The AI intelligence explosion is likely to become an economy-wide cost story, not just a tech story. BlackRock recently estimated that a modest shift in labor’s share of costs could generate $110 trillion in combined present value and lift corporate earnings by 31% — with beneficiaries extending far beyond technology.
Recursive self-improvement compresses the timeline. Frontier labs are automating their own R&D, meaning the AI cost revolution may arrive sooner and cut deeper than most investors assume — and legacy software companies may have half the adaptation window they expect.
The market may be underestimating three things at once: the broad earnings uplift from AI-driven margin expansion, the speed of disruption for many industries, including legacy software, and the emerging secular growth category of AI governance and compliance.
CALIBRATING THE INTELLIGENCE EXPLOSION
Start with two numbers. The first: 68.3 billion. That is Epoch AI’s projection for the global effective workforce — humans plus AI equivalents — by 2035, roughly 15 times today’s human labor force of 4.7 billion (Figure 1). The second: nine months. That is roughly how long OpenAI says it will take to deploy “hundreds of thousands of automated AI research interns,” with a fully automated research workforce targeted within two years.1
Investors have recently been dissecting the “SaaSpocalypse” — the near-$1 trillion rout in software stocks triggered by Anthropic’s impressive product launches and the market’s belated reckoning with agentic AI (Figure 2). That conversation matters, but it is focused on the tools. The far more consequential story is the factory that builds the tools: frontier AI labs are automating their own research at a pace that could reshape how investors think about the entire technology stack — and the timeline on which every sector faces disruption.
There is, however, a third number that reframes the entire debate — and it comes not from a technologist but from one of the most influential investment minds on Wall Street.
THE $110 TRILLION COST REVOLUTION
In his 2026 outlook, BlackRock’s Rick Rieder — the firm’s Chief Investment Officer of Global Fixed Income and a top-contender for the recently filled US Federal Reserve Chair role — frames AI not as a revenue story but as a cost revolution.2 The distinction is critical. While Wall Street has been arguing about which software companies lose revenue to AI agents, Rieder is focused on the far larger prize: What happens to corporate profitability across every sector when AI compresses labor’s share of total costs?
The arithmetic is startling. Labor currently accounts for roughly 55% of total US business-sector costs. Rieder’s illustrative scenario asks what happens if AI and related automation reduce that share by just five percentage points — from 55% to 50% — and if three-quarters of those savings accrue to the companies deploying AI while the remaining quarter flows to AI service providers.
The result: approximately $1.2 trillion in annual labor cost savings, translating into roughly $878 billion in incremental after-tax corporate profits per year. Discounted at reasonable rates, the present value of the corporate piece alone is on the order of $82 trillion, with another $27 trillion accruing to AI providers. The combined figure: $110 trillion (Figure 3).
Put differently: a 9% reduction in labor costs translates mechanically into roughly 31% higher corporate earnings.3 And that estimate assumes a modest five-percentage-point shift in cost structure — not the wholesale transformation that Epoch AI’s projections imply. If the intelligence explosion delivers even a fraction of what the frontier labs are targeting, the earnings uplift could be substantially larger.
This reframes the SaaSpocalypse entirely. Yes, some software companies lose. But the corporate sector as a whole wins — and wins big. The value that migrates away from labor and from certain software vendors does not disappear. It shows up as margin expansion across every industry that successfully deploys AI agents, from financial services and healthcare to logistics and manufacturing. The stock market, in aggregate, is a beneficiary of the intelligence explosion, even if specific sectors within it are disrupted.
INSIDE THE RECURSIVE LOOP
What makes Rieder’s cost-revolution framework so powerful is that it may be conservative, because it may not fully account for the self-reinforcing dynamics inside the labs that are building these AI systems.
Dean Ball, a researcher at the Foundation for American Innovation, has laid out the mechanics with unusual clarity. The major frontier labs — OpenAI, Anthropic, and Google DeepMind — have already begun automating large fractions of their research and engineering operations. This is not a future aspiration. It is happening now, behind closed doors, and Ball argues it may be the single most important development in AI this year.
