How CalPERS Is Reinventing a $600 Billion Portfolio for the AI Era

Stephen Gilmore, investment chief at CalPERS, is dismantling the architecture it has operated under for decades. What it is building in its place may define institutional investing for the next generation.

Background: The Institution Behind the Experiment

To understand what CalPERS is attempting, you first have to understand what CalPERS is — and what it carries.

The California Public Employees' Retirement System is not simply a large investment fund. It is the financial backbone of retirement security for approximately two million Californians: teachers, firefighters, nurses, administrators, and public servants whose professional lives were built on the promise that their pensions would be there when they needed them. It manages roughly $600 billion in assets. It is the largest public pension fund in the United States. And it is, by the nature of its obligations, one of the most conservatively governed pools of capital on the planet.

That context matters. Because what investment chief Stephen Gilmore is doing at CalPERS is not the kind of experiment a hedge fund or a venture capital firm runs. Those institutions can move fast and absorb the cost of being wrong. CalPERS cannot. The stakes of failure are not a fund vintage that underperforms or a portfolio company that does not exit. The stakes are the retirement incomes of millions of people who spent their careers in public service and have no alternative safety net.

Which makes what Gilmore is doing all the more striking. He is proposing — and actively implementing — one of the most fundamental structural changes in the history of the institution. He wants to break down the walls between asset classes entirely.

The Architecture of the Problem

For most of the past century, large institutional investors like CalPERS have organised themselves around asset class categories. You have an equities team. A fixed income team. An alternatives team — private equity, real estate, infrastructure, hedge funds. Each team has its own mandate, its own benchmark, its own risk budget, its own performance evaluation framework. Each team optimises within its lane.

This structure was not arbitrary. It emerged for good reasons at a particular moment in the history of capital markets. Asset classes behaved differently from each other. Equities and bonds were negatively correlated for long stretches — when stocks fell, bonds rose, and the portfolio held. The diversification benefit of holding both was real, measurable, and reliable enough to build a governance structure around.

It also made the complexity of managing a multi-hundred-billion-dollar portfolio comprehensible to boards, beneficiaries, regulators, and the public. If you could say "we have X percent in equities, Y percent in fixed income, Z percent in alternatives, and here is how each of those performed against its benchmark," you had a governance framework that was defensible, auditable, and legible to people who were not professional investors.

The problem is that the world those categories were designed to describe has changed. Substantially and in ways that the traditional structure is increasingly poorly equipped to handle.

The correlation between equities and bonds — the foundational assumption of the 60/40 portfolio and the diversification logic that underpins most institutional asset allocation — broke down dramatically in 2022 and has not fully recovered. Interest rate cycles, inflation dynamics, and the growing integration of global capital markets have created environments where the hedging relationship between stocks and bonds simply does not behave as historical models predicted.

Simultaneously, the boundaries between asset classes have blurred in ways that the old categories cannot capture. Infrastructure increasingly looks like fixed income in its cash flow profile but behaves like private equity in its liquidity and governance characteristics. Private credit has grown to the point where it competes directly with public fixed income for capital allocation. Real assets carry inflation-hedging characteristics that belong in a macro risk framework but are typically managed by teams that do not talk to the macro team.

The result, in a fund the size of CalPERS, is a governance structure that is actively preventing the integration of information and the cross-category analysis that modern markets require. Teams sitting in separate silos, managed against separate benchmarks, with separate risk budgets, cannot easily identify the opportunities or the risks that live at the intersection of those silos. And increasingly, that is where the most important signals are.

The Total Portfolio Approach: What It Actually Means

Gilmore's response to this structural problem is what he calls "total portfolio" investing. The idea, which has been adopted in different forms by a small number of leading sovereign wealth funds and endowments — most notably the Future Fund in Australia, where Gilmore previously served as CIO — treats the entire portfolio as a single integrated system rather than a collection of separately managed sub-portfolios.

In a total portfolio framework, capital allocation decisions are made at the portfolio level, not the asset class level. The question is not "how much should we allocate to equities versus fixed income?" but "what combination of positions, across all available instruments and asset types, produces the best risk-adjusted return for the total portfolio given our liability profile, our liquidity needs, and our investment horizon?"

That sounds like a subtle shift. In practice, it is an architectural revolution.

It requires dismantling the siloed team structure and replacing it with an integrated investment function where analysts, portfolio managers, and risk officers work across categories rather than within them. It requires building a unified data infrastructure that can aggregate and analyse positions, risks, and opportunities across all asset classes in real time — something that most large institutions currently cannot do, because their data systems were built to serve siloed teams and produce category-level reports, not integrated portfolio-level intelligence.

It requires redesigning performance measurement and accountability frameworks that have historically been built around how each category performs against its benchmark. In a total portfolio world, the benchmark is the total portfolio return against its liability obligations — and attributing that return to individual decisions and individual teams is both more complex and, ultimately, more honest than asking whether the equities team beat the MSCI World Index.

