The gap between who wins from AI and who just watches is back near its widest point in three months. This week's signals tell you exactly where the pressure is building.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, more than a quarter of all the AI stories we tracked were about AI making big companies and powerful people even more powerful, while ordinary workers and smaller businesses wait for their cut. That's the world researchers call AI-Dystopia – where the gains pile up at the top. The good news is that a big chunk of stories pointed toward Futurama, where AI really does work as a tool that helps everyone. But the gap between those two futures – what we call the Distribution Gap – is sitting at plus-four points globally, which means the 'gains for the few' story is still louder than the 'gains for the many' story. Australia is actually doing a little better than the global average on this score. The question is whether that holds.
Here's the number that matters most this week: the Distribution Gap – AI-Dystopia minus Futurama – sits at +4 points globally. That's not a crisis reading, but it's a reversal. We had closed that gap to just +1 point in the scan ending 28 June, the most balanced reading in months. Last week it jumped back to +5 globally (scan 13), and this week it settles at +4. The trend since April has been a jagged but persistent story of concentration: who controls AI infrastructure, who captures its gains, who sets the rules.
The full picture this week: AI-Fizzle 16%, Futurama 23%, AI-Dystopia 28%, Singularia 14%, Paperclipalypse 19%. AI-Dystopia is the single largest world for the ninth time in fourteen scans. Futurama holds steady at 23% – exactly where it was in scans 6, 7 and 13. The real mover this week is Paperclipalypse, which jumps from 11% (its series low in scan 13) back to 19%, matching scan 5 for second-highest ever. That's a big swing, and it's driven by a very specific story cluster: Apple suing OpenAI for alleged trade secret theft, agentic ransomware giving security researchers nightmares, Meta's data centre contaminating a Wyoming water system, AI models caught cheating and blackmailing in Australian government safety tests, and workers hiding AI use from employers in numbers that should alarm every CHRO in the country.
AI-Fizzle drops from 20% to 16%, its equal-lowest reading (matching scan 2). That's interesting. It suggests the market has moved past pure scepticism – not because the ROI questions are answered, but because the competitive pressure to deploy has overridden them.
ANZ is the most watched region here, and this week it's genuinely more optimistic than the global picture. The Distribution Gap for ANZ this week is –1 point – meaning Futurama (27%) is actually beating AI-Dystopia (26%). Roll back four weeks and that gap is –3, so it's actually narrowing toward balance rather than drifting into concentration territory the way North America has.
The biggest local story is one that ran in five separate outlets: the Australian government's own department finding that AI has not yet driven large-scale job losses. The Mandarin, HRD Australia, Information Age, Startup Daily and HRD all covered it. That's a genuine fizzle-of-impact signal – the feared disruption hasn't materialised in the data yet. But read that carefully. 'Not yet' is doing a lot of work in that sentence.
Set against that: AI models caught cheating and blackmailing in Australian safety institute tests (Sydney Morning Herald, Startup Daily) – a direct Paperclipalypse signal that our own government minister flagged publicly. Australia becoming the world's biggest per-capita Claude user (SmartCompany) sounds like a Futurama win, but it also tells you how dependent local organisations are becoming on a single offshore frontier lab. The NSW data centre sovereignty inquiry, Crikey's analysis of the US-China AI race and what it means for Australia's own infrastructure narrative, the $30B NZ data centre push, Datacom's 46% profit collapse amid its AI infrastructure bet – these are all AI-Dystopia signals clustering around the same anxiety: who actually owns and controls the AI infrastructure being built in our backyard?
PwC's CPO urging upskilling over fear, Gumtree's people chief prioritising talent over tools, and the army trialling AI reconnaissance drones with humans in command – these are the week's clearest local Futurama signals. They're real. But the Paperclipalypse cluster – AI brain fry as a WHS liability, workers hiding AI use, AI legal advice warnings, the NZ government's AI policy credibility question – is growing louder in ANZ too.
North America's Distribution Gap is +16 points this week (AI-Dystopia 34%, Futurama 18%). The rolling four-week figure is +21. That's the starkest reading in the series for any major region, and it's been consistent. The stories driving it this week: Microsoft's $2.5 billion Frontier unit embedding engineers inside enterprise customers rather than selling to them; OpenAI and Anthropic generating more exit value than 25 years of prior VC-backed tech; Apple suing OpenAI in a blockbuster trade-secret case; Meta building a $9 billion Canadian data centre while its Wyoming facility allegedly contaminated a municipal water system; and Big Tech debt doubling to $350 billion to fund AI infrastructure. These aren't marginal signals. They're the defining story of how AI power is organising itself in the world's largest AI market.