The logic is recursive. Today’s most capable AI models — Claude Opus 4.6, GPT-5 and its successors, and Gemini — are already strong enough to contribute meaningfully to AI research. They can write and debug code, run experiments, tune hyperparameters, generate and test hypotheses, and in some cases propose novel model architectures. The labs are scaling this involvement rapidly. Within a year or two, Ball projects, the effective “workforces” of each frontier lab could grow from the low thousands of human researchers to tens of thousands, then hundreds of thousands — digital researchers that, as Ball puts it, “will neither sleep nor eat nor take breaks. And their only objective is to make themselves, the AI systems, smarter” (Figure 4).
If each generation of AI helps build a better successor, the result is a positive feedback loop — not a mythical overnight singularity, but a concerted, compounding acceleration in the pace of improvement. The historical rate of AI capability growth has roughly doubled every four months (Figure 5). Recursive automation could steepen that curve considerably. The physical constraints are real — compute availability, training time, energy infrastructure — but within those limits, the trajectory points toward an exponential increase in effective AI research capacity.
WHAT EXPONENTIAL PROGRESS MEANS FOR MARKETS
The convergence of Rieder’s cost-revolution math and Ball’s recursive-improvement thesis produces an investment framework that many investors may not fully appreciate. If AI capability is accelerating faster than consensus assumes, the cost revolution arrives sooner and cuts deeper than Rieder’s already-impressive base case. And the beneficiaries are not confined to the technology sector.
The bull case for AI infrastructure gets steeper and longer. Companies like Nvidia, TSMC, and the hyperscale cloud providers are already benefiting from a massive compute buildout. If recursive self-improvement works, compute demand doesn’t plateau after a few model generations — it compounds. Energy infrastructure plays, including natural gas, nuclear restarts, and grid equipment, are similarly exposed to upside. The US power grid was not built for an economy running hundreds of thousands of simultaneous AI research processes.
The bull case for broad equities is Rieder’s core insight. A 31% increase in corporate earnings from labor-cost compression alone would represent one of the largest margin expansions in modern market history. And it accrues to any company that deploys AI effectively, not just the companies that build it. Rieder is explicit: “The key is to own the beneficiaries of that transition — businesses with scalable models, durable cash flows, and credible plans for harnessing AI to lift productivity.”
Financial services firms, healthcare companies, industrial conglomerates with complex operations, and professional services firms all stand to capture a share of the $82 trillion in present value that flows from labor-cost reduction.
The bear case for legacy software could also arrive faster than anyone expects. This is the less obvious implication of recursive improvement. A company that looks safe today because current AI agents “aren’t good enough yet” to replace its specialized workflow might find that gap closing in 12 to 18 months, not the five years its management is banking on.
The SaaSpocalypse of February 2026 was triggered by tools that barely existed six months earlier. If recursive improvement is real, the next wave of disruption may hit sooner and harder. Boris Cherny, the head of Anthropic’s Claude Code, shipped 22 and 27 code commits on consecutive days, all produced by Claude. An OpenAI engineer posted simply: “100% — I don’t write code anymore.” The progression from 10% AI-written code in 2023 to 100% at the frontier in early 2026 is the pace investors should be calibrating against (Figure 6).
THE GOVERNANCE PREMIUM
There is a category of investment implication that the market has barely begun to price: the value of AI governance, compliance, and oversight infrastructure. As AI builds AI and corporate deployment accelerates to capture Rieder’s labor savings, the demand for guardrails scales at least as fast as the capability it is guarding.
Europe’s AI Act is already imposing requirements on high-risk systems. The US is moving toward disclosure and audit standards for AI in finance, healthcare, and critical infrastructure. Every enterprise deploying AI agents in regulated industries will need tools to monitor, log, and verify what those agents do. Audit trails, explainability layers, and compliance dashboards are not glamorous, but they are becoming as essential as cybersecurity was a decade ago — and cybersecurity, once it became mandatory enterprise spending, produced a generation of high-growth, high-margin companies. It is still early days, but AI governance is likely on the same trajectory.
PRICING THE FEEDBACK LOOP
The market’s current positioning tells a familiar story: overweight the hardware, underweight the disruption risk to legacy software, and largely neutral on both the broad corporate margin story and the governance opportunity. That allocation made sense when AI progress felt roughly linear and adoption curves were slow and predictable.
It makes less sense if the labs are right about recursive self-improvement and if Rieder is right that the primary economic impact of AI is cost compression rather than new revenue creation. The Epoch AI projections — a 30% compound annual growth rate in effective cognitive labor through 2035 — are built on exactly this acceleration. Even discounted heavily, they imply a pace of change that stress tests virtually every sector’s forward estimates.
