Two very different institutions. The Australian Government in Canberra and the frontier AI labs in San Francisco spent the quarter underwriting the same outcome from opposite directions. One is subsidising the ownership of innovation. The other has been subsidising the consumption of intelligence. One subsidy is structural. The other is already beginning to unwind.
That convergence sits at the centre of Aura Ventures' investment strategy. Sovereignty, because Australia has made it increasingly clear that critical capability, from energy and advanced manufacturing to AI infrastructure and software, must be built domestically rather than imported. Subsidy, because both government policy and the economics of frontier AI continue to shape where returns accrue. The question that intrigues me most is which subsidy is likely to persist.
The May Federal Budget reinforced a shift that has been building for several years. Through the Future Made in Australia agenda, expanded support for AI commercialisation, and continued investment in sovereign capability, the Commonwealth has made an explicit policy choice: Australia should own more of its strategic technology stack rather than lease it from overseas.
That shift has only sharpened following the Fable controversy, which highlighted the risks of building national capability on technology controlled elsewhere. This also holds true for enterprises and businesses. Whether the issue is access, pricing, availability or policy, dependence on a small number of frontier model providers has become a strategic risk rather than merely a technical one.
The consequence has been a renewed emphasis on open-weight models, domestic capability and software businesses that retain control of their own economics. The most powerful expression of that policy is not a grant programme. It is the ESVCLP regime.1
The ESVCLP tax exemption is often described as a concession. It is better understood as sovereign co-investment.
Eligible income and capital gains generated by qualifying venture investments can flow to investors exempt from Australian tax. Rather than collecting its share of successful outcomes, the Commonwealth deliberately leaves that value with investors to encourage domestic venture capital and sovereign innovation. Viewed through that lens, the government is not simply encouraging venture investing; it is increasing the after-tax return available from it.1
For Australian investors, that creates an increasingly distinctive source of alpha.
Traditional asset classes like listed equities, commercial property and fixed income remain fully exposed to taxation on investment returns. Venture capital held through an ESVCLP structure occupies a different position. As tax becomes an increasingly important component of long-term compounding, after-tax returns matter more than headline returns.
In that sense, ESVCLP has evolved beyond being a niche venture incentive. It has become one of Australia's most compelling tax-alpha investment structures, directly aligned with the country's strategic objective of building sovereign technology capability.
The second subsidy has come from the AI industry itself. For much of the past three years, frontier AI labs appear to have priced inference below its long-term economic cost, prioritising adoption and ecosystem growth ahead of immediate profitability. Customers benefited from extraordinarily inexpensive intelligence while providers absorbed much of the infrastructure burden.
The last 3 months marked an important inflection point. Flagship model pricing began rising rather than falling. Tokenisers improved efficiency but also altered effective pricing. Capacity constraints became increasingly visible across leading providers, and competition shifted from acquiring users at any cost towards generating sustainable economics.
The direction of travel is now clearer than it has been since generative AI emerged. The subsidy that made intelligence exceptionally cheap is beginning to unwind as OpenAI and Anthropic contemplate a listed world.
That does not weaken the AI opportunity. It changes where value is created.
As inference costs normalise, businesses whose economics depend primarily on reselling frontier model access will face increasing pressure. By contrast, companies that own customer workflow, proprietary data, orchestration software and deployment infrastructure remain largely insulated from changes in underlying model pricing.
At the same time, open-weight models continue to narrow the capability gap with proprietary frontier systems while dramatically reducing deployment costs. The strategic implication is significant. Competitive advantage is moving away from the model itself and towards the software layer that determines how intelligence is deployed. Control over workflow increasingly matters more than ownership of the underlying model.
This is precisely the layer Aura Ventures focuses on.
Companies that own customer relationships, automate expensive workflows and improve unit economics regardless of which foundation model ultimately powers the application. Their value compounds because they are not structurally dependent on any single frontier provider.
The quarter also reinforced another long-term reality. The world's most valuable technology companies increasingly remain private for much longer than previous generations. By the time public investors gain access, much of the value creation has already occurred, and retail investors provide the liquidity window for earlier investors.
The opportunity for venture investors is not simply earlier access. It is participation during the years when technological uncertainty declines while enterprise value compounds most rapidly.
That remains particularly true across AI infrastructure and enterprise software, where today's category leaders continue to scale almost entirely within private markets.
Our thesis generation and portfolio design sit at the intersection of these structural trends.
We invest in Australian founders building sovereign technology capability. We back software businesses that own workflows rather than simply consume models, and we take significant ownership while these companies remain private, before public markets have an opportunity to price that growth.
The pattern remains consistent. The companies creating durable enterprise value are not those benefiting from subsidised compute. They are the businesses solving expensive customer problems with AI while retaining control of their own economics.
Closing Perspective:
Every investment cycle is shaped by a subsidy. The strongest long-term returns rarely come from chasing the subsidy itself. They come from owning the businesses that remain valuable after it disappears.
One subsidy, the under-pricing of frontier AI, is already beginning to fade as the economics of compute normalise. The other appears increasingly structural. Australia's commitment to sovereign capability, combined with the ESVCLP framework, represents an enduring alignment between public policy and private capital. It enhances after-tax returns while encouraging investment into precisely the businesses the country wants to build.
The frontier has been subsidising intelligence. Canberra is subsidising ownership and now Venture capital allows investors to participate in both, but only one of those subsidies is likely to endure.
1.Tax treatment under the ESVCLP regime is subject to eligibility requirements and individual investor circumstances. For non-Australian investors, Australian withholding tax may apply to certain distributions depending on the nature of the payment and the investor's jurisdiction, including the application of any relevant double tax agreement. Investors should obtain independent tax advice regarding their individual circumstances.