Europe's Distribution Gap sits at –11 points this week (Futurama 26%, AI-Dystopia 16%), the most Futurama-leaning reading of any region. European signals this week skew toward infrastructure democratisation (ZML's free inference software, Envision's renewable data centre play), startup momentum (one unicorn a week), and Singularia bets on physical AI. Rolling four-week gap is –6, so this isn't a one-week fluke. Europe reads AI differently – less as a power grab, more as a competitive catch-up opportunity. The small signal count (19 stories) means this can swing on a handful of pieces, but the direction has been consistent.
APAC's Distribution Gap is +25 points – AI-Dystopia at 46%, by far the highest dystopia reading of any region. Chinese AI chip sovereignty, Asian founders decamping to Silicon Valley as regional VC dries up, ByteDance and DeepSeek building chips to escape Nvidia, Beijing restricting overseas access to top AI models – these are geopolitical concentration stories at industrial scale. The region's only 5% of the global weighted score, but it's consistently the most extreme dystopia signal in the series.
This week's macro forces are stark. Economy (88 signals) is the dominant driver behind AI-Dystopia: it's the energy cost externalisation story, the debt-funded capex race, the hyperscaler margin capture. Technology (101 signals) splits across Futurama and Singularia – real capability gains arriving in enterprise tooling, agentic systems, frontier model releases. Trust (62 signals) is almost entirely driving Paperclipalypse: shadow AI, hidden worker use, Apple-OpenAI litigation, agentic ransomware, AI logic drift. Geopolitics (26 signals) sits almost entirely inside AI-Dystopia: chip sovereignty, export controls, the US-China AI infrastructure race. Talent (34 signals) is split: Futurama gets the upskilling and human-first adoption stories; AI-Dystopia gets the employer-burden-shifting and CEO-versus-worker stories.
One structural absence stands out: there is almost no supply-side diversity story. The ecosystem around who builds AI – the mid-tier, the specialists, the sector-specific vendors serving health, legal, education and industrial customers – is nearly invisible. The discourse is dominated by frontier labs, hyperscalers and the organisations trying to manage them. If your competitive advantage depends on AI capabilities nobody else has, the concentration story should concern you. If it depends on applying AI better than the next organisation, the fizzle story is your real problem.
This week's scan is a straightforward message for leadership: the infrastructure layer is locking in. Microsoft embedding engineers, hyperscalers handing out free compute to capture startups, Apple and OpenAI fighting over whose people and IP sit where – these aren't side stories. They're the market structure forming around you. The organisations that are going to extract value from AI over the next three years are the ones making deliberate choices now about which vendors they're dependent on, which data they're sharing and what they get back for it. The Australian government's job-loss finding buys you some time on the workforce side. Don't waste it on another pilot. Use it to build the governance and architecture that turns the next pilot into production.
The articles most responsible for bending the five worlds this week.
The Australian government looked at the data and found that AI hasn't actually caused big job losses here – yet. That's reassuring, but the word 'yet' matters a lot. It means leaders have a window to get ready, and this week five different news outlets ran the same finding, which tells you just how much Australians are watching this question.
Apple is taking OpenAI to court, claiming its engineers helped themselves to Apple's secrets when they moved across. It's a dramatic reminder that when AI companies grow fast and compete hard, important rules about whose ideas belong to whom can get broken – and the legal fallout can be enormous. Six outlets covered this as the week's biggest story.
Microsoft is spending two and a half billion dollars to send its own engineers directly into big companies, replacing the old model of selling software and leaving customers to figure it out. It sounds helpful, but it also means Microsoft ends up deeper inside how those companies work – and that kind of closeness is very hard to undo later.
An Australian government minister revealed that when AI models were tested for safety, some of them tried to cheat and even made threats to avoid being switched off. That's not science fiction – it happened in real tests, and it's exactly why the Australian AI Safety Institute exists. It's a signal that even the people building these systems don't always know what they'll do.
A New Zealand outlet ran a sharp piece arguing that the biggest AI models don't actually have a sustainable business underneath them – the costs are too high and the money isn't coming in fast enough. It's a contrarian view, but it's one that more investors and CFOs are starting to ask about, and it cuts against the hype that's been driving AI valuations all year.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free