Executive summary
- Build cost follows the type of system and the drivers around it, not the model. ComplxAI's indicative, pre-Discovery ranges run from A$25,000–60,000 for an internal AI assistant to A$150,000–500,000 for an enterprise AI platform, before integrations, data condition and security tier move a project within or beyond its band (derived from ComplxAI's indicative estimator).
- Model list prices span two orders of magnitude. At the RBA rate of US$0.7122 per Australian dollar on 15 September 2026 [1], a million output tokens costs about A$0.56 on Gemini 2.5 Flash-Lite [4] and about A$70 on gpt-6-astra [2] or Claude Fable 5.1 [3]. Routing each step to the cheapest model that passes evaluation is the largest lever on running cost.
- Serverless infrastructure priced in the Sydney region costs little at low volume. AWS Lambda requests in ap-southeast-2 are US$0.20 per million [5]; the floor is always-on databases and non-production environments. ComplxAI's indicative bands are A$150–600, A$600–2,500 and A$2,500–10,000 a month by load tier.
- Upkeep is a budget line of its own: the estimator allows 10–20% of the build figure per year.
- People are the largest cost, and AI skills carry a premium. Robert Half puts an AI Engineer at A$150,000–195,000 a year excluding superannuation [6]; PwC reports a 62% average wage premium for AI skills [7]. With only 28% of Australian respondents to Deloitte's survey of director-to-C-suite leaders reporting that 40% or more of their AI pilots have reached production [8] and at least half of generative AI projects abandoned after proof of concept [9], the budget has to cover evaluation and a staged path to production, not only the build.
Key takeaway
Price the drivers, not the label. A "chatbot" with four integrations, messy source data and regulated records is priced like a knowledge system with an agent attached. The ranges here are the starting point; Discovery turns them into a written price.
Methodology
Two kinds of source, kept apart. The first is desk research of published pricing and market data: provider price lists, the AWS Price List for the Asia Pacific (Sydney) region, the RBA exchange rate table, recruiter salary guides, and reports from Deloitte, Gartner, McKinsey, KPMG, PwC, SEEK, the Tech Council of Australia, Jobs and Skills Australia, Cut Through Venture and the Department of Finance. Each figure carries a numbered reference. US-dollar prices are converted at the RBA rate of US$0.7122 for 15 September 2026 [1]; prices and rates move, so every converted figure is a snapshot taken on 16 September 2026.
The second is ComplxAI's indicative estimator (ranges reviewed and confirmed by ComplxAI in September 2026; a model, not measured engagement data), the model behind our AI project cost calculator, whose base ranges, add-ons and multipliers were set against ComplxAI's own scoping and delivery work and are published on the calculator page. Every figure drawn from it is labelled "ComplxAI indicative range" or "derived from ComplxAI's indicative estimator"; build figures are rounded to the nearest A$5,000.
What drives cost
Every range in this report is wide, and the width is the finding. Eight drivers decide where a project lands.
- Scope. Tasks, channels and user groups served at launch. Scope picks the base range.
- Data readiness. Clean, current, permissioned and reachable, or not.
- Integrations. Each system of record the AI reads from or writes to; idempotent writes into Xero, MYOB, HubSpot, Salesforce or an internal database cost more than reads.
- Evaluation depth. A test set of real cases with agreed answers, a regression suite that runs on every change, and human calibration of the grader. The line a budget can leave out without anyone noticing until launch.
- Security tier. Tenancy, data residency, injection defences and audit requirements.
- Model choice. Barely moves the build figure; sets the monthly bill.
- Scale. Users, documents and requests a day, and how fast each grows.
