Whether it is the advent of electricity, the internal combustion engine, or the Internet, every technological revolution in history has delivered much the same experience for investors: short‑term pain followed by long‑term gain.
The costs and benefits invariably follow a J‑curve.
First come the costs. Heavy capital investment in the new technology marks the initial developmental phase. This is typically followed by a temporary decline in productivity and corporate profitability.
Then comes a turbulent period of societal adjustment, when some jobs are destroyed by the emerging technology and new ones are created.
Only after this do the benefits of the new technology’s diffusion begin to materialise — higher output, revenue and employment, alongside other gains for society, such as more leisure time.
Take electrification in the late 19th century.
Companies first spent heavily to rewire their factories, disrupting existing processes but with little to show for it in terms of productivity or profit.
Only after they completely reorganised their operations around the new technology — changing factory layouts and introducing more continuous production — did the real pay‑off arrive, in the form of higher output per worker and lower costs. It was at this point, long after the initial spending wave, that rewards finally began to flow to investors.
There is a useful lesson here for those trying to understand how the latest general‑purpose technology — artificial intelligence — might evolve.
It is already obvious that it too is progressing along a J-curve. As the Pictet Research Institute’s latest research shows, AI appears to be at its very early stages of development.https://am.pictet.com/ch/en/intermediaries/investment-research/demographics-and-technology
At this juncture, many industries are experiencing a productivity dip: significant investments in data infrastructure, organisational restructuring and workforce retraining are temporarily outweighing measurable returns.
Put another way, cognitive work processes are being redesigned more quickly than employees can adapt, causing productivity declines despite considerable technological advancements.
Source: Pictet Research Institute. Note: Based on empirical synthesis from Acemoglu (2025), Acemoglu & Restrepo (2020), Atkeson & Kehoe (2007), Autor (2024), Brynjolfsson et al. (2021), Brynjolfsson & Hitt (2003), Devine (1983), IFR (2024b), Jorgenson & Stiroh (2000), McKinsey (2023)
This is as true for companies developing AI tech as it is for firms deploying it.
Tech firms such as Meta, Nvidia and Microsoft have poured more than USD600 billion this year alone into building out AI data centres,https://www.reuters.com/business/global-software-data-firms-slide-ai-disruption-fears-compound-jitters-over-600-2026-02-06/overhauling production processes and slashing headcount - yet so far, without any material improvement in their bottom line.
That's not to say there is no evidence of tangible benefits.
In the health industry, for example, AI-human teams have cut radiologists’ workload while boosting disease detection rates and reducing aggressive cancer in subsequent years. https://ecancer.org/en/news/27721-ai-supported-mammography-screening-results-in-fewer-aggressive-and-advanced-breast-cancers-finds-full-results-from-first-randomised-controlled-trialElsewhere, vision-guided technology in distribution centres are achieving double-digit gains in picking speed. In call centres, meanwhile, chatbots are handling routine customer inquiries, freeing human staff for complex tasks and cutting overall costs.
Benefits are also emerging in agriculture. AI‑powered crop‑monitoring tools are helping farmers raise yields while using fewer water and fertilisers.
Taking all this into account, it would seem that the evolution of AI is faithfully tracing the path followed by previous general purpose technologies such as the Internet or electricity.
But there is one potential difference, and it is a significant one.
There is every chance that its cost phase could prove to be deeper and longer-lasting than anything the industrialised world has previously experienced.
With AI increasingly viewed as a threat to both white and blue-collar jobs, its adoption likely to face considerable societal and political resistance.
All of which suggests investors should brace for an extended period of AI-provoked upheaval.
Managing AI’s economic and social cost: the fiscal burden
AI is already causing significant social externalities, or negative side-effects, that financial markets – and government bond markets in particular - do not adequately discount.
Research shows AI’s adoption is beginning to displace lower-skilled or entry-level jobs in services industries as varied as law, banking, software development, advertising and accountancy.
