Giant Pension Funds Pair AI With Humans and Risk Juniors

The Thinking Ahead Institute says the world’s largest asset owners are building a hybrid they call HI x AI, pairing human judgment with machine speed. Norway’s Government Pension Fund Global ranked as the largest in 2025, at $2.11 trillion (21,268 billion kroner).

Staff there told an internal survey they get an average 20 percent output lift from AI tools. The Institute’s warning sits next to that number: if the tools eat junior work, the humans who are supposed to stay in charge may never be trained.

The Hybrid Model the Top Funds Are Betting On

Jessica Gao, director at the Thinking Ahead Institute, has been blunt about the slogan. The future investment model is likely to be a hybrid one, the Institute writes in its work on the world’s top asset owners, done with a U.S. investment publication. AI does not replace the call. It feeds the people who still have to make it.

That is already more than a white paper. Eleven of the top 20 asset owners were implementing AI in 2025, and 12 were adding other new technologies. In a separate TAI exercise, the Global Asset Owner Technology and AI Peer Study, Gao said around 58% of participants already describe AI as integral to their long-term strategy, and 67% expect large industry-wide benefits over the next five to ten years.

THE HYBRID SCOREBOARD

• Largest fund: Norway’s Government Pension Fund Global, $2.11 trillion in 2025.
• Top-20 rollout: 11 of the top 20 implementing AI in 2025, with 12 adding other new tools.
• Peer conviction: 58% call AI integral to long-term strategy; 67% expect industry-wide benefits in five to ten years.

Those figures travel well in a slide deck. They do not say who still reviews the output, who trains the next reviewer, or who pays the bill when a model is wrong. Gao’s own test is narrower than the adoption count: do not switch the tools on until the fund can say how staff and software will work together.

Norway’s 20 Percent Comes With an Asterisk

Norges Bank Investment Management, which runs the oil fund, made 2025 a forced march. It launched Tech Year 2025 with mandatory training, hackathons, workshops and technology days at every office, plus a network of 71 AI ambassadors. The average productivity increase of 20 percent comes from that campaign’s own staff survey, published in the 2025 annual report: all employees now use AI tools in daily work.

Nicolai Tangen, chief executive of NBIM, took the same number to parliament on 5 May 2026. All 680 employees use AI daily, he said, and more than half use coding tools and write code themselves. The latest internal survey, in which staff report their own usage, already shows the fund has reached its target of a productivity gain of over 20 percent.

This is not because we have replaced people with machines, but because our people are using AI as a tool and doing their jobs better. Portfolio managers analyse more data. Legal staff process cases more quickly. We meet more companies and make better decisions.

Nicolai Tangen, CEO, Norges Bank Investment Management, Storting hearing, 5 May 2026

A month later his deputy put a pin in the statistic. Trond Grande, deputy chief executive of NBIM, said the 20 percent figure is not a very scientific number. It is more an ambition, he said, and more an expression that the tools should mean something real for the way the fund works. The effect NBIM has tried to measure is what staff themselves guess when they use the software.

HOW THE OIL FUND WIRED AI IN

1. 2018: The fund moves onto a cloud IT platform and starts years of cleaning and structuring its data, the base Tangen now says makes responsible AI use possible.
2. 2025: Tech Year 2025 rolls out mandatory training, hackathons and a 71-person ambassador network; staff complete over 3,500 digital courses, with 10,000 more courses on tap.
3. Year end 2025: The annual report says every employee uses AI tools daily and the internal survey prints the 20 percent lift.
4. 5 May 2026: Tangen tells the Storting that 680 people use AI every day and that trading costs fell 30 percent from 2023 to 2025.
5. June 2026: Grande tells a seminar the 20 percent print is an ambition, not a lab result.

The fund still had a strong year on the market side. It returned 15.1 percent in 2025, or 2,362 billion kroner, 0.28 percentage point under its benchmark, with 71.3 percent of the book in equities. Tangen is careful to split two AI stories that get mixed up in headlines: the tools inside the building, and the fact that AI enthusiasm helped lift the stocks the fund already owned. Over five years NBIM earned NOK 1,614 billion from the seven largest U.S. technology companies. At the end of 2025 the 10 largest companies, seven of them U.S. tech names, were 21.3 percent of the equity book, a concentration he called new in the fund’s history.

Where the Human Still Makes the Call

Gao’s rule is simple enough to print on a risk memo. Asset owners should not deploy AI unless they have a strategy for how employees and the technology will work together, and unless they make clear where the decision lies and where human overview must stay in the process. They also have to watch usage and cost, and they have to keep a talent pipeline.

GAO’S FOUR GATES BEFORE THE TOOLS GO LIVE

• A joint working plan: Staff need to know how the model fits the actual job, not only that a licence has been bought.
• A named decision owner: The file has to say who still makes the call and where a person must review the machine.
• A usage and cost watch: Prompts, seats and bills need the same scrutiny as any other vendor line.
• A live junior path: If AI takes the work that used to train new staff, the fund has to rebuild that path on purpose.