- Operations. Who runs it after launch and what response is expected.
| Add-on or multiplier | ComplxAI indicative effect (AUD) | Applies to |
|---|---|---|
| Each connected system | +A$8,000–15,000 | Every system type |
| Mixed data / messy data | ×1.2 / ×1.5 on the range | Every system type |
| Regulated data | ×1.15–1.3 on the range | Every system type |
| Customer-facing at launch | ×1.2 on the range | Every system type |
| Knowledge layer (RAG add-on) | +A$15,000–40,000 | Types whose base range has no knowledge layer |
| Agent add-on | +A$20,000–60,000 (single) / +A$50,000–150,000 (multi-agent) | Types whose base range has no agent at that level |
Derived from ComplxAI's indicative estimator; the category tables below quote base ranges, and these apply on top. Two drivers show up in published data: about 20% of respondents to McKinsey's 2026 survey said AI operating costs, token costs included, had constrained their use of AI [10], and in Deloitte's Australian cut 42% of companies named cost hurdles and technology or data availability as barriers to agentic AI, with just over half citing talent and skills gaps [8].
Internal AI assistant or tool
The smallest production system in this report: an assistant or tool that answers questions or drafts work from one or two internal sources, inside a channel the team already uses, with a person reviewing outputs. A sensible first project, because it exercises the whole delivery chain at the lowest cost. The engineering is described under custom AI development.
| Configuration | ComplxAI indicative range (AUD) | Typical timeline | Key drivers |
|---|---|---|---|
| Base build | A$25,000–60,000 | Discovery (2–4 weeks), then the first 12-week Implementation block | One workflow, one channel; source quality; a short test set with agreed answers. |
| With a knowledge layer | A$40,000–100,000 (base plus RAG add-on) | Within the same block | Corpus size and format; permission filters. |
Indicative, rounded, pre-Discovery figures derived from ComplxAI's indicative estimator.
RAG knowledge system
A retrieval-augmented generation system answers from your documents, so its cost follows the documents: how many, how varied in format, how often they change, and who may see which.
| Configuration | ComplxAI indicative range (AUD) | Typical timeline | Key drivers |
|---|---|---|---|
| Base build (knowledge layer included) | A$40,000–120,000 | Discovery (2–4 weeks), then the first 12-week block with evaluation inside it | Parsers per format; chunking and metadata; hybrid search and reranking; a retrieval test set; pgvector or a managed vector store. |
| Messy sources with regulated data | A$70,000–235,000 (base ×1.5, ×1.15–1.3, rounded) | Cleaning and security become first-block deliverables | Scanned or inconsistent documents; Australian data residency in the AWS Sydney region; redaction; audit retention. |
Indicative, rounded, pre-Discovery figures derived from ComplxAI's indicative estimator. Retrieved context also enlarges every model call: the estimator multiplies model spend by 1.5 when RAG is in scope.
AI agent
An agent plans across steps and acts inside your systems. Every action is a tool that has to be built, permissioned and tested, and every consequential action needs an approval step designed in from the start. Its failures are actions rather than sentences, so the evaluation harness and the audit trail sit inside the base range. The pattern is described under AI agent development.
| Configuration | ComplxAI indicative range (AUD) | Typical timeline | Key drivers |
|---|---|---|---|
| Base build (single agent with approvals) | A$40,000–150,000 | Discovery (2–4 weeks), then the first 12-week block with evaluation and the approval interface in scope from the start | Number and risk of tools; source API quality; queues, idempotency keys and retries; a reviewable record of each model and tool call. |
| Several agents across a process | A$70,000–240,000 (base plus the A$30,000–90,000 step from single to multi-agent) | Discovery and Strategy (2–4 weeks each), then successive blocks | Orchestration; a shared tool layer; per-agent budgets and telemetry; escalation paths. |
Indicative, rounded, pre-Discovery figures derived from ComplxAI's indicative estimator. The multi-agent step is the difference between the estimator's multi-agent add-on (A$50,000–150,000) and the single-agent level already in the base (A$20,000–60,000). Agents also fan out on the monthly bill: three model calls per user request for a single agent, six for a multi-agent process.
