---
title: Bespoke software has a landlord
date: 2026-10-28T12:00:00+11:00
author: Portable
canonical_url: "https://portable.com.au/articles/bespoke-software-has-a-landlord"
section: Articles
---
This is a monumental week for artificial intelligence. This week the company behind Claude asked public markets to value it at up to US$2 trillion, the largest float in history. That's a milestone birthday for the AI boom, and not a gentle one. What happens from here is very consequential. If this float fails, then the AI boom is over and we’ll be somewhere new.

The AFR's James Thomson called it ["the ultimate market referendum"](https://www.afr.com/chanticleer/anthropic-just-kicked-off-biggest-pressure-test-in-investment-history-20261027-p618w1) on artificial intelligence, and warned that if investors go cold, the hairline cracks in the AI trade become fissures, and the correction finds its way into all of our super. He's right. But I read the prospectus with a different set of concerns. I'm not watching this as an investor. I'm watching it as a tenant.

## How did a handful of companies end up owning the brain?

There is a lot in the 335-page prospectus, including roughly 70,000 words of risk factors. Someone noted that's about the length of *Frankenstein*, which is either very funny or not funny at all, depending on how much of your business runs on it. But there’s a lot in these boring numbers and it’s the way they connect to all of the other major tech companies. Five customers account for about a quarter of Anthropic's revenue. Almost half its first-half revenue came through the cloud marketplaces of Amazon, Google and Microsoft. Those same three companies are also its suppliers, its investors and its competitors, and Amazon alone could own about 20 per cent of it after the float. Thomson calls this "stunningly circular". It reminds me of the type of hubris one encounters walking into a casino, where the house sells you the chips, runs the tables and owns the bank across the road.

Then there's the bill. Anthropic has signed up for something like US$800 billion of computing power over the next decade. Its revenue has gone from about US$4.6 billion last year to a run rate above US$33 billion, which is extraordinary, but quarterly growth has slowed from around 100 per cent to under 50 per cent. Across the industry, usage rose about 90 per cent last quarter while spending rose just 19 per cent. We are all using a lot more and paying a lot less for each unit, and someone has to pay for the data centres. There is no way around it. The price of intelligence today is subsidised, and subsidies always end. Remember when Ubers were cheap?

## What does it actually look like when the landlord sells?

A few weeks ago I wrote that we are entering [the era of beautifully bespoke software](https://portable.com.au/articles/the-era-of-beautifully-bespoke-software). Building costs are collapsing, and for the first time in my career courts and government departments may not have to be on a vendor roadmap they share with ten thousand other customers. They can have the thing that fits. I still believe that, deeply. But put yourself in the shoes of a CIO at a mid-sized government department. Over the past eighteen months you've finally escaped the platform you were locked into for a decade. Your team has built a beautiful, purpose-built case management tool, and your caseworkers love it. And under the hood, every single interaction runs through one model, from one company, sold to you through one cloud.

Now imagine the float goes badly. Nothing dramatic happens, because that's not how these things break in the real world. There's no blackout and no Black Mirror episode. Instead, a public company under pressure to show margin puts its prices up. Rate limits tighten. The model you built on gets retired with ninety days' notice and a canned email. Or the hyperscaler that already owns a fifth of the company takes the rest, and your bespoke tool becomes a line item on someone else's cloud invoice.

We have swapped dependence on a platform's roadmap for dependence on a model provider's pricing. Like all erosions, it's slow and inconspicuous, and by the time you notice it the ground has already moved. I wrote about the same pattern in [big tech and the erosion of access to justice](https://portable.com.au/articles/bigtech) back in 2021; the architecture of these companies shapes what happens to the people who depend on them.

## So is this a bubble, and does it matter if it pops?

Portable was born out of the last dot-com bust. When that bubble burst, investors lost fortunes and Amazon's share price fell more than 90 per cent. But the crash left something behind, and it's the part of the story we tend to forget. Telecommunications companies had laid tens of thousands of kilometres of fibre optic cable for demand that hadn't arrived yet, and most of it sat dark. A generation of very good engineers was suddenly looking for work. That glut is what made broadband cheap, and cheap broadband gave us YouTube, cloud computing and pretty much everything we now take for granted online. The people who did well in the decade after weren't the ones who owned the pipes. They were the ones who knew what to build with them.