According to the IMF, entry level employment in AI-vulnerable industries is already 3.6% lower than in sectors where demand for AI is weak.https://www.imf.org/en/publications/wp/issues/2024/09/13/the-labor-market-impact-of-artificial-intelligence-evidence-from-us-regions-554845
Even companies at the epicentre of the AI innovation wave are replacing workers with machine learning.
Microsoft, for example, has cut 7% of its workforce since January 2025, while Amazon has slashed 30,000 jobs in just three months to January 2026. The Big Four consultancy firms have also announced thousands
of redundancies mainly in junior roles.
If unemployment rises and inequality grows, governments will come under pressure to act.
Policy options include more generous and better-targeted subsidies for retraining and upskilling, regulatory measures to stop or slow job cuts, such as mandatory employment, or even a one-off windfall tax on AI hyper-scalers.
While this will be costly for the public purse, doing nothing would lead to a worse outcome. Failing to act could leave an entire generation disincentivised to work, one lacking in the skills that can boost an economy’s productivity.
In a typical productivity boom, where the adoption of a new technology smooth and orderly, an economy experiences higher real interest rates and lower breakeven rates – meaning inflation expectations are anchored.
But in disorderly transition in which governments offer no or insufficient support for displaced workers, the anticipation of mass unemployment incentivises households to save rather than spend, weighing on economic growth and leading to a reduction in real interest rates.
It is our view that governments will act to counter any severe AI-related disruption in the labour market. And this will have several investment implications for government bond markets.
Increased public spending will add to already-elevated levels of public debt, forcing central banks to keep interest rates artificially low to prioritise fiscal stability.
As risk-free rates fall, governments have greater incentives to borrow more from domestic investors than foreign creditors, since a deep, home-currency investor base offers stable funding and reduces vulnerability to external swings.
There are signs that this is already happening in the US. There, the share of short-term bills used to finance government borrowing – which are primarily bought by domestic investors - has risen to around 21% of marketable debt, slightly above the 15-20% range recommended by the Treasury Borrowing Advisory Committee (TBAC).TBAC is made up of private market participants with whom the Treasury consults as part of the quarterly refunding process.
This has had the effect of shortening the duration of the US government’s liabilities, which leaves the Treasury vulnerable to any unforeseen spike in interest rates even if affords some flexibility to adjust the cost of funding over time.https://www.jec.senate.gov/public/vendor/_accounts/JEC-R/debt/Monthly%20Debt%20Update.html
This should, all else being equal, steepen bond yield curves. In order to manage the higher interest-rate sensitivity of public debt, policymakers may increasingly rely on “financial repression” – or keeping real rates artificially low or capping yields. In which case, bond yields become less of a free-market signal, where true inflation or default risk may not be accurately priced in, distorting the asset class’s informational role.
Share of short-term sovereign bond issuance vs total, 12-month moving sum, %
Source: Institute of International Finance
In the UK, with taxes already high, the government is more likely to lean on extra borrowing than further tax hikes, which could lead to higher gilt yields or a weaker pound.
Europe may choose to slow AI adoption with regulation (even if its strong industrial base can still benefit from the global AI capex boom).
However, any extra AI-related borrowing would come on top of existing social spending commitments. This could push the euro zone’s debt‑to‑GDP ratio even higher, further constraining fiscal space and encouraging investors to demand a higher premium.
Japan will also be adding to its already high level of public borrowing to counter AI disruption.
The risks for developed economies, therefore, are two-fold: many governments will be under pressure to take on more debt, and much of it will be issued at shorter maturities, raising rollover and duration risk.
In effect, the higher welfare bill associated with AI disruption points to persistently larger net supply of government bonds and, unless fully offset by central banks or private demand, a structurally higher term premium and thus higher yields over time or more repression.
Whatever form that repression takes, it ultimately amounts to debt
monetisation, which is typically negative for currencies and relatively more supportive for real assets such as gold.
