“Everything Up to 2027 Is Already Priced In”

"It's as if there's an intelligence within language itself. We created language, but language seems to carry intelligence."

Cyla Research — A retired AI investor on the long buildup nobody noticed, the breakthrough that convinced him, and where the opportunity actually lives.

Michael spent his career in tech hardware, much of it in semiconductors. He rode the industry through the internet, consumer electronics, phones, and the mobile internet, and when he retired he used that background to catch the HBM memory boom — the chips that feed AI data centres. Now he invests in AI full time. I sat down with him for half an hour to ask the question most of us are quietly asking: is it too late to get in?

The setup nobody watched

His first move was to push back on the premise that AI arrived suddenly at all.

"People say 'now it's the AI era,' but this process started long ago," he told me. Twenty years back, computers could already read handwritten Chinese characters — badly, but they could. IBM built speech recognition that dazzled in the lab and flopped in the real world, defeated by accents and dialects. Then machines conquered chess, and later Go, the game he considers the highest form of human strategic play. "It's like a movie — there's always a setup. Things don't explode out of nowhere."

What changed with large language models, in his view, wasn't the intelligence. It was the door. The system that beat Go was a narrow tool most people could never touch — "most people can't play Go." Language is different. "If you solve a problem through something everyone can use, then everyone's in."

The moment he became a believer

He paid for his first AI subscription in early 2023 and treated it like a lab. The moment that converted him from curious user to investor came about a year later, when he sat with an Oxford PhD student and fed the model questions about the student's own cutting-edge research. The answers were startlingly current — the machine was operating near the frontier of a specialist field. But the deeper shift, he says, was long context: the model's ability to hold an entire conversation, twenty exchanges deep, and stay coherent. "It's as if there's an intelligence within language itself. We created language, but language seems to carry intelligence." Early models hallucinated and lost the thread. Then, gradually, they didn't. "That's when I felt AI had produced something like real intelligence."

One breakthrough a year

From there he sketched a timeline that doubles as an investing map. 2024 was the year of long context — suddenly everyone was summarising, drafting, reporting. 2025 was reasoning — models that could pull a hundred websites and think over them for a long stretch (his family used one to pick a hotel everyone actually liked). 2026 is agents — AI that carries out long, complex tasks through code. "When I first retired I tried to write AI programs and simply couldn't. Now it's no problem at all."

He's watched the research showing that roughly every seven months, the length of task AI can handle doubles. Will it continue? Here he reached for an analogy any parent recognises: a teenager growing ten centimetres a year. "This year ten, next year ten. But we all know a person can't keep growing forever. At some point it stops."

So is it too late?

Which brings us to the money question. Is AI priced in?

"A lot of AI stocks really are already priced in — no doubt about it," he said. His sharper claim: markets have largely absorbed the growth story through the end of 2027, because that's where most forecasts see the curve flattening. "How to invest for 2028 and beyond is genuinely challenging. There's no clear conclusion right now."

But he bristled at the phrase "chasing the top." "For a stock that's going to rise ten times, going up two or three times first is completely normal. A good one doesn't ten-x in a day — it zigzags." Waiting for a dip that never comes, he suggested, is its own way of losing.

His actual advice for ordinary people was strikingly un-flashy. Invest where you have an edge: if you spent years building generators, or working in semiconductors, or running any specific industry, you know things about how AI will hit that industry that outside investors don't. And never invest money you might need. "If you're investing money you'll need next month, that's the wrong money to invest." He flatly refused to name stocks or amounts.

Tractors, factories, and what comes next

On jobs, he was calmer than most commentary I read. Every tool has forced this migration — tractors emptied the fields and filled the factories; computers did it again. "You hand the machine what it can do, and humans go do what the machine can't." The catch is the condition attached: people need the ability to keep learning, to migrate with the tool. Those who can will be empowered. Some who can't may be replaced. "That's the challenge." He pointed to the new work already appearing — data labelling, data-centre construction, AI education — unglamorous, but real.

Seven hundred papers

And if he were thirty today, with no connections? "You absolutely have to get your hands on it. Buy a subscription, use the models, use the new features — start with the things related to you." Take whatever you already do — design, code, running a travel agency — and add AI's intelligence on top of it. Then he added the line I haven't stopped thinking about. In 2024, using AI as his reading partner, he worked through roughly 700 research papers. Seven hundred, in retirement. "You have to have volume first. Anything has to have volume before it yields anything."

The market may have priced in the next eighteen months. It hasn't priced in what you do with them.