The noise around AI is deafening. Every second LinkedIn post is someone claiming AI will either save the world or end it. But whilst everyone’s arguing about whether robots will steal our jobs, something more practical is happening: the market is maturing, and the patterns are becoming clearer.
I’ve been looking at the data behind the hype, and there are five trends worth paying attention to if you’re building something, hiring someone, or wondering what your job looks like in a few years.
Language models eat everything else
Since 2021, language-focused AI has completely dominated. By 2024, more than 250 language models had been released, dwarfing everything else combined. This isn’t an accident.
Language is the easiest way for humans to talk to machines. It’s also the easiest way for machines to make sense of what we actually want. The commercial applications are obvious and scalable: customer support, writing, search, coding, and education.
All the stuff businesses actually need.
There’s also a practical reason: training data is everywhere, and it’s cheap. You can scrape billions of words from the internet. You can’t scrape the physical world quite as easily.
What this means for startups: building another ChatGPT wrapper won’t cut it anymore. The winners will be the ones who understand a specific domain deeply and can embed AI into real workflows. Context is becoming the only sustainable advantage.
Take Freight Logistics, for example: An AI that knows how port delays ripple through supply chains, understands seasonal shipping patterns, and integrates with existing TMS systems is solving real problems. Another chatbot asking “How can I help you today?” isn’t. Context is becoming the only sustainable advantage.
The money got smarter (and more concentrated)
Corporate investment in AI peaked at $350 billion in 2021. Since then, it hasn’t disappeared; it’s just gotten more targeted, and the geographic split is becoming starker.
2021 was peak FOMO. Everyone wanted an AI strategy, even if it meant buying a company with no revenue and three researchers. The M&A market was absolutely bonkers, accounting for massive chunks of that corporate investment. Then it dropped sharply as companies got more selective about what they were actually buying.
Now we’re in what I’d call the platform phase. But it’s not just about being smarter with capital, it’s about a shift in where the opportunity lies. AI infrastructure and governance investment went from essentially nothing to $32 billion in 2024. That’s not incremental growth, that’s the market screaming that the tooling layer is where the real money is.
Meanwhile, private investment tells a different story about geography. The US has recovered strongly by 2024, whilst other regions stayed relatively flat. The innovation gap isn’t narrowing, it’s widening.
For founders, this means two things. First, you can’t just wave “AI” around anymore. Second, if you’re not in the US, you’re probably going to be an AI consumer rather than a creator unless you find a very specific niche.
Industry owns the research now
Since 2015, industry teams have overtaken academia in building notable AI systems. By 2024, it wasn’t even close.
The reasons are straightforward. The best researchers followed the money and the GPUs to OpenAI, Google DeepMind, Anthropic, and Meta. Modern AI research needs massive compute, which costs millions. Universities can’t compete with that.
There’s also the fact that these companies ship products and papers simultaneously. The line between research and go-to-market has basically disappeared.
What this means: startups probably can’t out-research the big players. But they don’t need to. The opportunity is in building on top of frontier models, not competing with them. Distribution, domain expertise, and user experience are where you win.
Another question: The research undertaken in Academia was also like an ad for how good that Institute was. Thus attracting more potential students (who pay to go there and learn). If Industry now owns research, what does this mean for Universities?
The patent map tells a story
In 2019, China and the US filed more AI patents than the rest of the world combined. The gap is stark; many countries had either no data or negligible filings.
This reflects a few things. Patents can be as much about signalling as innovation, especially for China. Countries with mature IP systems reward filing. And increasingly, innovation is clustering around the big AI hubs.
For most of the world, this means becoming AI consumers rather than creators, unless there’s a serious investment in compute access and education. For startups outside the US-China axis, the opportunity lies in localisation, regulation-specific niches, and underserved domains.
Infrastructure exploded (autonomous vehicles didn’t)
Looking at where private money actually went in 2024, one area absolutely dominated: AI infrastructure, which exploded to $32 billion. Meanwhile, autonomous vehicles remained the smallest category, which tells you something about solving real problems versus chasing headlines.
This makes sense when you think about it. Everyone built language models. Now we need the scaffolding: guardrails, monitoring tools, fine-tuning platforms, efficiency tools, and synthetic data generation. The tooling gap is massive, and that infrastructure investment spike shows just how massive.
Autonomous vehicles, despite years of hype about robotaxis revolutionising transport, barely registered. Turns out building something that works reliably enough for public roads is harder than building something that works in controlled environments like warehouses. I think the market has figured out that the problem isn’t big enough, or solvable enough, to justify the investment. It sounds good and makes people go “woah” when they see it, but are autonomous vehicles really solving a problem? I’d argue no.
Healthcare dipped from its 2021 peak but still significantly outperformed autonomous vehicles. After years of hype about AI revolutionising medicine, it seems the market is getting realistic about regulatory hurdles and implementation challenges, but hasn’t given up entirely.
The implication: we’re entering an arms race for the picks and shovels of AI deployment. The smart money is backing what others need to build on, not necessarily what makes the best headlines.
What does all this mean?
Let’s be realistic about what’s happening, because the patterns here look suspiciously familiar.
We’ve seen corporate investment peak and crash, M&A drop sharply from its 2021 highs, and money flow toward infrastructure plays. To me, this looks less like natural market evolution and more like the post-COVID investment reality check hitting AI. 2021 was peak “growth at any cost” – cheap money, digital transformation hysteria, everyone throwing cash at anything that might scale. Then investors remembered that businesses should actually be profitable.
AI isn’t going to make jobs vanish overnight. But it is changing what skills matter. The most valuable people may not be AI engineers, but AI orchestrators: people who understand the technology, can work with it effectively, and direct it toward real business outcomes.
For startups, the window for launching “yet another AI tool” might well be closing fast. The opportunity now is in embedding intelligence into existing workflows, owning specific verticals, and building infrastructure that others depend on.
I don’t think the AI boom is over by a long shot, but it is growing up. And in mature markets, generalists get squeezed. Specialisation, speed, and domain knowledge are your edge now.
The patterns are clear if you’re willing to look past the hype. The question is whether you’re building for the world as it is, or the world as the headlines suggest it might be. Because if the last few years have taught us anything, it’s that reality has a way of reasserting itself.
Before you go, can you help me out?
Nearly 1100 people have subscribed to this newsletter in ~6weeks, which makes me feel that it’s resonating.
If you found this useful, here are three ways you could support me:
- Share it – Forward this to a founder, hiring manager, or job seeker who you think might find it useful and/or interesting.
- Follow StartUpHiring on LinkedIn if you’re not already – https://www.linkedin.com/company/startuphiringaustralia
- Subscribe on Substack – Get early access to podcasts, templates, AI prompts, and bonus content at https://startuphiring.com.au/
- Send feedback – What topics would help you most? What’s working? What isn’t? Just hit reply.
I’m building this into the go-to resource for startup hiring. Whether you’re hiring your first employee or looking for your next role, I want this to be genuinely useful for you.
These small actions make a massive difference in helping me reach more people who need this content.
Thanks for being here. Have a brilliant week.
Weekly newsletter drops
- Monday 10:30am – Monday by the Numbers: Data-driven insights on startup hiring trends
- Wednesday 1:30pm – The Main Issue: Comprehensive guide with advice for teams & talent, plus The List and latest podcast
- Friday 10:30am – Hot Takes Friday: My unfiltered opinion on something
Need hiring help?
If you’re struggling with a specific hiring challenge, I still offer 1:1 strategy calls through my consultancy, Crew. Book a 15-minute clarity call here
Simon