Emerging economies, in our view, are better positioned to endure the pain phase from a public-debt standpoint: their social welfare commitments are lower than their developed counterparts, while their government debt represents around 70-76% of GDP, compared with around 120% for advanced counterparts.https://www.imf.org/external/datamapper/GGXWDG_NGDP@WEO/OEMDC/ADVEC/WEOWORLD
The dispersion in bond yields across developed and developing markets might also have consequences for equity investors.
This is because the equity risk premium – or the extra compensation investors seek for holding equities over risk-free assets – should become more attractive in economies that are willing to, and can afford to, mitigate the societal costs of AI adoption without taking on more debt, or monetising debt.
By contrast, countries lacking the financial resources to engineer a smoother transition look exposed to the negative effects of AI. Their financial assets would, all being equal, command a higher risk premium. Under a worst-case scenario, these countries may opt to limit access to AI, which could further curtail their long-term growth.
Investments in the AI supply chain: winners and losers
But changes in equity risk premia isn’t the only development stock investors will need to consider as the AI J-curve progresses.
AI’s fraught adoption will create myriad commercial risks and opportunities. This is clearly the case for firms within the AI supply chain. But it is also true for companies attempting to wring efficiencies from the technology and those whose business models might be rendered obsolete by it.
Underscoring how difficult it is for investors to assess the benefits of AI, a recent study found that 95% of organisations that have piloted generative AI tools like ChatGPT and Copilot have reported zero return for their investment.https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
Equity investors should therefore take a more selective approach to AI exposure. In our view, they should focus on identifying and investing in the bottlenecks in the infrastructure behind the boom – semiconductors, memory chips and hardware that enable capacity build‑out and computer power, rather than on the broader software and application providers, where earnings outlook may lag reality.
When it comes to sectors, heavy industries, such as energy, material and heavy equipment manufacturers, are well positioned to capitalise on AI’s efficiency improvements and boost profitability. They represent an industry where productivity gains have been harder to achieve and material efficiency improvements are still to come.
From trough to the era of abundance
Investors should view AI as a long build-out, rather than an imminent economic and investment game changer.
While global AI spending is rising fast – projected to reach USD 2.5 trillion in 2026, up 44% year on year – it remains modest relative to the size of the economy.https://www.gartner.com/en/newsroom/press-releases/2026-1-15-gartner-says-worldwide-ai-spending-will-total-2-point-5-trillion-dollars-in-2026
US tech firms’ AI spending is estimated to be about 1% of GDP, smaller than recent investment booms including the US shale craze of the mid-2010s and the dot-com bubble of the 1990s.https://www.bis.org/publ/bisbull120.pdfCommercial property and mining surges in Japan and Australia during the 1980s and 2010s, respectively, were over five times larger relative to GDP.
Change in investment, % of GDP, in major industrial overhauls
Source: BIS, https://www.bis.org/publ/bisbull120.pdf
More abundant AI inputs will ultimately support the range and volumes of tasks we perform, supporting new forms of work, novel industries and business models as well as higher living standards. It will bring a future well worth the J-curve.
Against this backdrop, investors need to factor in the social and fiscal adjustment costs ahead, recognising that AI’s long‑term rewards are likely to come with a deeper, more uneven and more fiscally burdensome transition than many currently assume.
However, this is not to say that the pain phase of the transition – with all the fiscal strain and social tensions it could bring – is not worth it. This is because of the Jevons Paradox, a phenomenon observed during almost every major technological shift in history: when technology makes a resource or task much more efficient, its effective price falls, so people and businesses find many more uses for it and total consumption can actually rise.
If AI follows this historical pattern – which we think it will – more abundant AI inputs will ultimately expand the range and volume of tasks we perform, supporting new forms of work, novel industries and
business models as well as higher living standards, rather than permanently displacing humans.
It will create a future well worth its longer-lasting J-curve.
