The Pension Dilemma: When Technocratic Urgency Meets Technological Optimism

I thought I had made an airtight case.

Among friends who care about fiscal sustainability and intergenerational fairness, I would walk through the demographic arithmetic with confidence. Fewer workers supporting more retirees. Sluggish growth. Rising longevity. No serious political appetite to raise the retirement age. The numbers point in one direction. Something has to adjust. The math is brutal.

But what I encountered was not agreement. It was reluctance. Not denial of the figures, not confusion about the trends, just a deep discomfort with the conclusion. I found that frustrating. If we take sustainability seriously, why resist the implications?

Lately, though, I have begun to wonder whether that hesitation to reconcile with unsustainable fiscal policies by making cuts to popular pension programs had more merit than I initially allowed. Not because I care any less about fiscal sustainability. I still believe the demographic pressures are real and that pretending otherwise is irresponsible.

But I may have been too quick to treat the problem as a purely linear equation. Retirement systems are not just budget lines. They reflect decades of contribution and expectations of fairness. And economies do not evolve in straight lines. Productivity, labor force participation, immigration, and technology can shift the trajectory in ways that static projections cannot fully capture.

I am not abandoning the math. I am reconsidering whether arithmetic alone is enough to justify imposing near term burdens, especially on those who are already feeling economically squeezed.

The OECD’s 2025 Pensions at a Glance report lays it out plainly. The number of people aged 65+ per 100 working-age adults is projected to jump from 33 in 2025 to 52 by 2050. It was 22 in 2000. The working-age population across OECD countries will fall by 13 percent over the next 40 years. In ten countries, including Japan, Korea, Italy, Spain, and Poland, it will shrink by over 30 percent.

Public pension spending already averages 8.1 percent of GDP across the OECD, up from 6.7 percent in 2000. In Greece and Italy, it exceeds 16 percent. Without reform, the OECD’s Long-Term Model projects fiscal pressure increasing by nearly 6.25 percentage points of GDP between 2024 and 2060. The European Commission projects total ageing-related expenditure rising from 24.6 percent of GDP in 2019 to nearly 27 percent by 2040.

To maintain current commitments, governments face an impossible trilemma: raise taxes on a shrinking workforce, borrow at levels that crowd out other spending, or cut benefits and face the political firestorm. Most are choosing to borrow, deferring the reckoning. My position has been: raise retirement ages now, while adjustments can still be gradual, before we hit a wall in the 2030s that forces brutal cuts, the kind that breed populist revolts.

I still think that logic is sound. But I’ve started to wonder whether I’m solving for the wrong future.

In my research on AI geopolitics and the compute stack, I operate on the assumption that artificial intelligence will be transformative. Not like social media, which largely reshuffled existing value, but like electrification, which restructured what economies can produce.

When I look at the trajectory of frontier AI systems, it is difficult to see a linear story. Training compute for leading models has grown exponentially over the past decade. Algorithmic efficiency has improved alongside hardware, so performance gains are not solely dependent on brute force scaling. Systems that struggled with narrow tasks only a few years ago now autonomously write complex software, conduct multi step research, generate technical documentation, and assist in medical and legal workflows. The scope of tasks AI can reliably complete has expanded rapidly and continues to expand.

More importantly, intelligence itself is compounding. AI systems are increasingly used to improve AI systems, optimizing chip layouts, accelerating model training, generating code, and compressing research cycles. If intelligence becomes partially scalable through capital rather than solely through human labor, the structure of productivity growth changes. This is no longer just incremental efficiency improvement. It is the potential automation of cognitive labor at scale.

It converges with robotics, where improvements in perception, control systems, and dexterity are meeting falling hardware costs. It converges with energy technology, where solar costs have fallen roughly 85 percent since the early 2000s and battery costs nearly 90 percent over the past decade, lowering the marginal cost of compute and automated production. It converges with AI assisted discovery in biology, materials science, and manufacturing, shortening development timelines and accelerating innovation across sectors.

And here is where I began to notice a cognitive dissonance.

In fiscal debates, I default to technocratic caution. I assume baseline growth of 1 to 2 percent in advanced economies, consistent with post 2008 stagnation and demographic drag. Pension models rest on that assumption. Fewer workers imply slower output growth. A higher dependency ratio implies structural strain. That framework has been historically sound.

But in my work on AI and technological transformation, I routinely forecast nonlinear change. I argue that intelligence is becoming partially capitalizable. I suggest that productivity regimes can shift. I point to reinforcing exponential curves.

I was, in effect, holding two worldviews at once: technocratic pessimist in public finance, technologist optimist in frontier innovation.

If even part of the acceleration thesis materializes at scale, the implication is not marginal growth improvement. It is the possibility of a materially higher productivity regime, one in which sustained real growth above recent baselines becomes plausible for a period of time.

The demographic squeeze that drives the pension dilemma intensifies in the 2030s and 2040s, precisely when compounding AI capability, maturing robotics, and structurally cheaper energy would be expected to reshape economic conditions.

What if we are locking in politically costly reforms based on a growth baseline that exponential technological change is already rendering obsolete?