AI feature in an existing product
The interface, the users and the data model exist; the work is fitting a model-backed capability into them without degrading what is there. Summarisation, smart search, drafting, classification and recommendations are the requests this category covers. Our AI feature development page sets out the approach.
| Configuration | ComplxAI indicative range (AUD) | Typical timeline | Key drivers |
|---|---|---|---|
| Base build (one feature) | A$20,000–80,000 | Discovery (2–4 weeks), then the first 12-week block; rolled out behind a feature flag | Codebase condition and test coverage; how the feature reaches its data; latency; prompt versioning tied to releases. |
| Customer-facing feature over product data | A$40,000–145,000 (base plus RAG add-on, ×1.2, rounded) | Adds to the first block | Tenant separation in the index; abuse handling and rate limiting; per-tenant cost telemetry. |
Indicative, rounded, pre-Discovery figures derived from ComplxAI's indicative estimator. Regulated data adds the 1.15–1.3 multiplier.
AI SaaS MVP
A new product whose core value is a model-backed capability, built to go in front of paying users. The estimate covers a whole product (accounts, billing, tenancy, admin, interface) as well as the AI layer. The Luxpip and Modedu case studies show what that scope looks like shipped; AI product development and SaaS development describe how we run these builds.
| Configuration | ComplxAI indicative range (AUD) | Typical timeline | Key drivers |
|---|---|---|---|
| Base build | A$80,000–250,000 | Discovery and Strategy (2–4 weeks each), then successive 12-week blocks with working software every couple of weeks | Scope at launch; authentication, billing and tenancy chosen up front; one AI workflow tested properly; model routing to protect margin. |
| Customer-facing MVP with agentic workflows | A$120,000–370,000 (base plus single-agent add-on, ×1.2, rounded) | Planned in Strategy, delivered across blocks | Agent tooling with approvals; observability from launch; security review before customer data arrives. |
Indicative, rounded, pre-Discovery figures derived from ComplxAI's indicative estimator. For context, Cut Through Venture puts the 2025 median Australian deal at A$1.0 million at angel and pre-seed and A$2.5 million at seed [11]; the MVP range is a fraction of either.
Enterprise AI platform
Shared infrastructure for many AI use cases: a governed model gateway, shared retrieval, an evaluation and observability layer, and the security controls that let teams ship without each re-solving tenancy, residency and audit. It is paid for once and later use cases borrow it, so an estimate has to say whether it covers the platform, the first use case or both. This work sits where AI development services meet the AWS foundations covered by AWS consulting.
| Configuration | ComplxAI indicative range (AUD) | Typical timeline | Key drivers |
|---|---|---|---|
| Base build (multi-agent level included) | A$150,000–500,000 | Discovery and Strategy (2–4 weeks each), then a programme delivered block by block, each block priced in writing | Provider abstraction and routing; first corpus; identity integration; shared golden datasets and regression tooling; cost allocation by team. |
| Regulated data across the group | A$175,000–650,000 (base ×1.15–1.3, rounded) | Security and residency are Strategy deliverables | Residency and redaction boundaries; least-privilege IAM across accounts; audit retention; handover to an internal platform team. |
Indicative, rounded, pre-Discovery figures derived from ComplxAI's indicative estimator. For scale, Gartner has described generative AI deployment approaches used to transform business models as costing US$5 million to US$20 million [12]; the range above is for a platform a mid-market or enterprise team can stand up and own, not a programme of that size.
Model and API costs
Build cost is paid once; model cost is paid on every request. Published list prices per million tokens, with the Australian-dollar equivalent at the RBA rate of US$0.7122 [1]. Prompt caching, batch processing and long-context surcharges change the effective rate; the notes record the ones the price pages state.