The practical investment framework comes down to four key points.
AI infrastructure demand is real but Nvidia and the hyperscalers benefit from recursive AI development, but capex cycles in compute buildouts have historically been volatile.
The disruption timeline for legacy software is shorter than consensus: companies relying on a five-year adaptation window may have half that, or less
Broad corporate earnings stand to benefit massively from Rieder’s cost revolution, and the market has not yet priced a 31% earnings uplift into the non-tech sectors that stand to capture the largest share of $82 trillion in present value.
Governance and compliance are the next secular growth As AI capability compounds, so does the institutional demand for oversight infrastructure.
The SaaSpocalypse discussion asked the right question: what happens when AI agents can do the work? The recursive loop asks the harder one: what happens when they get better at it every single month? And Rieder’s framework supplies the answer that matters most for portfolios: the value does not vanish when labor’s share declines. It migrates — to corporate margins, to AI providers, and to the investors positioned on the right side of the largest cost revolution in modern economic history.
William P. Sterling, Ph.D.,
Global Strategist
1 These figures were quoted in Dean W. Ball, “On Recursive Self-Improvement,” Hyperdimensional Substack, February 5, 2026.
2 Rick Rieder, “The Odds Are Changing: Investing in 2026,” Blackrock Advisor Center, January 2, 2026.
3 In highly competitive sectors, AI-driven efficiencies may primarily manifest as disinflation and quality gains rather than sustained margin expansion.
This represents the views and opinions of GW&K Investment Management and does not constitute investment advice, nor should it be considered predictive of any future market performance. Data is from what we believe to be reliable sources, but it cannot be guaranteed. Opinions expressed are subject to change. Past performance is not indicative of future results.
Indexes are not subject to fees and expenses typically associated with managed accounts or investment funds. Investments cannot be made directly in an index. Index data has been obtained from third-party data providers that GW&K believes to be reliable, but GW&K does not guarantee its accuracy, completeness or timeliness. Third-party data providers make no warranties or representations relating to the accuracy, completeness or timeliness of the data they provide and are not liable for any damages relating to this data. The third-party data may not be further redistributed or used without the relevant third-party’s consent. Sources for index data include: Bloomberg, FactSet, ICE, FTSE Russell, MSCI and Standard & Poor’s.
Keep
Reading
GW&K Investment Review 1Q 2026
Macro | Insight
In his Q1 2026 Economic Letter, Harold G. Kotler shares a timely reminder: patience, discipline, and diversification matter.
Economic And Market Implications Of The US-Iran Conflict
Macro | Insight
Global Strategist Bill Sterling discusses potential market implications — especially for global energy — and key considerations for navigating volatility from the conflict involving Iran.
We use cookies to improve your experience on our website. To accept cookies click Accept & Close, or continue browsing as normal. For more information or to learn how to opt out of cookies, please see our cookie policy.
The Coming Intelligence Explosion
When AI Builds AI, Who Wins On Wall Street?
Global Perspectives | March 2026
CALIBRATING THE INTELLIGENCE EXPLOSION
Start with two numbers. The first: 68.3 billion. That is Epoch AI’s projection for the global effective workforce — humans plus AI equivalents — by 2035, roughly 15 times today’s human labor force of 4.7 billion (Figure 1). The second: nine months. That is roughly how long OpenAI says it will take to deploy “hundreds of thousands of automated AI research interns,” with a fully automated research workforce targeted within two years.1
Investors have recently been dissecting the “SaaSpocalypse” — the near-$1 trillion rout in software stocks triggered by Anthropic’s impressive product launches and the market’s belated reckoning with agentic AI (Figure 2). That conversation matters, but it is focused on the tools. The far more consequential story is the factory that builds the tools: frontier AI labs are automating their own research at a pace that could reshape how investors think about the entire technology stack — and the timeline on which every sector faces disruption.
There is, however, a third number that reframes the entire debate — and it comes not from a technologist but from one of the most influential investment minds on Wall Street.
