW · QLD · VIC~$120.4Mthe EMR incumbent — bought in three states
Orion HealthNSW · QLD · VIC~$32.3Mhealth-information-exchange / EMPI — three states
DedalusNSW · QLD~$20.2Mclinical systems / eMeds
Fred IT / SafeScriptNSW · VIC~$15.6Mreal-time prescription monitoring — both states run it
AgfaNSW · QLD~$11.6Menterprise imaging / PACS
AlcidionNSW · QLD · VIC~$7.2Mpatient-flow / virtual-care (Miya) — three states
(infrastructure: NTT ~$86.8M, Telstra ~$74.6M, Data#3 ~$36.6M span NSW+VIC — counted as health-ICT, not digital-health)

The lifecycle trace — FDA-approved AND bought in Australia (imaging/device AI, the clearest funded→approved→bought→deployed signal): of 9 vendors FDA-cleared and bought here, Annalise.ai and Agfa are the only two visible at three stages (approved + bought + deployed); the other seven — Canon, Philips, Fujifilm, GE HealthCare, Varian, Siemens, Compumedics — are FDA-cleared and bought here but with no public deployment trace yet.


3. Programs in flight

  • NSW — Single Digital Patient Record (SDPR): ~$198M visible across 33 notices — the statewide EMR build (Epic-anchored), with a granular layer of ~25 device-integration contracts (Philips, GE, Siemens, Varian, Elekta) wiring instruments into the record. The largest single clinical system in the dataset is Sectra's enterprise imaging/PACS at ~$98M.
  • VIC — federated, not flagship: no single statewide EMR; instead per-service EMR projects (Eastern, Barwon, Western Health, an Oracle/Cerner migration) plus central public-health systems — the COVID Test-Trace-Isolate build (~$36M) and Orion Health EMPI + health-messaging (~$26M). VIC's spend skews to system-integration and services.
  • QLD — ieMR: Oracle/Cerner integrated electronic medical record remains the dominant line (#2), cumulative 2019–26.

4. The gaps — and why the community layer matters

The procurement record has structural blind spots, and the gaps are themselves the finding:

  • Ambient scribes leave almost no trace. The hottest AU clinical-AI category (Heidi, Lyrebird) shows up in deployments (RCH, Cabrini, PHN GPs) but in zero procurement records — they're SaaS/sub-threshold, so contracts can't see them. If you are evaluating one, our AI-scribes guide is the checklist to take into the trial.
  • Cleared-but-not-bought-here. AU/ANZ-founded, FDA-cleared vendors (Artrya, Volpara, See-Mode) appear in the regulator data but not (yet) in any AU purchase trail — innovation clearing offshore first.
  • Private-provider deployments are invisible. Aidoc (live at Integral Diagnostics) and Annalise (I-MED) are deployed via private radiology networks that don't publish contracts.

Implication: the contract/registry layer (this brief) answers "what did the public system buy?" — but it cannot see the scribe boom, private deployments, or front-line experience. That is exactly the layer the verified clinician community is built to capture.


5. Methodology & coverage (honesty bar)

  • Sources & licences: AusTender OCDS (CC-BY 3.0 AU), data.qld CKAN (CC-BY 4.0), buy.nsw & tenders.vic contract disclosures, FDA AI/ML device list, MRFF grant recipients (CC-BY 4.0), ClinicalTrials.gov. Hospital-performance baseline (AIHW MyHospitals) is held as context only — deliberately NOT joined to vendor data, because attributing a performance change to a system go-live would be a causal claim the data can't support.
  • How built: 10 datasets, each put through an independent adversarial verification agent that re-pulled the data, recomputed every figure, and audited the method; the two browser-scraped state datasets (NSW, VIC) additionally passed a live, to-the-cent spot-check against the source site. A real error was caught and fixed in every dataset.
  • Confidence: "digital-health-specific" is a keyword/vendor classification published with its method; generic infrastructure (hardware/network/cyber/hosting) is explicitly carved out of the digital-health figures. Procurement search engines OR-match, so recall is bounded — this is a floor.
  • As-at: 2026-06-13. Spine: observatory-data/spine/ (records.json, entities.json, coverage.json); the interactive version is the deployment radar and the per-product directory.

6. Citations

Every number above resolves to a per-product document in observatory-data/products/ (#1 federal, #2 QLD, #3 FDA, #4 ED performance, #5 vendor footprint, #6 lifecycle, #7 MRFF, #9 trials, #11 NSW, #12 VIC), each carrying its own source URLs, as-at date, and verification record.


Pre-publication checklist (per Observatory plan): (1) 1-hour defamation-lawyer eyeball — this brief reports public-record facts with citations and makes no quality/safety claims, which is the low-risk path; (2) confirm CC-BY attributions rendered; (3) coverage-honesty statement intact. Incorporated-association status is not required for a facts-only brief, but is required before the review/community layer goes live.