Insights

Savannah, Reef, Invasion or Farm? Four futures for the AI industry (Part 2 of 2)

By Will Hardman · · 8 min read

In part 1 I argued that two “big questions” - questions I expect to see answered within the next three years - will shape the future of the AI industry and determine the ripple-out effects on of the rest of the economy. The questions were:

  1. Substitution vs complementarity: will AI agents fully substitute and replace (a significant fraction of) human knowledge workers or merely empower them to achieve more?
  2. Concentration vs diffusion: who generates the tokens that get used for economic work - and at what price?

Today, I’ll have a go at painting these possible futures and asking what it suggests for business today. I’ll conclude with list of “no regret” actions, actions that every leadership team should be executing on today:

  1. Watch for and study your “AI native” challengers
  2. Ready your data and knowledge to build an “AI moat”
  3. Plan your investments to avoid AI vendor lock-in

Two questions, each with two possible outcomes gives four possible scenarios. Following a tradition in scenario analysis circles, I have given each a memorable name on the theme of ecosystems.

Scenario 1: The Savannah

In which a small handful of “charismatic megafauna” shape the landscape and everything else adapts to live around them.

Complementarity + Concentration

In this scenario, AI becomes much like cloud computing: a small handful of providers, offering the best models at any price point, compete to acquire and retain enterprise workloads. Product features like deep ecosystem integration, non-portable agents and artifacts or memory and personalisation make providers sticky. On the other hand, aggressive competition drives token costs down to commodity levels.

Corporate and consumer spend on AI levels out with investment directed towards use cases where productivity gains are real. Even if this is the scenario we end up in, there will be substantial growth in AI spend from today. For comparison, the cloud computing market is forecast to be worth around $875B in 2026 - or around 4-5X current spend on AI products and inference.

The Savannah is a great scenario for the hyperscalers - whose datacentre businesses will grow. It is a reasonable outcome for the frontier labs only if they can impose switching costs on their customers. If they cannot, their marginal share of each token will fall. For the rest of the economy, AI will simply be another line item cost required to remain competitive.

Complementary AI, by definition, will not lead to large-scale, structural unemployment.

Scenario 2: The Coral Reef

In which a stunning diversity of species colonise and compete across every niche of the ecosystem.

Complementarity + Diffusion

In this scenario, proprietary data and organisational knowledge - inaccessible to frontier models - become moats that firms seek to exploit. Vertical AI firms that have pooled industry data offer focused models with an edge over their generalist rivals. Larger firms host their own models, perhaps adapted from capable open-weights releases. The capability at the frontier might be ahead, but for large swathes of industry, “good enough” + “proprietary data” proves to be a winning recipe.

There are some indications that something of a diffusion may be happening right now. For example, Vercel, an agentic infrastructure provider, reported on 22nd August 2026 that the proportion of open model workloads on their platform had jumped from 28% to 62% in just two months this year. (Though the same report disclaims that actual spend remains concentrated among closed providers.)

The impact of this scenario on the capital and investment markets should not be underestimated. It’s the nightmare outcome for any institution that has poured money into a lab or piled on debt to facilitate its own AI R&D. As the terminal value of the labs declines, the economic shockwaves will reverberate through the capital markets and the economy more widely.

For the hyperscalers, the prospects are less gloomy. After all, so long as models are being used they need to be hosted somewhere and the infrastructure build-out will not be in vain.

Beyond this, the prospects for nations and businesses also look good. Eventually, firms will learn to reorganise their operations around the technology and no major shocks to employment are to be expected. It is perhaps the most optimistic - and least capital-concentrating - scenario of the four.

Scenario 3: Invasive Species

In which a new arrival outcompetes native species and runs rampant through a habitat, changing its character forever.

Substitution + Diffusion

What if our agents get good enough to replace large swathes of knowledge work but these capabilities diffuse out of the frontier labs and into open weights or vertical AI companies? This is perhaps the most radical of the four scenarios - but no less credible than the others.

Cheap automation poses an immediate threat to every business based on knowledge work. Incumbents will seek to defend their positions by virtue of the moats I listed earlier but for many of them, this may not be enough. Already the term “AI native” has been coined for new market entrants seeking to disrupt their industries with lean, human-light business models.

