AWS Summit Singapore on 6 May opened with two Australian-adjacent companies showing real agents running in production. AWS Summit Sydney on 13–14 May opened with a leadership reshuffle and a partner-channel narrative. Eight days later, the federal budget proposed a tax structure that gives the senior engineers who would have built the Australian equivalents a clear financial reason to leave.

The gap is not a coincidence. It is the visible end of a talent pipeline problem the budget just made worse.

What Singapore put on stage

At Singapore, Grab’s managing director of strategic finance Ken Lek walked through Project Grabhouse: a rebuild of the company’s data foundation on Amazon S3 and Apache Iceberg, with the Bricks platform on top letting teams “build automation workflows and deploy artificial intelligence agents at scale without writing any code.” Manual data reconciliation dropped by up to 60%. Lek’s framing: “The first decade of Grab was about scale, the next decade is about intelligence.”

Certis CEO Ng Tian Beng then brought robots on stage. Max, a multi-service autonomous concierge. Ace, an autonomous patrol dog. Both coordinated through Mozart, the company’s operational platform on Bedrock. Ng’s line: “We don’t use AI just to generate insights. We use it to orchestrate decisions in our operations, to coordinate resources and to drive action in the real world.”

Two companies. Two flagship in-production agentic stories. Both at scale. Both on the keynote stage.

What Sydney put on stage

Sydney led with Pip Gilbert’s appointment as the new A/NZ head of partner, Chris Casey moving in as managing director to replace Rianne van Veldhuizen, and a partner-channel story (47% to 68% of A/NZ work flowing through partners). One of Gilbert’s three stated priorities for 2026 was “advancing proof-of-concept to production deployment.” Her own words on the shift: “Getting through the proof-of-concept into production – that’s been a material shift in the past year.”

That is a tacit admission. The reason “POC to production” is a 2026 priority is that the 2024–25 GenAI pilots largely did not ship at scale. There was no Sydney equivalent of Grabhouse on the keynote stage because there is not yet a Sydney equivalent of Grabhouse in production.

That part is fixable. Australian engineering teams know how to ship. The question is whether the people who would do that shipping have a reason to do it here.

The 12 May budget answers that question

The tax reform chapter of the 2026–27 budget proposes, from 1 July 2027, to replace the 50% CGT discount with cost-base indexation and impose a 30% minimum tax on net real capital gains. Baker McKenzie’s read is straightforward: cost-base indexation for assets held more than 12 months, with a 30% floor on net gains.

It is a proposal, not law. Consultation is pending. A startup carve-out is “under consultation,” which means it does not yet exist on paper.

That ambiguity is the problem. People making a 3–5 year equity bet today are pricing the worst-case scenario, not the eventual carve-out.

The maths that breaks the equity bet

Henry Innis, CEO of Mutinex, worked it through publicly. A senior data scientist takes a startup role for $400K of options vesting over five years, accepting a $60K annual pay cut against her stable corporate salary. Under today’s CGT discount, her effective tax on the liquidity event is about 23.5%, take-home about $306K. Under the proposed regime, that becomes about 47% and roughly $212K. Innis’s conclusion: “The maths on that pay cut no longer work. She stays at the stable job. The startup hires someone less experienced, or doesn’t hire at all.”

That is the supply line for Australian agentic case studies. The senior engineer who would have built Grabhouse-equivalent infrastructure in Sydney is now $94K worse off for doing it.

Where she goes instead

Singapore is actively pricing the same engineer in. The ONE Pass (AI and Tech) track that replaces Tech.Pass on 1 January 2027 accepts vested ESOPs and share schemes toward the S$30,000 monthly threshold, with a fixed-cash floor of S$22,500. The Ministry of Manpower’s stated rationale is that top AI and tech professionals are often paid in equity. Singapore’s CGT on share gains is zero.

New Zealand has no general capital gains tax on shares. The US, after the OBBBA changes, exempts up to $15M of qualified small business stock gains under Section 1202 for stock issued after 4 July 2025.

So the scoreboard for risk-taking technical labour reads: three jurisdictions actively pulling that labour in by taxing equity lightly, and one pushing it out by taxing equity more heavily while saying a carve-out might arrive later.

What the sector said out loud

Tech Council CEO Kate Cornick, in her first week in the role: “There is work to do to ensure Australia’s startup community doesn’t become collateral damage as a result of proposed changes to CGT.” Heidi Health CEO Dr Thomas Kelly, the same day: “In the past 12 hours alone, I have had my team asking about relocating overseas because they see these settings as making Australia less competitive for people who want to build ambitious global companies.”

Twelve hours. Not twelve months.

What to do about it before July 2027

Rerun your ESOP maths now. If your option grants were modelled against the 23.5% take-home number, redo them at 47%. If the equity story you sold candidates 18 months ago breaks at the new rate, get ahead of it. Cash adjustments, accelerated vesting on liquidity, or a refresh grant. Do not sit on it waiting for the carve-out to be designed, because your engineers are doing their own arithmetic on a much shorter clock.

For your top five technical hires, write down what their equity is worth net in Sydney, Singapore, Auckland, and a US Section 1202 company. If the difference is enough to decide a job change, assume your competitors have already done the same exercise and put a number on the offer.

And then ship the production agentic story. The reason there was no Sydney equivalent of Grabhouse on the keynote stage is that nobody had one to show, and the policy fight will not change that. A shipped product is the only thing that holds local relevance regardless of where the consultation lands. The companies that put real agents into real production this year, not the ones waiting on the tax outcome, are the ones who get on the Sydney 2027 keynote stage.

The Sydney keynote was not weak because Australian engineers cannot build the work. It was weak because the marquee stories are not yet built, and the budget just made the supply side worse.

If your team is making equity-design decisions in the gap between the budget proposal and the consultation outcome, talk to us.

Sources