If you haven’t already, subscribe and join our community in receiving weekly AI insights, updates and interviews with industry experts straight to your feed.

DeepDive 

Your weekly immersion in AI 

If you’re an AI founder, you’ve probably spent the last few years asking how you can keep competing with the AI firms that have billions to spend. Every new frontier model release has seemed to widen the gap – more compute, bigger training runs, and (of course) bigger budgets. 

It’s reasonable to feel like the future of this sector belongs to hyperscalers. But at the beginning of June, Microsoft AI made an announcement that we think is very encouraging for startups. 

The incumbent launched seven new MAI models spanning reasoning, coding, transcription, voice, and image generation. But the most important part of the announcement isn’t the models themselves – it’s the strategy behind them. 

Microsoft described its approach as building a ‘hill-climbing machine’ – a system designed for continuous optimisation across reinforcement learning, infrastructure, deployment feedback, and specialised workflows.

And if this approach is the right one, it changes what real success in the AI sector could look like. 

The era of ‘one giant model’ may be ending

The dominant assumption in AI has been that bigger models are better. That belief has shaped much of the startup ecosystem. Companies raced to raise capital, secure GPUs, and build ever-larger systems in the hope of competing at the frontier.

But this latest announcement suggests that a more nuanced market is emerging. Microsoft repeatedly emphasised efficiency, reinforcement learning environments, enterprise tuning, and real-world deployment systems. Even its flagship reasoning model, MAI-Thinking-1, was presented as part of a larger optimisation pipeline.

The implication here is that next-gen AI leaders could be the companies that build the smartest systems around models – not just the single smartest model itself. 

That’s a huge opportunity for startups. Because you’re not going to win by outspending giants – but you could gain serious traction by being more focused, and closer to customer problems. 

Specialisation could become a startup advantage

This launch from Microsoft shows that AI is becoming more operational and domain-specific. 

The company discussed reinforcement learning environments for specialised Excel agents. It announced collaborations with healthcare organisations (specifically the Mayo Clinic), to build models grounded in real clinical workflows. It focused heavily on deployment efficiency, not just raw intelligence.

All of that suggests that context is becoming just as important as technological capability – and context is where startups can really shine, by understanding and solving very specific problems in very targeted ways. 

A startup with deep expertise in a particular industry may now have a genuine advantage over larger, more general-purpose players. Because the value is shifting away from the foundation model alone and toward the surrounding system:

• Workflow integration
• Proprietary data
• User feedback loops
• Domain expertise
• Operational reliability 

And that’s a much more accessible game for startups to play.

AI startups don’t need to beat Microsoft

You don’t have to try to compete directly with frontier labs. But Microsoft’s announcement actually points toward a more realistic and potentially more profitable path.

The AI stack is becoming more layered, more specialised, and more operational.

There will still be a handful of companies building giant foundation models. But around them, an enormous ecosystem of startups will emerge – and that’s where many of the most sustainable businesses may be built.

So go and climb hills. Know your niche, and believe in the power of constant refinement over occasional breakthroughs. 

We want your perspective 

Open this newsletter on LinkedIn and tell us what you think: what advantages do startups still have in the AI race?