The $9 Trillion Signal

What Jensen Huang's GitHub data tells every business leader about AI agent ROI.

Jensen Huang stood on stage in Taipei last week and said something that should reshape how every executive thinks about AI investment.

He wasn't talking about chips. He was talking about a multiplier.

30 million software developers. $3 trillion in salaries. That same workforce is now producing nearly three times as much output — effectively $9 trillion of productivity from $3 trillion of salaries. Jensen Huang · GTC Taipei, June 1 2026

His evidence: GitHub commit volume grew from 300 million in 2023 to 500 million in 2025 — then nearly tripled to 1.4 billion in just the first months of 2026. Same number of developers. Three times the output. No new headcount.

That is not a software story. That is a business model story. And it applies far beyond engineering.

The framework: from labour cost to productive multiplier

Huang's argument is transferable across every business function. The logic:

• A function has a defined labour cost (salary base).
• AI agents multiply the output of that labour without proportionally increasing headcount.
• The same salary base now generates a larger economic return.
• That surplus funds expansion — more hires, more throughput, more market coverage.

In software development, the multiplier is approximately 3x — documented at scale across 1.4 billion commits.

The question for every other function: what is your multiplier, and is the evidence there yet?

The honest answer in 2026: the evidence is building fast — but the maturity curve varies significantly by domain.

1.Software development — the benchmark

This is Huang's anchor data point — and the most robustly evidenced domain. GitHub commits grew from 300M (2023) → 400M (2024) → 500M (2025) → 1.4B (early 2026) with no proportional increase in developer count.

Huang's conclusion: the $3T global software salary base is now generating $9T in economic output. AI coding agents now plan tasks, write tests, debug, and open pull requests autonomously — not just autocomplete.

Sources

2.Customer support — the scale engine

Customer support has the most dramatic and well-documented AI agent ROI outside engineering. The mechanism is asymmetric economics: an AI agent's marginal cost per interaction approaches zero after deployment; a human agent costs $6–12 per conversation.

• 84% resolution — Salesforce deployed Agentforce on help.salesforce.com in October 2024. By Q4 FY2025 it had handled 380,000+ conversations, resolving 84% autonomously with only 2% requiring human escalation.
• ROI range — Average enterprise ROI: $3.50 for every $1 spent. Leading organisations: up to 8x.
• Market scale — Gartner projects conversational AI will save $80 billion in contact-centre labour costs globally by end of 2026.

Sources

3.Marketing & sales — the revenue engine

Salesforce's 2026 State of Sales report — surveying 4,050 professionals across 22 countries — found that AI and AI agents are the #1 growth tactic for sales teams this year.

• 2–3x improvement in pipeline velocity for lead qualification, personalised outreach, and meeting-booking agents (Gartner, 2026).
• SDR agents pay back in a median of 3.4 months — fastest payback of any agent category (BCG/Forrester 2026).
• Cost per qualified opportunity fell from $487 (human-only) to $224 in hybrid AI+human pods — a 54% reduction (Bridge Group SDR Metrics 2026).
• Top-performing sales organisations are 1.7x more likely to use AI agents than struggling teams (Salesforce State of Sales 2026).
• Revenue increase of up to 4% through higher win rates, faster proposals, and improved retention (Forrester TEI for Microsoft Copilot, March 2025).

Sources

4.Finance & accounting — the accuracy engine

Finance is the domain with the clearest ROI case after software and customer support, because every output is quantified by design.

• Financial modelling in Excel: 30–40% faster (Microsoft enterprise pilot data).
• Document drafting: 50–60% faster in Word (Microsoft enterprise data).
• Accounting procedure time cut by 90% in a Google Cloud / Finnit deployment.
• Forrester TEI for Microsoft 365 Copilot (25,000-employee composite): 116% ROI over 3 years, $19.7M NPV.
• Law firm DWF reduced a 7-day contract draft to 7 hours using AI agents.
• Knowledge workers save 30–60 minutes per day on routine tasks (EPC Group, 70+ Fortune 500 Copilot deployments).

Sources

5.Operations & supply chain — the efficiency engine

Operations is where ROI is most measurable because the inputs and outputs are quantitative by nature. The mechanism is waste and latency elimination, not revenue generation — but the economic impact compounds at volume.

• Finance and ops agents accelerating close processes by 30–50% (Gartner, 2026).
• Bradesco (82-year-old Latin American bank): agentic AI freed 17% of employee capacity and cut lead times by 22% (IDC case study).
• Early adopters reporting 20–30% faster workflow cycles (BCG/Forrester 2026).
• McKinsey: AI agents could add $2.6–$4.4 trillion in annual value across business domains.

Sources

6.HR & talent — the throughput engine

HR is the domain where ROI is real but hardest to directly quantify — it shows up in speed, quality, and retention metrics rather than direct revenue.

• New-hire onboarding time reduced by up to 30% (Forrester TEI — Microsoft 365 Copilot, March 2025).
• 15–25% improvement in intent-to-stay for employees who use AI tools vs. those who don't (McKinsey, Gartner, Microsoft research — consistent across three independent datasets).
• Fortune 50 company (200,000 employees): an Intelligent Employee Assistant across Teams and Salesforce automated IT/HR service-desk tasks (OneReach.ai / IDC case study).
• AI agents in production for resume screening, interview scheduling, and candidate evaluation across large enterprises (IDC, 2026).

Sources

The honest maturity map

The unified ROI framework

Huang's $3T → $9T equation gives us the template. Apply it to any function:

1. Define your labour base — what is the total salary cost of the team or function?
2. Identify the multiplier opportunity — what % of that team's time is on tasks an agent could handle autonomously?
3. Model the output expansion — if agents reclaim 40% of a sales team's administrative time, what does revenue per seller look like?
4. Measure tokens, not just tasks — every token that previously required human labour to produce is recovered margin.
5. Track the leading indicator before you deploy — in software it was GitHub commits; in sales it is pipeline velocity; in finance it is close cycle time; in support it is resolution rate.

The organisations capturing Huang's multiplier are not the ones with the most ambitious AI strategies. They are the ones who picked the highest-value, most measurable workflow — deployed there first — and built from a proven base.

The number that matters as much as the ROI

Forrester and Anaconda 2026 data cite evaluation gaps (64% of leaders), governance friction (57%), and model reliability (51%) as the top blockers. McKinsey puts the average ROI on AI investment at 5.8x — but only 25% of AI initiatives deliver expected ROI, and only 16% reach enterprise-wide scale.

The gap between Huang's $9T signal and most organisations' AI reality is not a capability gap. It is a deployment, measurement, and governance gap.

The conclusion

The GitHub commit surge is not a software metric. It is a signal about what happens when you give a skilled workforce a 3x force multiplier.

The question is not whether AI agents deliver ROI. The evidence in 2026 says they do — in software emphatically, in customer support convincingly, and in finance, sales, and operations compellingly.

The question is whether you have deployed them where your labour is most expensive, your output is most measurable, and your competitive disadvantage from not acting compounds fastest.

If you can hire a software engineer and generate $9 trillion worth of productive work, why wouldn't you want to hire more software engineers? Jensen Huang · GTC Taipei, June 1 2026

The same logic applies to every hire in your organisation.

What multiplier are you seeing in your deployments? And which domain has surprised you most — positively or negatively? I'd genuinely like to know.

Full source list

Primary sources

Analyst & research sources

Case studies & industry reports