From Compute to Control

I spent 15 October at 6DAI Sydney, 6 Degrees Media's enterprise AI conference at the Hyatt Regency on Sussex Street. I went looking for who could help an enterprise understand what its AI actually costs.

The answer surprised me. Fact: eight of the 20 sponsoring companies now offer some form of AI spend measurement, attribution or control, most of it launched or announced in 2026. [partners page; company releases]

Not long ago, the enterprise AI question was whether the technology could do the work. The sessions in Sydney were about something else: Cisco's AI chief technology officer on "The Price of Intelligence – Tokenomics at Scale," a CXO panel on "The Buck Stops Where?", and an audience Q&A on "Techonomics – managing AI spend while still funding growth." [6 Degrees Media, final 6DAI Sydney 2026 agenda, 12 Oct 2026] The conversation has moved from acquiring AI to operating it, economically and responsibly.

My read: each of these meters is built to explain AI spend well inside the platform it serves. Enterprises, though, run AI across many platforms at once. The next piece of work is the one that connects them: a single, auditable record of what AI did across the business, what it cost and who approved it. Call it the ledger.

1. Enterprise AI is moving from compute acquisition to operational control. Verified: the 6DAI Sydney agenda centred on agent identity, accountability and AI spend rather than access to capacity, and buyers are already measuring returns, with CBA reporting about $200 million in FY26 gross AI benefits.

2. Eight of 20 sponsors now offer AI cost measurement or controls. Verified: most of these features were launched or announced in 2026, each designed around the platform or ecosystem it serves.

3. A cross-platform AI ledger is the next opportunity. Our analysis: with agents running across many platforms and Australian AI budget ownership split five ways, a shared record linking agent spend, approvals and business outcomes looks set to become a distinct layer of value.

A useful way to see the shift is to treat an AI agent like a new hire. When a company first tries AI, it asks one question: can this do the task? When it deploys hundreds or thousands of agents that act on their own, it has to ask the questions it asks about every employee. Who is this? What information can it see? What decisions can it make? How much can it spend? Who is accountable for what it does?

The afternoon keynote from AWS and Okta, "The Next Billion Identities Won't Be Human," started exactly there. The next billion identities an enterprise manages, they argued, will belong to AI agents rather than people, and access controls, accountability and governance need to be built for that so organisations can adopt agentic AI with confidence. [6 Degrees Media keynote announcement; final agenda, 12 Oct 2026]

a16z's Terry Angelos made the same point from the payments world this week. An agent, he wrote, needs "a payment policy, not just a payment credential": rules for what it may buy, from whom, up to what amount, and who makes it right when it gets something wrong. [a16z, "Wallet Wars", 15 Oct 2026] Replace "payment" with "enterprise system" and you have the brief for the enterprise AI control stack.

Put the conference evidence together and five connected functions emerge. The first three are covered here; economics and the ledger follow below.

Identity → Data → Governance → Economics → Ledger

Identity: who is the agent? Fact: Okta made Agent SSO generally available on 24 August 2026, extending single sign-on to AI agents. [Okta, 24 Aug 2026] APRA's April letter to industry observed that identity and access management had "not yet adjusted to nonhuman actors such as AI agents." [APRA, 30 Apr 2026] An agent that cannot be identified cannot be permissioned or held to account.

Data: what can it access and trust? A whole session, "The Accelerators," was given to data readiness, with speakers from Nasdaq, Transurban, Suncorp, UTS and Sharon AI. Fact: Fivetran completed its merger with dbt Labs on 1 June 2026 under the banner "the data infrastructure for trusted AI agents." Alation relaunched as AIOS and added AI Governance connectors for Amazon Bedrock, SageMaker, Databricks MLflow, Copilot Studio, Microsoft Foundry and Snowflake Cortex. Bigeye now positions data observability as "Enterprise AI Trust." [Fivetran, 1 Jun 2026; Alation, 17 Sep 2026; Bigeye] An agent is only as reliable as the data it is allowed to use.

Governance: what can it do? Fact: 90% of Australian CIOs in Thoughtworks' 2026 survey said AI adoption is moving faster than governance can adapt. [CFOtech Australia on Thoughtworks Global CIO Survey 2026, 8 Oct 2026] The sponsors are answering in product. Airia, a Visionary in Gartner's Magic Quadrant for AI Governance Platforms, ran a fireside with Fresh Clinics on governing agents and MCP; Boomi launched its Agent Control Plane from Conshohocken and Sydney; OutSystems, UiPath and Thoughtworks build governance into their agent platforms; and Rimini Street offers "Rimini Govern for AI" as a managed service. Governance is moving out of policy documents and into the systems where agents run.