THE $110 TRILLION COST REVOLUTION
In his 2026 outlook, BlackRock’s Rick Rieder — the firm’s Chief Investment Officer of Global Fixed Income and a top-contender for the recently filled US Federal Reserve Chair role — frames AI not as a revenue story but as a cost revolution.2 The distinction is critical. While Wall Street has been arguing about which software companies lose revenue to AI agents, Rieder is focused on the far larger prize: What happens to corporate profitability across every sector when AI compresses labor’s share of total costs?
The arithmetic is startling. Labor currently accounts for roughly 55% of total US business-sector costs. Rieder’s illustrative scenario asks what happens if AI and related automation reduce that share by just five percentage points — from 55% to 50% — and if three-quarters of those savings accrue to the companies deploying AI while the remaining quarter flows to AI service providers.
The result: approximately $1.2 trillion in annual labor cost savings, translating into roughly $878 billion in incremental after-tax corporate profits per year. Discounted at reasonable rates, the present value of the corporate piece alone is on the order of $82 trillion, with another $27 trillion accruing to AI providers. The combined figure: $110 trillion (Figure 3).
Put differently: a 9% reduction in labor costs translates mechanically into roughly 31% higher corporate earnings.3 And that estimate assumes a modest five-percentage-point shift in cost structure — not the wholesale transformation that Epoch AI’s projections imply. If the intelligence explosion delivers even a fraction of what the frontier labs are targeting, the earnings uplift could be substantially larger.
This reframes the SaaSpocalypse entirely. Yes, some software companies lose. But the corporate sector as a whole wins — and wins big. The value that migrates away from labor and from certain software vendors does not disappear. It shows up as margin expansion across every industry that successfully deploys AI agents, from financial services and healthcare to logistics and manufacturing. The stock market, in aggregate, is a beneficiary of the intelligence explosion, even if specific sectors within it are disrupted.
INSIDE THE RECURSIVE LOOP
What makes Rieder’s cost-revolution framework so powerful is that it may be conservative, because it may not fully account for the self-reinforcing dynamics inside the labs that are building these AI systems.
Dean Ball, a researcher at the Foundation for American Innovation, has laid out the mechanics with unusual clarity. The major frontier labs — OpenAI, Anthropic, and Google DeepMind — have already begun automating large fractions of their research and engineering operations. This is not a future aspiration. It is happening now, behind closed doors, and Ball argues it may be the single most important development in AI this year.
The logic is recursive. Today’s most capable AI models — Claude Opus 4.6, GPT-5 and its successors, and Gemini — are already strong enough to contribute meaningfully to AI research. They can write and debug code, run experiments, tune hyperparameters, generate and test hypotheses, and in some cases propose novel model architectures. The labs are scaling this involvement rapidly. Within a year or two, Ball projects, the effective “workforces” of each frontier lab could grow from the low thousands of human researchers to tens of thousands, then hundreds of thousands — digital researchers that, as Ball puts it, “will neither sleep nor eat nor take breaks. And their only objective is to make themselves, the AI systems, smarter” (Figure 4).
If each generation of AI helps build a better successor, the result is a positive feedback loop — not a mythical overnight singularity, but a concerted, compounding acceleration in the pace of improvement. The historical rate of AI capability growth has roughly doubled every four months (Figure 5). Recursive automation could steepen that curve considerably. The physical constraints are real — compute availability, training time, energy infrastructure — but within those limits, the trajectory points toward an exponential increase in effective AI research capacity.
WHAT EXPONENTIAL PROGRESS MEANS FOR MARKETS
The convergence of Rieder’s cost-revolution math and Ball’s recursive-improvement thesis produces an investment framework that many investors may not fully appreciate. If AI capability is accelerating faster than consensus assumes, the cost revolution arrives sooner and cuts deeper than Rieder’s already-impressive base case. And the beneficiaries are not confined to the technology sector.
The bull case for AI infrastructure gets steeper and longer. Companies like Nvidia, TSMC, and the hyperscale cloud providers are already benefiting from a massive compute buildout. If recursive self-improvement works, compute demand doesn’t plateau after a few model generations — it compounds. Energy infrastructure plays, including natural gas, nuclear restarts, and grid equipment, are similarly exposed to upside. The US power grid was not built for an economy running hundreds of thousands of simultaneous AI research processes.
The bull case for broad equities is Rieder’s core insight. A 31% increase in corporate earnings from labor-cost compression alone would represent one of the largest margin expansions in modern market history. And it accrues to any company that deploys AI effectively, not just the companies that build it. Rieder is explicit: “The key is to own the beneficiaries of that transition — businesses with scalable models, durable cash flows, and credible plans for harnessing AI to lift productivity.”