And it requires, perhaps most critically, building the analytical capability to identify and act on the cross-category relationships and opportunities that the total portfolio framework is designed to surface. This is where artificial intelligence becomes not a support tool but a foundational capability.

The Role of AI: Not a Feature, but the Foundation

The total portfolio approach that Gilmore is implementing at CalPERS is, at its core, an information integration challenge. The question it is trying to answer — what combination of positions across all available instruments produces the best risk-adjusted return for the total portfolio? — requires the simultaneous processing of more data, across more dimensions, with more complex interdependencies, than any human team can manage without technological assistance.

This is precisely the kind of problem that AI is structurally well-suited to address. Machine learning models can identify relationships between asset classes, geographies, sectors, and macro variables that are not visible through conventional quantitative analysis. They can process real-time market signals, news flows, regulatory developments, and economic data simultaneously and flag the implications for portfolio positioning faster than any analyst team.

More specifically, agentic AI — systems that do not just analyse but act, within defined parameters and governance guardrails — can execute the kind of continuous portfolio monitoring and rebalancing that a total portfolio approach requires. In a siloed structure, rebalancing is a periodic, deliberate process. In a total portfolio structure, the portfolio is understood as a dynamic system that requires continuous calibration as market conditions evolve.

Companies with the best AI-driven financial outcomes are nearly twice as likely as other companies to say they are using AI in advanced ways — executing multiple tasks within guardrails — and 1.9 times more likely to be operating in autonomous, self-optimising ways. CalPERS, in adopting the total portfolio approach with an integrated AI layer, is making exactly this bet: that continuous, AI-assisted portfolio optimisation across a unified data infrastructure will produce better risk-adjusted outcomes than periodic, human-led rebalancing within siloed category structures.

The data infrastructure required to support this is itself a multi-year project. CalPERS manages positions across public equities, fixed income, private equity, real estate, infrastructure, and inflation-linked assets, held through a combination of direct positions, external manager mandates, and derivative overlays. Aggregating all of that into a single, real-time view of total portfolio risk and return — integrated with the external market data, economic indicators, and alternative data sources that an AI layer needs to function effectively — is a genuinely complex engineering challenge.

It is also a governance challenge. A $600 billion fund does not get to experiment with AI-driven autonomous decision-making without a governance framework that the board, the beneficiaries, regulators, and the public can understand, scrutinise, and hold accountable. Building that framework — defining the parameters within which AI can act autonomously, the thresholds at which human oversight is required, and the accountability structures for decisions made by the system — is as important as the analytical capability itself.

The People Question: What Total Portfolio Investing Requires From Human Talent

One of the most underappreciated dimensions of what Gilmore is attempting is the talent transformation it requires. The analysts, portfolio managers, and risk officers that CalPERS has historically employed were developed — and hired — to be deep specialists within specific asset class categories. The equity team knows equities. The real estate team knows real estate. That specialisation is valuable and not being discarded.

But the total portfolio approach requires a different layer of capability on top of that specialisation: the ability to think across categories, to understand how a position in one part of the portfolio interacts with positions in another, to speak the language of total portfolio risk and return rather than category-level benchmark performance. That is a different skill set, and it does not exist in abundance in the talent market.

PwC's 2026 Global AI Jobs Barometer found that skills needed for the most AI-exposed jobs are changing more than twice as fast as for the least AI-exposed jobs, and the most AI-exposed junior roles are seven times more likely to demand traditionally senior skills like leadership, strategic thinking, and stakeholder management. CalPERS is encountering this dynamic in its own talent strategy. The people who will thrive in a total portfolio, AI-integrated investment environment are not those with the deepest specialisation in a single category. They are those who combine solid domain expertise with the analytical range to work across the portfolio and the judgment to know when AI-generated signals warrant action and when they require human override.

Building that capability internally — through hiring, through reskilling, through changes to how analysts are developed and evaluated — is a parallel track to the structural and technological transformation. And it may, in the long run, be the harder one. Technology can be procured. Data infrastructure can be built. Culture, judgment, and the capacity to think across artificial boundaries that have existed for decades are developed over time, through deliberate investment in people who are given the mandate and the support to think differently.

Industry Context: The Broader Shift in Institutional Investing

CalPERS is not operating in isolation. The total portfolio approach it is implementing reflects a broader movement across the most sophisticated institutional investors in the world — a recognition that the traditional asset class framework, built for a different information environment and a different market structure, is no longer optimally serving the purpose it was designed for.