If the demand doesn't arrive for US$800 billion of compute, we get our own dark fibre. Data centres sit half empty and inference gets cheap, maybe ridiculously cheap. The market splits in two: frontier labs charge a premium for the very best models, while open-weight models, including Chinese ones, push the price of "good enough" towards nothing. Most real work doesn't need the frontier models. A court form, a caseworker's notes or a council permit flow needs good enough, designed properly by people who understand the problem.

The wrappers will wash out, and they should. Start-ups and SaaS "AI features" that only made sense on subsidised tokens will fold. Consultancies that sold pilots and never delivered anything will be found out. CFOs will take over the AI budget, and every line item will have to show its return. Only when it makes sense.

The pain will come home, too. Australian Retirement Trust says more than a quarter of its portfolio is AI-related. A correction hits household wealth, then tax receipts, then government budgets. Anyone who works with the public sector, and that's us, should be planning for tighter spending in 2027 and 2028.

## How do we build for a future where we don't own the brain?

I don't believe the answer is to slow down. That would be the wrong lesson entirely. The answer is to build like the landlord might sell.

That means treating the model as a component and not as the foundation; if swapping providers means a rebuild, you haven't built a product, you've signed a lease. It means owning the things that last: the problem definition, the data, the prompts, the evaluations, and the hard-won judgement about what good looks like. It means pricing for tomorrow and not today, because a business case that only works at subsidised token prices doesn't work. And it means asking every vendor, including us, the awkward question: what happens to this service if the price of the model doubles, or the model disappears?

We've started asking that of our own work, from [Sweep](https://portable.com.au/articles/sweep-wage-compliance-platform-launch-2026) and [Hear Me Out](https://portable.com.au/articles/hear-me-out-legal-design-web-award-innovation) to the tools our teams use every day, and the [guardrails we've written about](https://portable.com.au/articles/trust-in-the-time-of-ai-guardrails-that-work) for experimenting with AI safely. We’re building in the infrastructure that our clients own or have procured, so it takes a lot of that decision making away for now, but I think it’s going to become very relevant in 2027.

## Restoration

I don't know whether this float soars or flops. Nobody does, and I think ultimately it’ll come down to sentiment. What I do know is that the technology isn't going anywhere, and that a crash, if it comes, will make intelligence cheaper and more widely available than it has ever been. Whether that turns out to be good news depends on who owns it, and on whether we've built anything that is actually ours.

AI evolves quickly. Where we direct our energy today matters. The next decade needs us to build software that fits the problem, and doesn't fall apart when the landlord sells.

## References

Australian Financial Review, ["Anthropic targets Aussie investors, as IPO filing shows $114b losses"](https://www.afr.com/world/north-america/anthropic-s-ipo-filing-shows-explosive-growth-risks-and-staggering-costs-20261027-p618rz), 27 October 2026

James Thomson, ["If Anthropic's IPO goes wrong, the AI bubble will pop"](https://www.afr.com/chanticleer/anthropic-just-kicked-off-biggest-pressure-test-in-investment-history-20261027-p618w1), Australian Financial Review, 27 October 2026

The Information, ["Anthropic's confidential filing reveals customer concentration"](https://www.theinformation.com/briefings/anthropics-confidential-filing-reveals-customer-concentration)

Fast Company, ["Anthropic's IPO filing is the start of the put-up-or-shut-up era of AI"](https://www.fastcompany.com/91629897/anthropics-ipo-filing-is-the-start-of-the-put-up-or-shut-up-era-of-ai), October 2026

The Information, ["Anthropic's rapid but slowing growth could weigh on IPO valuation"](https://www.theinformation.com/articles/anthropics-rapid-slowing-growth-weigh-ipo-valuation), October 2026

Ron Schmelzer, ["Anthropic's IPO reveals the biggest risk to the AI boom"](https://www.forbes.com/sites/ronschmelzer/2026/10/26/anthropic-ipo-ai-boom-valuation-risks/), Forbes, 26 October 2026 (citing Ramp data via Reuters Breakingviews)

Andrew Apostola, ["The era of beautifully bespoke software is here"](https://portable.com.au/articles/the-era-of-beautifully-bespoke-software), Portable, 20 October 2026