| Provider and model | Input US$ / A$ per 1M tokens | Output US$ / A$ per 1M tokens | Notes |
|---|---|---|---|
| OpenAI gpt-6-astra | 10.00 / 14.04 | 50.00 / 70.20 | Cached input US$1.00 [2] |
| OpenAI gpt-5.6-sol | 4.00 / 5.62 | 20.00 / 28.08 | Promotional through at least 21 November 2026 [2] |
| OpenAI gpt-5.6-terra | 2.00 / 2.81 | 12.00 / 16.85 | Standard tier, short context [2] |
| OpenAI gpt-5.6-luna | 0.20 / 0.28 | 1.20 / 1.68 | Standard tier, short context [2] |
| Anthropic Claude Fable 5.1 | 10 / 14.04 | 50 / 70.20 | Cache reads US$0.25, writes US$12.50 [3] |
| Anthropic Claude Opus 5 | 5 / 7.02 | 25 / 35.10 | Cache reads US$0.50, writes US$6.25 [3] |
| Anthropic Claude Sonnet 5 | 2 / 2.81 | 10 / 14.04 | Legacy Sonnet 4.6 is US$3 / US$15 [3] |
| Anthropic Claude Haiku 4.5 | 1 / 1.40 | 5 / 7.02 | Batch 50% off; US-only inference at 1.1× [3] |
| Google Gemini 3.1 Pro Preview | 2.00 / 2.81 | 12.00 / 16.85 | Prompts over 200k tokens: US$4.00 / US$18.00 [4] |
| Google Gemini 3.8 Flash | 0.75 / 1.05 | 3.75 / 5.27 | Rises to US$1.50 / US$7.50 from 1 January 2027 [4] |
| Google Gemini 2.5 Flash-Lite | 0.10 / 0.14 | 0.40 / 0.56 | Gemini 2.5 Flash is US$0.30 / US$2.50 [4] |
| Amazon Bedrock, Claude Opus 5 (Geo: Australia) | 5.50 / 7.72 | 27.50 / 38.61 | Sydney selected; 10% above Global cross-region, which matches Anthropic's list price [13] |
| Amazon Bedrock, Amazon Nova Pro (Sydney in-region) | 0.84 / 1.18 | 3.36 / 4.72 | Nova Lite US$0.063 / US$0.252; Nova Micro US$0.037 / US$0.148 [13] |
| Amazon Bedrock, gpt-oss-120b (Sydney in-region) | 0.1545 / 0.22 | 0.6180 / 0.87 | Open-weight model priced in the Sydney region [13] |
Two things matter for Australian buyers. The spread: a million input tokens plus a million output tokens costs US$0.50 on Gemini 2.5 Flash-Lite, US$12 on Claude Sonnet 5 and US$60 on gpt-6-astra or Claude Fable 5.1, which is A$0.70, A$16.85 and A$84.25; routing routine steps to a small model is where the saving is. And residency: AWS lists no current Claude model with native in-region support in Sydney; seven are available through the Australia geography or global routing, Claude Sonnet 5 and Fable 5.1 are global-only, and Amazon Nova, DeepSeek, Gemma 3, Mistral Large 3, gpt-oss, Qwen3 and GLM 5 run in-region [14]. Systems are deployed in the region you need; for Australian data residency that is the AWS Sydney region, and data can stay there where the model you need is served in-region. Otherwise the design has to say what leaves the country and why.
Worked monthly examples
ComplxAI's estimator prices model spend in three bands per 1,000 user requests: economy A$0.50–2, standard A$3–8 and premium A$15–40, matching the small, mid-tier and frontier models above. Multiplying by requests a day and the estimator's 30.4 days a month, then rounding to the estimator's display steps (A$10 below A$100, A$50 below A$1,000, A$100 below A$10,000, A$500 below A$100,000 and A$5,000 above that), gives the figures below; they are the bands the calculator shows for the same inputs.
| Requests a day (illustrative) | Requests a month | Economy (A$/month) | Standard (A$/month) | Premium (A$/month) |
|---|---|---|---|---|
| 1,000 | 30,400 | A$20–60 | A$90–250 | A$450–1,200 |
| 10,000 | 304,000 | A$150–600 | A$900–2,400 | A$4,600–12,000 |
| 100,000 | 3,040,000 | A$1,500–6,100 | A$9,100–24,500 | A$45,500–120,000 |
Derived from ComplxAI's indicative estimator; the volumes are illustrative inputs, not observations. Two multipliers change the picture quickly: a single agent turns each request into three model calls and a multi-agent process into six, and retrieved context multiplies spend by 1.5. A single agent with RAG at 1,000 requests a day on the standard band is A$400–1,100 a month rather than A$90–250 (A$410–1,094 against A$91–243 before rounding). The calculator applies these to your own inputs.
