"Change drives growth": the lesson from the Scottish Mortgage investment trust for the AI era

A listed investment company does not win on the average of its holdings. At Scottish Mortgage, a handful of very large winners deliver most of the performance. On 23 September, at its annual conference, the Baillie Gifford trust added one more criterion to that rule: how fast a company can transform itself. From SpaceX's rockets to Chinese batteries and artificial intelligence agents, the same lesson came back for an hour and forty minutes.

A growth rule: a few winners do almost everything

In a growth portfolio, returns are not spread evenly. A holding can at worst lose what was put in. Another can multiply it by ten, or by a hundred. The whole craft of management plays out on that asymmetry.

Tom Slater, one of the two managers of Scottish Mortgage, spells out the implication: the trust's performance rests on a small number of very large winners. Many investments will not develop as hoped. The job is therefore to let the successful ones run.

But what do you do when a winner becomes enormous?

SpaceX provides the answer. Scottish Mortgage first invested in 2018, with about $200 million. At the IPO in June 2026, the holding was worth close to $5 billion, according to the trust. At the end of June, it represented more than a quarter of the portfolio, a level of concentration the trust itself described as very unusual.

"We don't equate large positions with risk," says Tom Slater. Selling too early would have kept positions of equal, more reassuring size. It would also have cut the return. Now that SpaceX is listed, the trust can sell, in tranches, as selling windows open. The weight is down to just over 15%, and more selling windows will open in the coming weeks.

Letting winners run does not rule out selling. Tesla shows the other side. After thirteen years, Scottish Mortgage sold the rest of its holding. It had made about $6 billion for investors, 13 times the capital invested. But, according to the trust, its valuation now reflected too much of the future.

Netflix also left the portfolio as its growth matured. Meanwhile, the trust added to higher-conviction ideas: Nvidia, Cloudflare, the digital bank Nu and the e-commerce group Sea.

When a cost collapses, markets are born

Why keep so much SpaceX when Lawrence Burns, the other manager, himself acknowledges high starting multiples? Because, in his view, the company is driving down the price of a scarce resource: access to orbit.

Before SpaceX, putting a kilogram into orbit cost $18,000 to $19,000. With Falcon 9, the list price is around $2,700. Starship, still in testing, targets about $900. Once reusable, it could go down to around $200, according to Lawrence Burns. The trust's July update even mentioned about $100.

Each step down this staircase opens a new business. The first is called Starlink: internet access through thousands of satellites. According to Lawrence Burns, it already generates $17 billion of annualised revenue, up almost 70% in a year, with an operating margin close to 40%.

If SpaceX builds several businesses the size of Starlink, the investment can still pay off well over ten years, Lawrence Burns believes. Opulion has covered SpaceX's new hosted-compute business and what the listing changes for Baillie Gifford's trusts.

The lesson goes beyond space. To spot tomorrow's big winners, look for costs that are collapsing. That is where markets are born that nobody is measuring yet.

Artificial intelligence, from the power socket to the application

AI is the portfolio's other major theme. The trust describes it as a chain, from electricity to end uses, and it is present at every layer.

Anthropic shows how fast things move. The trust invested in 2025, when the company was worth about $200 billion. According to the trust, it is now worth nearly $1 trillion and ranks among its top ten holdings. Opulion has analysed Anthropic's place in the two trusts.

Where will durable value settle? Lawrence Burns gives two answers.

First, with chipmakers and their equipment suppliers. Whatever the winning application, and whatever the winning model, far more chips will be needed. He sees in them a kind of "royalty" on the whole development of AI.

Second, with companies able to reorganise quickly. He cites a founder who knows how to pivot, like Tobi Lütke at Shopify, a culture that makes it possible, the best talent. And sometimes data that nobody else has, like MercadoLibre's on payments and credit.

So should we fear a bubble?

Tom Slater does not dismiss the question. He starts from demand, which keeps growing. According to OpenAI, its largest customers consume about eight times more than the median company. The cloud giants are putting their cash into it, then debt and equity markets. The ride may be bumpy. But for him, volatility is inevitable: the real risk is the permanent loss of capital.