The Institute has been using the HI x AI phrase for more than a year. In a 2024 peer study of 26 large asset owners, together running $6.3 trillion, it already treated the stack as the way funds would cope with uncertainty. Attracting and retaining staff ranked as the second most pressing issue those funds named, behind complexity and workload. The quality of people and technology, and how those two strands join, was the path it said they would have to walk.

A later TAI look at 176 asset managers, representing about $39 trillion, found high conviction that human intelligence combined with explainable AI will separate firms. It also found the money still follows people. Technology takes 12% of operating expenditure on technology, against 38% on frontline investment talent. That split is the hybrid in budget form: the industry talks as if software will swallow the shop, then keeps paying for the humans who are supposed to remain accountable.

AI Makes the Junior Years Harder to Get

Gao’s sharpest line is about the people who are not in the room yet. If AI replaces the work done in more junior roles, she said, the junior has a hard path to become mid-level and senior, because that is years of experience accumulating to support them on that path. Funds should focus on how to keep that career track and change the learning process under the new tools.

The junior has a hard path to become mid-level and senior, because that’s years of experience accumulating to support them to go down that path.

Jessica Gao, Director, Thinking Ahead Institute

Tangen can say, correctly, that NBIM has not swapped 680 people for a model. Headcount held. The quieter change is inside the day. When a language model drafts the first pass on a filing, a meeting brief or a risk note, the hours a new analyst used to spend doing that pass are the hours that used to teach pattern recognition. Graduate-level analysis is the first work these tools absorb, and that is the same apprenticeship year a pension fund used to buy with a first job.

Leadership advisers who sit with asset owners are seeing the same shift without a collapse in junior hiring. Across those funds, junior hiring is not shrinking, but the bar is. One leader put it as the bar has changed, not the need. What a junior role is for, when AI handles more of the first-cut analysis, is now an open design question. The hire they want is less a person who already knows today’s tool, which will move, and more a person who can learn the next one.

That is a polite way of stating Gao’s problem. You can keep the same number of desks, print a 20 percent lift, and still thin out the years that turn a junior into someone fit to overrule a model. The hybrid then becomes a slogan over a shop that is top-heavy in title and thin in trained judgment. Tangen’s own defence, that people are doing their jobs better rather than being replaced, only holds if the next cohort still gets a job that teaches the job.

Eleven of the Top 20 Have Turned AI On

The Institute’s last full public ranking of the top 300 pension funds’ record assets put that group at $24.4 trillion at the end of 2024, up 7.8 percent on the year. The top 20 held 42.4 percent of that total. Those are 2024 books. The 2025 snapshot Gao is walking around is about behaviour, not a new league table: 11 of the top 20 were putting AI to work, and 12 were layering in other new technology.

WHO IS ACTUALLY USING THE TOOLS

The gap between those rows is the story. Norway can make every employee touch a model because it spent years on data, then a full year on training, then told parliament the headcount is 680 and the tools are not optional in practice. Most of the rest of the top 20 are still in the implementing column. Gao’s 58 percent is a strategy claim, not a completed build. The manager study’s 12 percent tech spend shows how much of the industry is still paying for people first and wiring the stack second.

That order is rational until the junior work disappears faster than the training plan arrives. A fund can be honest when it says AI is integral to the five-year plan and still have no answer for who will be qualified to sit over the model in 2032.

Trading Costs Fell While Headcount Held

The cleanest number Tangen has is not the survey. Between 2023 and 2025, through a range of measures of which AI was a large part, NBIM reduced trading costs by 30 percent, equal to NOK 4.8 billion. Management costs over 2013 to 2025 have run at 0.05 percent of assets, a line the executive board likes to hold up against other large funds. Those are the kind of savings a board can audit.

The 20 percent output claim is a different species. It is a staff self-score after a year in which using the tools was the culture. Grande has now said out loud that it is an ambition. Tangen still uses it as proof that the fund is ahead, and as proof that it did not fire its way to the result. Both things can be true in the same building: the cost line moved, the survey moved, and nobody can yet say what a 25-year-old analyst will know in five years if the first-cut work never crosses their desk.

Gao’s list is the practical response, and it is heavier than a licence budget. Name the decision owner. Watch the bill. Keep a path from junior to senior that still accumulates judgment. Norway has shown that a 680-person shop can put a model in every workflow and print a 20 percent lift on a questionnaire. The next test, the one the Institute is actually selling under the HI x AI label, is whether that shop can still grow the people who are allowed to say the model is wrong.

Disclaimer: This article is news reporting and analysis of how large asset owners are using AI inside their organisations. It is informational only and is not investment, pension, legal or career advice. Readers should consult a qualified financial adviser, investment professional or employment lawyer before acting on any fund, savings or workplace decision. Figures, rankings and programme details reflect the fund reports and research cited for the dates given in the piece and may change as later surveys and accounts are published.