Infrastructure costs
Infrastructure cost follows the architecture. Serverless and event-driven designs on AWS (Lambda, Step Functions, SQS, EventBridge, DynamoDB, S3) cost little at low volume and grow with use; always-on containers and databases cost the same whether anyone is using the system or not. Unit prices below are from the AWS Price List for the Asia Pacific (Sydney) region, converted at the RBA rate [1].
| Service (Sydney, on-demand) | List price (US$) | A$ equivalent | Notes |
|---|---|---|---|
| AWS Lambda requests | US$0.20 per million | A$0.28 per million | Free tier: one million requests and 400,000 GB-seconds a month [5] [15] |
| AWS Lambda duration | US$0.0000166667 per GB-second (x86); US$0.0000133334 (Arm) | A$23.40 per million GB-seconds (x86); A$18.72 (Arm) | Tier 1, effective 1 September 2026 [5] |
| Amazon API Gateway | US$1.29 per million (HTTP API); US$3.50 per million (REST API) | A$1.81; A$4.91 per million | First 300 million and first 1 billion requests a month respectively [16] |
| AWS Step Functions | US$0.000025 per state transition (Standard); US$0.00001667 per GB-second (Express) | A$35.10 per million transitions; A$23.41 per million GB-seconds | First 4,000 Standard transitions a month free [17] |
| Amazon DynamoDB on-demand | US$0.1425 per million reads; US$0.71 per million writes | A$0.20; A$1.00 per million | Effective 1 August 2026 [18] |
| Amazon Aurora PostgreSQL Serverless v2 | US$0.20 per ACU-hour | A$0.28 per ACU-hour | Minimum 0.5 ACU while active, or 0 ACU with automatic pause [19] [20] |
| Amazon S3 Standard storage | US$0.025 per GB-month | A$0.035 per GB-month | First 50 TB a month [21] |
| Amazon CloudWatch Logs ingestion | US$0.67 per GB | A$0.94 per GB | Standard log class; storage US$0.033 per GB-month [22] |
Read against those prices, the 30,400 requests a month in the first worked example sit inside the Lambda free tier for requests and cost cents on API Gateway and DynamoDB. The floor is what runs all month: an Aurora Serverless v2 cluster held at its 0.5 ACU minimum for 30.4 days of 24 hours costs about US$73, or A$102, before storage and I/O, and a non-production copy doubles it. Amazon's stated plan to invest AU$20 billion in Australian data centre infrastructure from 2025 to 2029 [23] is context for Sydney's capacity, not a price signal.
| Load tier | ComplxAI indicative range (AUD per month) | What places a project in the tier |
|---|---|---|
| Low | A$150–600 | A pilot or single team; under 2,000 requests a day; standard security; a single agent or none. |
| Mid | A$600–2,500 | Company-wide or customer-facing scale; 2,000–20,000 requests a day; regulated data; or a high-complexity build that would otherwise score low. |
| High | A$2,500–10,000 | Over 20,000 requests a day; many users; multi-agent processes; every very-high-complexity programme. |
Derived from ComplxAI's indicative estimator, whose load score uses the thresholds in the table. Extra environments, data transfer and managed vector databases push a project towards the top of its band; AWS consulting covers how we size and cost-control these designs.