Financial services firms, healthcare companies, industrial conglomerates with complex operations, and professional services firms all stand to capture a share of the $82 trillion in present value that flows from labor-cost reduction.
The bear case for legacy software could also arrive faster than anyone expects. This is the less obvious implication of recursive improvement. A company that looks safe today because current AI agents “aren’t good enough yet” to replace its specialized workflow might find that gap closing in 12 to 18 months, not the five years its management is banking on.
The SaaSpocalypse of February 2026 was triggered by tools that barely existed six months earlier. If recursive improvement is real, the next wave of disruption may hit sooner and harder. Boris Cherny, the head of Anthropic’s Claude Code, shipped 22 and 27 code commits on consecutive days, all produced by Claude. An OpenAI engineer posted simply: “100% — I don’t write code anymore.” The progression from 10% AI-written code in 2023 to 100% at the frontier in early 2026 is the pace investors should be calibrating against (Figure 6).
THE GOVERNANCE PREMIUM
There is a category of investment implication that the market has barely begun to price: the value of AI governance, compliance, and oversight infrastructure. As AI builds AI and corporate deployment accelerates to capture Rieder’s labor savings, the demand for guardrails scales at least as fast as the capability it is guarding.
Europe’s AI Act is already imposing requirements on high-risk systems. The US is moving toward disclosure and audit standards for AI in finance, healthcare, and critical infrastructure. Every enterprise deploying AI agents in regulated industries will need tools to monitor, log, and verify what those agents do. Audit trails, explainability layers, and compliance dashboards are not glamorous, but they are becoming as essential as cybersecurity was a decade ago — and cybersecurity, once it became mandatory enterprise spending, produced a generation of high-growth, high-margin companies. It is still early days, but AI governance is likely on the same trajectory.
PRICING THE FEEDBACK LOOP
The market’s current positioning tells a familiar story: overweight the hardware, underweight the disruption risk to legacy software, and largely neutral on both the broad corporate margin story and the governance opportunity. That allocation made sense when AI progress felt roughly linear and adoption curves were slow and predictable.
It makes less sense if the labs are right about recursive self-improvement and if Rieder is right that the primary economic impact of AI is cost compression rather than new revenue creation. The Epoch AI projections — a 30% compound annual growth rate in effective cognitive labor through 2035 — are built on exactly this acceleration. Even discounted heavily, they imply a pace of change that stress tests virtually every sector’s forward estimates.
The practical investment framework comes down to four key points.
The SaaSpocalypse discussion asked the right question: what happens when AI agents can do the work? The recursive loop asks the harder one: what happens when they get better at it every single month? And Rieder’s framework supplies the answer that matters most for portfolios: the value does not vanish when labor’s share declines. It migrates — to corporate margins, to AI providers, and to the investors positioned on the right side of the largest cost revolution in modern economic history.
William P. Sterling, Ph.D.,
Global Strategist
1 These figures were quoted in Dean W. Ball, “On Recursive Self-Improvement,” Hyperdimensional Substack, February 5, 2026.
2 Rick Rieder, “The Odds Are Changing: Investing in 2026,” Blackrock Advisor Center, January 2, 2026.
3 In highly competitive sectors, AI-driven efficiencies may primarily manifest as disinflation and quality gains rather than sustained margin expansion.
William Sterling, Ph.D.
Global StrategistDisclosures
This represents the views and opinions of GW&K Investment Management and does not constitute investment advice, nor should it be considered predictive of any future market performance. Data is from what we believe to be reliable sources, but it cannot be guaranteed. Opinions expressed are subject to change. Past performance is not indicative of future results.
Indexes are not subject to fees and expenses typically associated with managed accounts or investment funds. Investments cannot be made directly in an index. Index data has been obtained from third-party data providers that GW&K believes to be reliable, but GW&K does not guarantee its accuracy, completeness or timeliness. Third-party data providers make no warranties or representations relating to the accuracy, completeness or timeliness of the data they provide and are not liable for any damages relating to this data. The third-party data may not be further redistributed or used without the relevant third-party’s consent. Sources for index data include: Bloomberg, FactSet, ICE, FTSE Russell, MSCI and Standard & Poor’s.