Ethan Mollick, a professor at Wharton and author of Co-Intelligence, keeps both scenarios open. A boom-and-bust cycle remains possible, as with canals, railways or fibre optics. Even then, AI will not go away: it is useful, and cheap to run once the infrastructure is built. This moment could even look like a second industrial revolution.

China is judged on the ground

Scottish Mortgage has invested in China for almost twenty years. Its exposure has fallen from 24% at the end of 2020 to about 11% today. The trust sees it as a higher bar for investment, not a retreat.

Linda Lin, who heads Baillie Gifford's China equities team, has an image for those who read China through its headlines: it is "like judging a restaurant from the weather outside". Property is suffering, households are cautious, geopolitics is tense. But in green technology, advanced manufacturing, robotics and AI, the country now sets the pace.

The examples are concrete. BYD's Seagull costs about £8,000 in China. The carmaker makes its own chips, motors and batteries, and its fast charging adds 400 kilometres of range in five minutes. Lawrence Burns says he is "very worried" about Europe's mass-market car industry.

CATL supplied almost 40% of the world's electric-vehicle batteries last year. Tom Slater sees in it a rare culture: walking away from what worked last year if a better solution appears. Horizon Robotics, China's leader in autonomous-driving chips, completes the picture. Linda Lin has known its founder, Kai Yu, since he worked at Baidu. The relationship opens the door, she sums up, but research decides.

She also describes two ways of doing AI. American labs aim for the most powerful models. China bets on open, low-cost models, to spread AI into factories, schools and public services. ByteDance, the trust's largest Chinese holding, has both its own models and global distribution.

How does the trust manage political risk? On two levels. For each company, it builds scenarios: for BYD, it works mostly on the assumption that not a single passenger car will be sold in the United States. For the portfolio, it reduces the shared exposure, because American restrictions on capital would hit every Chinese holding at once.

One manager adds a nuance that many investors forget. Owning no Chinese company does not protect you from China risk. Tesla and Apple depend heavily on China, as a market and as a factory, without their share prices seeming to reflect it.

AI amplifies those who adapt

Everyone has access to the same AI models, at the same price. So why are the gaps between companies widening?

For Ethan Mollick, the difference lies less in data than in processes: how AI fits into work, and who makes the decisions. Lawrence Burns hears the same from the leaders he meets. AI is not a tool that levels, it is an amplifier of the best organisations. In a regulated industry, he was told, two years to approve a decision means several generations of AI.

The word "agile" was used by Ethan Mollick himself. Being "nimble and fast-moving, agile or ambidextrous, however you want to call it" is incredibly important. And nobody knows anything for sure, says someone who talks constantly with AI labs, governments and business leaders. So there is no point waiting for a ready-made solution: experiment, and keep what works.

The target keeps moving. For software developers, value has shifted from writing code to directing agents and writing tests. That new bottleneck will probably be solved in turn, and value will move again.

To tell whether a company really benefits from AI, Ethan Mollick announces four points. He puts two of them as questions and the other two as recommendations.

Why are we so often caught off guard? Hannah Fry, a mathematician at the University of Cambridge, gives the deeper reason. Our brains were shaped in a world that changed little. So we extend the current trend in a straight line, while some technologies accelerate on their own. Tom Slater spells out the implication: we underestimate how much everything will change.

And jobs? Tom Slater reports a message he often hears from business leaders: they worry less about too many humans than about a shortage. It leads him to a hypothesis. Perhaps fewer employees in large companies, but many more small companies, because a small, well-equipped team can now achieve far more.

What this conference teaches the reader

Opulion's reading. To read any growth investment company, three questions are enough. Where does its performance come from: a few holdings, or all of them? What does it do with its winners: does it let them run, does it know how to sell them? And can the companies it owns transform themselves faster than their market?

The last question goes beyond investing. Lawrence Burns says it: it is not just organisations but society as a whole that must adapt. Ethan Mollick insists: we must not be passive. We draw a simple conclusion. Agility is no longer a personality trait. It is becoming a criterion, for a company as for an investor.