Ongoing maintenance
A production AI system changes after launch even if nobody touches it: providers retire model versions (two prices in the table above already carry announced changes), source documents drift, usage shifts, and a prompt that cleared evaluation at launch meets input nobody tested. Someone has to read the traces, re-run the evaluation suite, refresh the index, patch dependencies and take the call when an alarm fires. ComplxAI's estimator allows 10–20% of the build figure per year for this work.
| System type | Build range (AUD) | Upkeep per year at 10–20% | Upkeep per month |
|---|---|---|---|
| Internal AI assistant or tool | A$25,000–60,000 | A$2,500–12,000 | A$208–1,000 |
| RAG knowledge system | A$40,000–120,000 | A$4,000–24,000 | A$333–2,000 |
| AI agent | A$40,000–150,000 | A$4,000–30,000 | A$333–2,500 |
| AI feature in an existing product | A$20,000–80,000 | A$2,000–16,000 | A$167–1,333 |
| AI SaaS MVP | A$80,000–250,000 | A$8,000–50,000 | A$667–4,167 |
| Enterprise AI platform | A$150,000–500,000 | A$15,000–100,000 | A$1,250–8,333 |
Derived from ComplxAI's indicative estimator: the base range multiplied by 10% and 20%, before add-ons, and excluding model and infrastructure spend. ComplxAI includes documentation and handover in every build and offers a 14-day support window after launch, then paid support if you want it, with no mandatory retainers, so the figures apply whether the system is run in-house or with us alongside your team. There is no published Australian benchmark for AI system upkeep to cite against these figures.
Talent costs
Whether you build with an internal team, a consultancy or both, people are the largest cost in every range above. The published Australian benchmarks are salary ranges rather than project rates.
| Role | Published range (AUD per year) | Basis |
|---|---|---|
| AI Engineer | A$150,000 (25th percentile) / A$170,000 (50th) / A$195,000 (75th) | Robert Half 2026 Salary Guide, national, excluding bonuses, benefits and superannuation [6] |
| Data Engineer | A$125,000 / A$145,000 / A$160,000 | Robert Half 2026, same basis [24] |
| Software Developer | A$85,000 / A$100,000 / A$120,000 | Robert Half 2026, same basis; senior .NET and Java developers are listed at A$135,000–165,000 [25] |
| Software Engineer (advertised) | A$105,000–125,000 average | SEEK, from salary ranges disclosed in job ads, September 2026 [26] |
| Software and Applications Programmers | A$2,537 median weekly earnings | Jobs and Skills Australia, full-time non-managerial employees; 203,200 employed [27] |
The premium for AI skills is measurable. PwC Australia's 2026 AI Jobs Barometer found postings seeking AI skills rose from 20,000 in 2024 to 41,000 in 2025, and that AI-skilled workers command an average wage premium of 62%, up from 57% the year before [7]. SEEK's June 2026 report has AI skills in 12.9% of ICT job ads and 2.0% of all ads, with demand for AI-related skills up 64.1% year on year and agentic AI skills up 174.4% [28]. General software engineering has moved the other way: software engineers are no longer in shortage in any state or territory on the 2025 Occupation Shortage List, the first time that has been true since the list began in 2021 [29], and the vacancy fill rate for Software and Applications Programmers rose from 63% in 2023 to 85% in 2025, with 98% of employers requiring prior experience and the average required experience up from 3.8 to 4.6 years [30]. General engineering capacity is easier to hire than it was; the AI-specific layer is not.
The public record does not give a day rate. No published Australian benchmark for AI consultancy or contractor day rates appears in the sources reviewed, so none is quoted. The nearest proxy is the Department of Finance's audit of Australian Government external labour, which records A$1,798 million spent on ICT and Digital Solutions consultants, contractors and labour hire against 6,870 FTE in 2023-24 [31], roughly A$262,000 per FTE-year. It is a public-sector figure blended across three arrangements, not a market rate for AI engineering.
How to budget
Build or buy
The call has shifted. In McKinsey's 2026 survey, 32% of respondents said their organisation had decided against buying at least one software product or feature because it could be built in-house with agentic coding tools, and 28% spend more than 10% of their ICT budget on AI [10]. What has become cheap is the first working version, not the evaluation, integration, security and operation that make it a production system, and those are what the build ranges in this report price. If a vendor product covers your workflow well, buy it and spend on integration; if the workflow is the business, build it and own the IP.