## More from Andrew Apostola

[The era of beautifully bespoke software is here](https://portable.com.au/articles/the-era-of-beautifully-bespoke-software)

[Big tech and the erosion of access to justice on a platform near you](https://portable.com.au/articles/bigtech)

[Guardrails that actually work | Trust in the time of AI](https://portable.com.au/articles/trust-in-the-time-of-ai-guardrails-that-work)

[We submitted to the Senate wage theft inquiry. Here's why.](https://portable.com.au/articles/beyond-wage-theft-enforcement)

[Portable secures $1.1M BRII Grant to enhance wage compliance with Sweep](https://portable.com.au/articles/sweep-by-portable-brii-grant)

[What Civily shows us about collaboratively designing justice, globally](https://portable.com.au/articles/civily-digital-family-law-justice-platform)

[Hear Me Out wins Most Innovative Website at the Australian Web Awards](https://portable.com.au/articles/hear-me-out-legal-design-web-award-innovation)

 

 

 

 

## Explore more of Portable's work

- ![](https://portable.com.au/assets/_transforms/images/article/connection-ownership-respect-in-ai-coding-trust-in-the-time-of-ai/82820/DKTwTOhhGY716lVn_TrustinthetimeofAI3_638b64f2391dff6e364e5709b4734889.webp)### [Connection, ownership, respect in AI coding | Trust in the time of AI](https://portable.com.au/articles/trustworthy-ai-assisted-development)
- ![](https://portable.com.au/assets/_transforms/images/article/how-to-experiment-and-scale-ai-safely-trust-in-the-time-of-ai/82675/8bp7iVqzSdQ3WoTJ_TrustinthetimeofAI2-1_638b64f2391dff6e364e5709b4734889.webp)### [How to experiment and scale AI safely | Trust in the time of AI](https://portable.com.au/articles/how-to-experiment-and-scale-ai-safely)
- ![](https://portable.com.au/assets/_transforms/images/article/guardrails-that-actually-work-trust-in-the-time-of-ai/82666/8auutsfSzEi3KrAe_TrustinthetimeofAI-1_638b64f2391dff6e364e5709b4734889.webp)### [Guardrails that actually work | Trust in the time of AI](https://portable.com.au/articles/trust-in-the-time-of-ai-guardrails-that-work)
- ![The era of beautifully bespoke software is here 1](https://portable.com.au/assets/_transforms/images/article/the-era-of-beautifully-bespoke-software/84718/The-era-of-beautifully-bespoke-software-is-here-1_638b64f2391dff6e364e5709b4734889.png)### [The era of beautifully bespoke software is here](https://portable.com.au/articles/the-era-of-beautifully-bespoke-software)
- ![Designing and developing amica, a digital solution for separating couples](https://portable.com.au/assets/_transforms/images/caseStudy/amica/7938/Xv5tlhAAACAAmk_n_638b64f2391dff6e364e5709b4734889.jpg)### [Designing and developing amica, a digital solution for separating couples](https://portable.com.au/work/case-studies/amica)
- ![Building trust in AI for Australia’s future thumbnail](https://portable.com.au/assets/_transforms/images/caseStudy/76398/Thumbnail_638b64f2391dff6e364e5709b4734889.png)### [Building trust in AI for Australia’s future](https://portable.com.au/work/case-studies/building-trust-in-ai-for-australias-future)
- ![Portable Report: Customising GenAI, What You Need to Know](https://portable.com.au/assets/_transforms/images/report/generative-ai-report/30338/ZvO1I7VsGrYSv60X_638b64f2391dff6e364e5709b4734889.png)### [Portable Report: Customising GenAI, What You Need to Know](https://portable.com.au/reports/generative-ai-report)
- ![Unlocking the potential of AI in statutory planning for Victorian councils](https://portable.com.au/assets/_transforms/images/caseStudy/mavlabs-ai-in-statutory-planning-for-councils/10682/aHnmkkMqNJQqIFnv_638b64f2391dff6e364e5709b4734889.jpg)### [Unlocking the potential of AI in statutory planning for Victorian councils](https://portable.com.au/work/case-studies/mavlabs-ai-in-statutory-planning-for-councils)