If you genuinely believe growth in the 2030s will significantly exceed current baselines, then aggressive pension reform today is not prudent stewardship, it is premature austerity. You would be inflicting real pain to solve a problem that accelerating growth may substantially mitigate.

This is not unprecedented. The United States emerged from World War II with debt-to-GDP of roughly 120 percent. It did not resolve that through austerity. It grew. Rapid postwar productivity gains expanded the denominator until debt-to-GDP fell below 40 percent by the early 1970s. If pension obligations grow at 2 to 3 percent and nominal GDP grows at 5 to 7 percent under AI-augmented scenarios, the same mechanism applies.

And the political cost of reform is not just a PR problem. France is the case study. In 2023, Macron raised the retirement age from 62 to 64 — modest, technocratically sound, endorsed by economists. He had to invoke Article 49.3 to bypass parliament. The result: months of strikes, collapsed approval ratings, and an acceleration of polarization that has made it overwhelmingly likely France’s next president will come from the far right or far left — both of whom promise to reverse the reform and pursue far more reckless agendas. The technocratic cure may prove worse than the disease.

This pattern is universal. Pension reform takes from a large, organized, electorally engaged constituency and gives to a diffuse, disengaged one. The backlash metastasizes into a generalized revolt against the governing class, benefiting precisely the political actors least equipped for responsible governance.

I would be intellectually dishonest if I did not confront the scenario where the bet fails.

We have been here before. In 1987, Robert Solow observed: “You can see the computer age everywhere but in the productivity statistics.” The IT revolution was visibly transforming offices across America, yet productivity growth slowed, from 2.9 percent annually (1948-1973) to 1.1 percent after 1973. This “Solow Paradox” persisted for over a decade before a brief 1990s acceleration that then faded.

The pattern may be repeating now. Goldman Sachs’ own chief economist acknowledged in February 2026 that AI investment has contributed “basically zero” to US economic growth so far. Apollo’s Torsten Slok was blunter: “AI is everywhere except in the incoming macroeconomic data.” A recent NBER study of 6,000 executives found positive AI claims not translating into aggregate productivity gains. MIT’s Daron Acemoglu estimates just 0.07 percent additional annual productivity growth — orders of magnitude below Goldman’s projections.

The optimists invoke a J-curve: heavy upfront investment, then exponential returns as organizations restructure. That was eventually true of electrification, which took 30 years to peak productivity impact. But “eventually” does a lot of work in that sentence, and for pension policy, timing is everything.

Suppose AI follows the internet’s trajectory rather than electricity’s — enormous consumer welfare gains poorly captured in GDP, massive value for a narrow set of firms, measured productivity growth closer to Acemoglu’s estimates than Goldman’s. Suppose robotics hits integration barriers. Suppose the energy transition faces grid and permitting bottlenecks.

In that world, the OECD projections are roughly correct. And we will have spent the late 2020s and early 2030s not reforming, betting on growth that failed to materialize. The reforms that could have been gradual if started in 2025 would need to be sudden and draconian by 2035. Debt would be structurally higher. The intergenerational transfer from young to old would reach levels that make today’s housing crisis look mild. And the political consequences would be exactly what the pessimist in me fears: populist revolts driven by a generation that rightly feels the system was rigged against them by leaders who gambled and lost.

I think the probability of transformative AI-driven growth arriving in time to materially mitigate the pension crisis is meaningfully above fifty percent. I think the convergence of AI, robotics, and cheap energy is structurally different from previous technology cycles. And I think the political costs of aggressive reform in the 2020s are real and underappreciated.

But betting a civilization’s fiscal solvency on a technological revolution arriving on schedule is one of the highest-stakes gambles imaginable. If growth disappoints, the adjustment required in the 2030s will be harsher than anything we are debating today.

So the right posture is adaptive reform.

Pension systems should be explicitly aligned with national capacity. Benefit growth, contribution rates, and retirement ages should adjust gradually based on transparent indicators such as long-term productivity growth, the old-age dependency ratio, structural debt levels, and the overall tax burden.

That would still be controversial. There is no painless path. But it is less destabilizing than front-loaded austerity calibrated to a stagnation baseline that may already be eroding.

And if even that cannot pass, if the political system cannot agree on adaptive guardrails, then I would rather take the growth bet than press for immediate structural cuts that large segments of the public do not perceive as urgent. France is a cautionary example. A modest, technocratically defensible reform triggered months of unrest and deepened polarization. The fiscal gains were marginal relative to the political damage.

In a decade likely to be defined by technological disruption and geopolitical strain, destabilizing democratic politics in the name of preemptive austerity may be the greater risk. If the choice is between front-loading highly visible cuts today or tolerating somewhat higher debt while pursuing less polarizing ways to stabilize the fiscal trajectory, I would rather err on the side of political stability than theoretical purity.

This is the analytical struggle of our moment. How do you respect demographic math without treating it as destiny? How do you embrace technological acceleration without mistaking it for guaranteed salvation?

I am more optimistic than many about the coming productivity regime. But optimism is not immunity from error. The least we can do is design institutions that reduce the odds of catastrophic miscalculation — and be honest about which risks we are prioritizing.

The future will likely be shaped in the 2030s. Between now and then, the most dangerous mistake may not be betting on growth. It may be destabilizing the present in fear of a future that no longer arrives.