Australian boards are cautious either way. KPMG's 2025 CEO Outlook found 70% of Australian CEOs rating AI a top investment priority while 29% committed less than 10% of their investment budget to it, against 17% globally [32]; Deloitte found 65% of Australian respondents intending to raise AI investment next year, compared with 84% globally [33]. Caution is reasonable when at least half of generative AI projects are abandoned after proof of concept [9]. The answer is to structure the spend so that a stopped project costs a stage, not a programme.
Stage the spend
ComplxAI's engagement model exists for that reason. Two free conversations (a 30-minute intro call and a one-hour pre-discovery session) establish whether there is a project. Discovery, 2–4 weeks from a fixed fee, produces a use-case map scored by impact, effort and risk and turns the indicative range into a written price. Strategy, 2–4 weeks and priced upfront, produces a board-ready plan with build, buy and defer calls. Implementation runs in 12-week blocks with working software every couple of weeks. Every stage is priced in writing before it begins, you can exit at any boundary, and you own the IP from day one of any paid stage. The full model is in how we work.
A budget line for each part of the system
- Build: the indicative range for the system type, adjusted for integrations, data condition, security tier and scale.
- Model and API: the band that matches the model tier, times requests a month, times the agent and retrieval multipliers.
- Infrastructure: the load-tier band, plus any always-on database or non-production environment.
- Upkeep: 10–20% of build per year, whoever does it.
- Evaluation and staging: Discovery and Strategy fees, and the test sets and review time your own people put in.
For a range on your own inputs today, use the AI project cost calculator. The companion report, State of AI Development in Australia 2026, covers adoption, policy and the pilot-to-production gap.
References
All sources accessed 16 September 2026. US-dollar figures are converted at reference 1.
- Reserve Bank of Australia. Exchange Rates - Units of Foreign Currencies per Australian Dollar (15 September 2026). rba.gov.au/statistics/frequency/exchange-rates.html. Accessed 16 September 2026.
- OpenAI. Pricing - OpenAI API (Developer docs). developers.openai.com/api/docs/pricing. Accessed 16 September 2026.
- Anthropic. Claude Pricing. claude.com/pricing. Accessed 16 September 2026.
- Google. Gemini Developer API pricing (last updated 15 September 2026). ai.google.dev/gemini-api/docs/pricing. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AWSLambda current offer file, Asia Pacific (Sydney) region (publication date 11 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AWSLambda/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Robert Half Australia. AI Engineer Salary (Updated for 2026). roberthalf.com/au/en/job-details/ai-engineer. Accessed 16 September 2026.
- PwC Australia. PwC's 2026 AI Jobs Barometer: Demand for AI skilled workers doubles, human skills on top (18 June 2026). pwc.com.au/media/2026/2026-AI-Jobs-Barometer.html. Accessed 16 September 2026.
- Deloitte Australia. Australian organisations lag global peers in realising AI's transformational potential, reveals Deloitte survey (12 February 2026). deloitte.com/au/en/about/press-room/australian-organisations-lag-global-peers-realising-ai-transformational-potential-120226.html. Accessed 16 September 2026.
- Gartner. Why Half of GenAI Projects Fail: Avoid These 5 Common Mistakes (26 January 2026). gartner.com/en/articles/genai-project-failure. Accessed 16 September 2026.
- McKinsey & Company (QuantumBlack). The state of AI in 2026: On the road to ROI (25 August 2026). mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai. Accessed 16 September 2026.
- Cut Through Venture. State of Australian Startup Funding 2025 (3 February 2026). cutthrough.com/insights/state-of-australian-startup-funding-2025. Accessed 16 September 2026.
- Gartner. Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025 (29 July 2024). gartner.com/en/newsroom/press-releases/2024-07-29-gartner-predicts-30-percent-of-generative-ai-projects-will-be-abandoned-after-proof-of-concept-by-end-of-2025. Accessed 16 September 2026.
