Six months ago, your biggest AI line item was probably OpenAI. Today it might be Anthropic. We noticed the shift in our own transaction data and dug into the numbers.
Budgetly processes payments for Australian SMEs across NDIS, healthcare, education, construction, professional services, nonprofits, childcare, agriculture, transport, and retail. When we looked at what our customers are spending on AI platforms specifically, the pattern was clear: the vendor mix is shifting faster than most finance teams can track.
What Budgetly’s transaction data shows
This data comes from real Budgetly card transactions made by Australian SMEs to OpenAI and Anthropic between November 2025 and April 2026. These are businesses across more than 20 industries, all based in Australia, paying for AI tools through their Budgetly cards.
| Period | OpenAI share | Anthropic share | Trend |
|---|---|---|---|
| Nov 2025 | 92% | 8% | OpenAI dominant |
| Dec 2025 | 92% | 8% | Stable |
| Jan 2026 | 88% | 12% | Anthropic growing |
| Feb 2026 | 88% | 12% | Stable |
| Mar 2026 | 53% | 47% | Near parity |
| Apr 2026 | 23% | 77% | Anthropic dominant |
Total monthly AI spend across both platforms more than doubled between November and April. Businesses are not just switching providers. They are spending significantly more on AI every month.
Transaction volumes confirm both trends. Anthropic transactions grew from dozens per month to hundreds, while OpenAI transactions declined steadily. More teams, more use cases, more cost, and a completely different vendor mix in under 90 days.
It’s not the model. It’s the interface.
OpenAI released four model upgrades over this period (GPT-5, 5.1, 5.2, 5.3) and still lost share. The models got smarter. The spend still moved. What changed was not intelligence. It was what you could do without being technical.
Anthropic’s advantage is Claude, the conversational interface, and the tools built around it. Three capabilities shifted the balance:
Computer Use. Claude can operate your desktop directly: open applications, click buttons, fill forms, move between tabs. A finance team member can tell Claude to pull a report from Xero, compare it to a bank statement, and flag discrepancies. ChatGPT can answer questions about how to do that. Claude can do it for you. That is a fundamentally different product.
Dispatch. You can delegate multi-step tasks to Claude from your phone and walk away. It runs the task in the background, handles errors, and reports back when it is done. This turns AI from “a tool you sit in front of” into “a tool that works while you do something else.” For time-poor SME teams, that distinction is everything.
Managed Agents. Launched in April 2026, this lets businesses deploy Claude as persistent automation with sandboxing, checkpointing, and error recovery. It moved Anthropic from “better chatbot” to “platform you build workflows on.” April is when Anthropic hit 77% of total AI spend in our data.
OpenAI’s product during this period remained centred on ChatGPT: a conversational interface where you type prompts and get answers. Powerful, but passive. Anthropic built tools that act on your behalf. For Australian SMEs that need to replace manual workflows (not just get advice about them), that is the difference that moved the money.
Why this matters for your finance team
The AI spend shift creates three specific problems:
Unpredictable cost growth
AI tool costs do not behave like traditional SaaS subscriptions. Usage-based pricing means a team that doubles their API calls in a busy week can double the bill without any approval. The CFO Survey from March 2026 found that 74.5% of Australian finance leaders are interested in AI automation, but most have no framework for controlling the costs that come with it.
Vendor sprawl without visibility
When teams switch providers rapidly, finance loses track of which subscriptions are active, which are redundant, and which are growing. The data shows businesses running both OpenAI and Anthropic simultaneously during transition months, paying for overlapping capabilities. Without real-time spend visibility, these overlaps persist for months before anyone notices.
Budget category confusion
AI tool spend does not fit neatly into existing budget categories. Is it a software subscription? A professional service? An R&D cost? When the provider changes every quarter, the categorisation problem compounds. Finance teams end up with AI costs scattered across multiple budget lines, making it impossible to answer: “How much are we actually spending on AI?”
How fast-moving AI costs expose broken workflows
If your finance team discovers AI spend changes only when the monthly statement arrives, you are already behind.
Consider what happened in the data: a business spending heavily on OpenAI in February saw Anthropic overtake it entirely by April. If finance was reconciling monthly, they would not have noticed the crossover in March until the April statement arrived. By then, the new provider’s costs had already grown to dominate the AI budget.
This is the same pattern we see with shared corporate cards and uncontrolled subscriptions. The problem is not the AI spend itself. The problem is discovering it after the money is gone.
How to control AI tool costs without slowing down your teams
The goal is not to block AI adoption. It is to give finance the same visibility and control over AI spend that you have over every other business expense.
Step 1: Create a dedicated AI spend category
Stop burying AI costs inside “Software” or “IT Services.” Create a specific budget category for AI tools and require all AI-related subscriptions (OpenAI, Anthropic, Google, Perplexity, Midjourney, and any API usage) to be coded there. This gives you a single view of total AI investment across the business.
Without this, AI costs hide inside three or four different budget lines and nobody can answer the basic question: what are we spending on AI?
Step 2: Set per-vendor and per-team spending limits
AI spend should have the same controls as any other business expense. Set monthly limits by team and by vendor. When a team wants to switch providers or increase usage, the approval happens before the spend, not after the statement.
The data shows that when teams switch providers, they often run both simultaneously for weeks. Pre-set limits force the conversation about decommissioning the old tool before the new one scales up.
Step 3: Track usage-based costs weekly, not monthly
Monthly reconciliation is too slow for usage-based AI pricing. A team running heavy API workloads can burn through a quarterly budget in two weeks. Weekly spend reviews, or better, real-time dashboards, catch runaway costs before they compound.
Set up alerts for when any AI vendor exceeds 80% of its monthly allocation. This gives teams a warning before they hit the limit, not a surprise after they have already blown past it.
Step 4: Require a business case for each new AI tool
Every new AI tool should have a documented business case: what workflow it replaces, what the expected monthly cost is, and what the measurable outcome will be. This prevents the “shadow AI” problem where teams sign up for tools without finance knowing.
The business case does not need to be a 10-page document. Three questions: What does it replace? What will it cost per month? How will we measure whether it is working?
Step 5: Budget for provider switching
The data shows businesses can shift the majority of their AI spend from one provider to another in a single month. Build this into your planning:
- Budget for 4-6 weeks of overlap when teams are migrating between providers
- Avoid annual commitments for AI tools. Quarterly or monthly billing gives you flexibility to switch when a better option appears
- Review AI vendor contracts quarterly, not annually. The market moves too fast for 12-month review cycles
The bigger picture for Australian SMEs
The CFO Survey found that 49.1% of Australian finance leaders say manual tasks consume 40% or more of their team’s time. AI tools are the obvious solution. But adopting AI without spend controls creates a new problem: uncontrolled, unpredictable costs that grow faster than the efficiency gains they deliver.
The businesses that will get the most value from AI are not the ones spending the most. They are the ones that know exactly what they are spending, which teams are using which tools, and whether the investment is delivering measurable results.
That requires the same discipline you would apply to any other business expense: visibility, controls, and accountability.