If enough of them succeed, we will see significant churn in not just who the most successful players are but where they are based. AI-native businesses may have far less trouble localising their services worldwide, freed from the constraints of needing to build local knowledge and expertise. On the flip side, no jurisdiction will have a labour cost advantage - so the scenario favours any country with good access to compute infrastructure and an AI-literate populace.

For businesses, the possibility of this scenario implies that they should be watching “AI native” competitors carefully and - if they are successful - buying or copying them. Many, of course, will not.

The implications for the wider economies of the world are interesting. Widespread success of “AI native” companies will surely result in a rise in unemployment. Whether temporary or structural is hard to say. This could be a major challenge for national governments already struggling to balance their budgets - but there is opportunity too: since the winning firms will be globally distributed, a shift from personal to corporate taxation could recoup excess profits from the winners and redistribute them to help society navigate the transition to a lower-labour set-point.

Scenario 4: The Farm

In which the landscape is fenced off and controlled by a single farmer. The fauna that live there are selected and husbanded for profit.

Substitution + Concentration

Here’s a thought experiment: suppose a single AI lab cracks RSI and the same recipe yields compounding improvements on other, research-intensive domains. The lab has the capability and the compute to deploy millions of these knowledge worker-equivalents and discovers two things:

  1. they have achieved full substitutability of human workers
  2. swarms of agents running on frontier models always outcompete those running on second-tier or specialised models.

Why, then, would they sell tokens to the likes of pharma companies, hedge funds and quantum computer developers when they could simply buy or outcompete them and claim the respective sectors?

Sure, this wouldn’t work everywhere. Domains where the verification loop is slow and expensive are less suitable - but a wide swathe of knowledge work could be in the crosshairs if we discover that cognitive labour is fully substitutable and RSI techniques transfer across domains well.

A monopoly on super-intelligence would be an economically concentrating event the likes of which the modern world has never seen. If there is to be a $100T company in my lifetime, this is how it will happen.

The impact of this centralisation would be most keenly felt by countries reliant on services businesses and intellectual capital, who will watch those sectors enter a precipitous decline. Those states lucky enough to host a leading lab (and right now, the USA and China are alone here) will at least be able to harvest a tax windfall from these new super-companies. For the rest (think the UK, Switzerland, Israel) the implications are sobering: structural unemployment, a personal and corporate tax base in precipitous decline and an inevitable fiscal crisis.

This is the scenario that the leading labs are simultaneously warning us about and driving towards. OpenAI has even proposed a “public wealth fund” seeded by AI companies. A reasonable prescription if you happen to be a state where a lab is headquartered; not so helpful for those that are not.

So what now?

For most firms, three of the four scenarios offer as much opportunity as they do risk. The smart move is to watch the industry and figure out in which direction we are headed.

  • If models are substituting for people, the effect will show up in hiring data and job losses even as firms post higher profits and invest further in AI.
  • If capabilities are diffusing, traffic to open-weights models will grow and niche or vertical AI firms will reach profitability and establish market footholds.

If I had to make three “no regret” recommendations to an executive team off the back of this analysis, they would be:

  1. Carefully watch your industry for AI-native challengers. Treat them seriously and - if they gain traction - react sooner rather than later. Don’t be outcompeted by an invasive species.
  2. Think about your data and know-how as potential moats. Work on initiatives to record and codify them. Have a strategy in place for exploiting your knowledge and data via a proprietary, “enterprise AI” capability. This will position you to flourish in a “coral reef”.
  3. Apply Satya Nadella’s test to your AI investments: be sure you can switch providers without losing everything you’ve built on their products and platforms. Don’t get locked in - which is how providers will shore up their pricing in a “savannah” scenario.

The purpose of this article is not to predict which scenario will play out, but to provide a framework for thinking about how AI might develop over the medium term. So long as we don’t end up in “The Farm”, forward-thinking businesses should be excited about what is to come and planning effectively for it.

Ready to move from pilot to production?

Book a free 30-minute consultation. You'll leave with a straightforward answer on where agentic AI can work for your organisation, and the first steps to get there.

Book a free 30-min consultation →