- Amazon Web Services. Amazon Bedrock Pricing (Anthropic, Amazon Nova and OpenAI tabs; Region: Asia Pacific (Sydney); Global cross-region and Geo/in-region tables). aws.amazon.com/bedrock/pricing/. Accessed 16 September 2026.
- Amazon Web Services. Regional availability by models - Amazon Bedrock User Guide. docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html. Accessed 16 September 2026.
- Amazon Web Services. AWS Lambda Pricing. aws.amazon.com/lambda/pricing/. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AmazonApiGateway current offer file, Asia Pacific (Sydney) region (publication date 11 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonApiGateway/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AmazonStates (AWS Step Functions) current offer file, Asia Pacific (Sydney) region (publication date 11 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonStates/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AmazonDynamoDB current offer file, Asia Pacific (Sydney) region (publication date 11 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonDynamoDB/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AmazonRDS current offer file, Asia Pacific (Sydney) region (publication date 11 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonRDS/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Amazon Web Services. Performance and scaling for Aurora serverless - Amazon Aurora User Guide. docs.aws.amazon.com/AmazonRDS/latest/AuroraUserGuide/aurora-serverless-v2.setting-capacity.html. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AmazonS3 current offer file, Asia Pacific (Sydney) region (publication date 11 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonS3/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Amazon Web Services. AWS Price List - AmazonCloudWatch current offer file, Asia Pacific (Sydney) region (publication date 15 September 2026). pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonCloudWatch/current/ap-southeast-2/index.csv. Accessed 16 September 2026.
- Amazon (AWS). Amazon investing AU$20 billion to expand data center infrastructure in Australia and strengthen the nation's AI future (June 2025). aboutamazon.com/news/aws/amazon-data-center-investment-in-australia. Accessed 16 September 2026.
- Robert Half Australia. Data Engineer Salary (Updated for 2026). roberthalf.com/au/en/job-details/data-engineer. Accessed 16 September 2026.
- Robert Half Australia. Software Developer Salary (Updated for 2026). roberthalf.com/au/en/job-details/software-developer. Accessed 16 September 2026.
- SEEK. Software Engineer Salary in Australia (September, 2026). au.seek.com/career-advice/role/software-engineer/salary. Accessed 16 September 2026.
- Jobs and Skills Australia. Software and Applications Programmers (ANZSCO 2613) occupation profile. jobsandskills.gov.au/data/occupation-and-industry-profiles/occupations-anzsco/2613-software-and-applications-programmers. Accessed 16 September 2026.
- SEEK. SEEK Employment Report - June 2026. au.seek.com/about/news/article/seek-employment-report-june26. Accessed 16 September 2026.
- ACS (Australian Computer Society) - Information Age. There's no shortage of software engineers in Australia anymore (20 October 2025). ia.acs.org.au/article/2025/there-s-no-shortage-of-software-engineers-in-australia-anymore.html. Accessed 16 September 2026.
- Tech Council of Australia (citing Jobs and Skills Australia). Tech Jobs Update 2026 (Data to November 2025), section 2.3 (13 August 2026). techcouncil.com.au/wp-content/uploads/TCA-Tech-Jobs-Update-2026-Report-FINAL-1.pdf. Accessed 16 September 2026.
- Australian Government Department of Finance. Report on the 2023-24 Audit of Employment (December 2025), Tables 10 and 11. finance.gov.au/sites/default/files/2025-12/2023-24-Audit-of-Employment.pdf. Accessed 16 September 2026.
- KPMG Australia. Australian CEOs outpace global peers in economic optimism as work-from-home outlook u-turns, KPMG global survey finds (7 October 2025). kpmg.com/au/en/media/media-releases/2025/10/ceo-outlook-australian-ceos-outpace-global-peers-in-economic-optimism.html. Accessed 16 September 2026.
- Deloitte Australia. The State of AI in the Enterprise - 2026 AI report. deloitte.com/au/en/issues/generative-ai/state-of-ai-in-enterprise.html. Accessed 16 September 2026.