This is where the eight meters sit. Each is designed around the platform or ecosystem it serves:

Boomi's chief executive summed up the question finance teams are now asking: "which agent spent this, on what, and who approved it." [Boomi, 2 Sep 2026]

The ledger is the layer that answers that question for the whole enterprise: what the agent did, what it cost, who approved it, and where it should be recorded. a16z's "You Need a New CFO" describes the same requirement from inside finance. As agents take on work, finance needs to know where each number came from, how it was calculated and which agents were involved, and the authors argue that "having controls enables speed." [a16z, 8 Oct 2026] David Haber adds the commercial reason: once an agent keeps a complete record of a customer journey, its outcome becomes attributable, and vendors can price on results. [a16z, "Winning the Relay Race", 14 Oct 2026]

This is not an empty market. Cross-cloud FinOps tools have existed for years, and several sponsors already reach beyond their own walls: Airia across model providers, Bigeye across agent platforms, Snowflake's gateway to third-party agents. But enterprise-wide reconciliation is genuinely hard, for three reasons.

Different units. AI spend arrives as credits, tokens, seats, platform units and GPU hours, each priced and metered differently.

Different identities. The same agent can appear under different names in an identity provider, a data platform and an automation tool.

Different owners. Fact: Australian CIOs in the Thoughtworks survey split AI budget ownership five ways: central IT (22%), business units (22%), shared between IT and the business (20%), executive or board level (18%), and "still evolving" (18%). [CFOtech Australia, 8 Oct 2026]

The platforms are already moving towards the problem. In the first half of 2026 Cisco bought the agent observability platform Galileo; Boomi acquired the AI gateway Lunar.dev; and Snowflake agreed to acquire Natoma, an enterprise MCP platform. [Cisco, 9 Apr and 22 May 2026; Boomi, 13 May and 27 Jul 2026; Snowflake, 27 May 2026] Each extends a platform's reach into the agent request path, which is where a ledger would start.

The room made clear who will demand this first. Fact: of the 17 speakers on the final agenda from organisations that buy rather than sell AI, ten came from financial services, including Colonial First State, Westpac, CBA, Zip, QBE Ventures, NGS Super, Suncorp, Lendi and Nasdaq. [final agenda, 12 Oct 2026; my count] The Buy Australian AI Partnership, launched on 27 August with the National AI Centre as principal sponsor, brings in ANZ, CBA, Cuscal, NAB and Westpac, institutions representing "more than $22 billion in annual procurement spend." [Stone & Chalk, 27 Aug 2026; Assistant Minister Charlton]

These buyers are already measuring results. Fact: CBA reported about $200 million in measured gross benefits from AI use cases in FY26, including reinvested capacity, and expects that to double in FY27 and exceed investment. [CBA FY26 results, 12 Aug 2026] And regulators are asking for the ledger's foundations. APRA expects "an inventory of AI tooling and AI use cases" and "ownership and accountability across the AI lifecycle." [APRA, 30 Apr 2026] ASIC begins a banking-sector AI review in the second quarter of 2026–27. [ASIC, 26-232MR, 30 Sep 2026]

Sovereign capacity fits the same story. Sharon AI, sharing its stand with VAST Data, pitched local control over where data sits and models run; in September the two launched DataEnclave for "banks, government agencies and other regulated organizations." [Sharon AI, 23 Sep 2026; Sharon AI's Q2 2026 figures are in Appendix A]

It is early, and the products are new. Three questions will show how this phase develops.

1. Which layer becomes the system of record. Identity, data, governance and economics each describe the same agent from a different angle. I'll be watching for the first shared record that joins them: what an agent did, under whose identity, on which data, at what cost.

2. Who owns it. Finance, IT or the business; the answer will shape which tools enterprises standardise on.

3. What gets disclosed. CBA has shown what an AI benefits disclosure looks like. The next step is benefits and costs reported side by side, use case by use case.

Sydney showed an industry moving from compute to control, with every layer of that stack well represented on the floor. Thank you to 6 Degrees Media and to everyone who took the time to talk with me on the day.

Sponsor tiers from the 6DAI Sydney 2026 partners page (accessed 15 Oct 2026).

As published by 6 Degrees Media (page last modified 12 Oct 2026).

Event

Identity, data and governance

Cost

Buyers and regulators

a16z essays

Capacity

Appendix figures

Carol Chen

Founder, Compute Notes

Builder of AI-native businesses and investor in AI infrastructure

LinkedIn: linkedin.com/in/carol-c-76461498

Website: computenotes.co

Network: network.computenotes.co

Get a free AI spend teardown → watt.computenotes.co

#AIInfrastructure #FinOps #EnterpriseAI #AIEconomics #Compute #PhysicalAI #Token