The Future Fund in Australia — where Gilmore developed the total portfolio methodology before joining CalPERS — has operated on this basis for years and has consistently delivered returns that outperform peers operating with conventional category structures. The Government Pension Investment Fund in Japan, the largest pension fund in the world at approximately $1.5 trillion, has been moving in a similar direction. Several major sovereign wealth funds in the Gulf and in Asia have adopted integrated portfolio frameworks that treat cross-asset relationships as a primary input to allocation decisions rather than an afterthought.

The common thread across all of these moves is the recognition that AI is delivering on efficiency and productivity, and twice as many leaders as last year are reporting transformative impact — but that capturing that impact requires structural change, not just technological adoption. You cannot generate the value of integrated, cross-category intelligence by deploying an AI tool on top of a siloed organisational structure. The structure itself has to change.

For most institutional investors, that realisation is still arriving. Only 25% of organisations have reached the Scaling phase of AI adoption, while the largest share — 47% — is still in Piloting, and 28% remain in the Understanding phase. The pension fund industry mirrors this pattern. Most large funds are running AI pilots in specific functions — risk analytics, manager selection, ESG scoring — without the integrated data infrastructure or the organisational design that would allow those tools to function as a coherent intelligence system.

CalPERS, in committing to the total portfolio approach with integrated AI as its foundational assumption, is making a bet that it can move from the 47% stuck in piloting to the 20% capturing the majority of the value. The scale of the commitment — $600 billion in assets, multi-year implementation timelines, fundamental organisational redesign — means this is not a hedge or a proof of concept. It is a strategic conviction.

Risks, Challenges, and the Honest Assessment

An honest account of what CalPERS is attempting cannot ignore the genuine risks and challenges involved.

The most immediate is implementation complexity. Restructuring the organisational design, data infrastructure, governance frameworks, talent model, and investment processes of a $600 billion fund simultaneously, while continuing to meet daily obligations to beneficiaries, manage existing manager relationships, and operate under the public scrutiny that a government institution faces, is an extraordinary management challenge. The history of large-scale institutional transformations is littered with initiatives that were sound in concept but poorly executed in practice.

The second is the governance challenge specific to a public institution. CalPERS operates under a level of transparency and accountability that private funds do not face. Its board meetings are public. Its investment decisions are subject to political scrutiny. Its performance is reported against benchmarks that the public and the media can evaluate. Explaining a total portfolio approach — which is inherently more complex and less legible than a category-based framework — to a board, to beneficiaries, to legislators, and to the public requires a communications strategy as sophisticated as the investment strategy itself.

The third is the model risk inherent in any AI-driven investment system at scale. Machine learning models trained on historical data can identify relationships that have existed in the past. They cannot guarantee that those relationships will persist in the future, particularly in the kind of tail-risk scenarios — a sudden geopolitical shock, a regime change in monetary policy, a liquidity crisis — where historical patterns are least reliable and human judgment is most needed. Building robust human override mechanisms, clear escalation protocols, and regular model validation processes into the governance framework is not optional. It is the difference between a system that amplifies human judgment and one that substitutes for it at the worst possible moment.

The Lesson for Every Organisation

The CalPERS story is about a pension fund. But the lesson it carries belongs to every organisation navigating the intersection of artificial intelligence and structural change.

The most common failure mode in enterprise AI is deploying sophisticated analytical tools on top of organisational structures that were not designed for the insights those tools generate. The AI produces signals that nobody has the authority or the mandate to act on. The data is integrated at the analytical layer but siloed at the decision-making layer. The technology becomes an expensive reporting tool rather than a genuine driver of better outcomes.

CalPERS is attempting the harder thing: redesigning the organisational structure around the capability the technology enables. Breaking down walls that have existed for decades. Redefining roles, responsibilities, and accountability frameworks to match the integrated intelligence a total portfolio approach requires. And doing all of it under public scrutiny, with the retirement incomes of two million people as the ultimate measure of whether it works.

PwC's analysis shows that capturing growth opportunities from industry convergence is the single strongest factor influencing AI-driven financial performance, ahead of efficiency gains alone. For CalPERS, that convergence is between asset classes. For a manufacturer, it might be between supply chain, product development, and customer analytics. For a healthcare system, it might be between clinical data, operational systems, and financial planning. The specific silos differ. The principle is the same.

The organisations that will define the next decade — in financial services, in technology, in healthcare, in retail, in energy, in every sector covered by The Time Global — are those that are willing to ask a question that most are still avoiding: not "how do we use AI to do what we already do, better?" but "if we assume AI from the start, what should we be doing differently?"

CalPERS has asked that question. The $600 billion experiment has begun. And the answer, when it comes, will be one of the most closely watched results in institutional investing history.

*Sources: Bloomberg, CNBC, PwC 2026 AI Performance Study, Deloitte State of AI in the Enterprise 2026, PwC Global AI Jobs Barometer 2026, BCG AI Radar 2026, Stanford AI Index Report 2026.*