00 · Executive summary
The thesis in one page
America's AI dividend will not be decided by the largest banks or the newest startups. It will be decided by roughly 200,000 companies in the middle of the economy that employ about 48 million people and produce about a third of private sector GDP.
Those companies are the most important place for AI to land and the least equipped to land it. Small businesses can buy a subscription and be done. The Fortune 100 can spend tens of billions a year and staff armies of engineers. The middle market has neither option, so it sits in pilot mode while the gap widens.
~1/3
The middle market's share of U.S. private sector GDP and employment.
NCMM
36%
Share of AI-using mid-market firms with AI fully embedded across core processes.
RSM, 2026
16%
Share of mid-market companies with fully governed, integrated data.
Kaufman Rossin, 2026
My argument is simple. If you want AI to show up in GDP, in wages and in jobs, you have to transform the middle. And the firms that will do it are not the transformation giants of the last thirty years. They are a new class of AI-native services firms that can deliver enterprise-grade strategy and engineering at mid-market speed and price, because AI has collapsed the cost of the work itself.
Most of the middle market has bought AI tools. Very little of it has changed how the work gets done. That gap is where the national dividend is being lost.
This paper covers why the middle matters, why it is stuck, and what leaders should do next.
01 · The barbell
Two ends served, the middle stranded
The AI market is shaped like a barbell. Both ends are well served. The middle, where most of the economy's weight sits, is not.
segment 01
Small business
Served by software
- ~31.7M firms
- Standard problems
- AI ships inside tools
- Little legacy to connect
Buy it and go
segment 02
Middle market
Carries the weight, gets neither
- ~200,000 firms, $10M to $1B
- ~48M jobs, ~1/3 of private GDP
- Problems tied to core systems
- Aging ERP, on-prem, thin IT
- Outbid for AI talent
Stranded in pilot mode
segment 03
Enterprise
Served by scale
- One bank: $19.8B on tech
- 65,000+ technologists
- Cloud-native data
- Moving to custom models
Build it in-house
The small business end. A five-person firm's problems are mostly standard: bookkeeping, scheduling, marketing, customer email. A deep bench of SaaS products now ships with AI built in. The owner buys a tool, turns it on, and the problem is largely solved. There is little to integrate because there is little legacy to integrate with.
The enterprise end. The largest companies have already reinvented their systems at least once. They run cloud-native data estates, employ tens of thousands of technologists, and are moving to open and custom models. One large U.S. bank alone budgets about $19.8 billion a year for technology and employs more than 65,000 technologists, per Fortune. At that scale, AI is a line item, not a bet-the-company decision.
The middle. A 30-year-old manufacturer, distributor or services firm with 700 or 2,000 employees gets neither. Its problems are too specific for off-the-shelf tools and too entangled with core systems for a plug-in. Its IT team is a couple dozen people keeping the lights on, not ten thousand building the future. It knows it has to change. It does not have a credible path to do it.
That is the stranded middle. And it is where the country's AI outcome will be decided.
02 · Why the middle market matters
The load-bearing wall of the American economy
The National Center for the Middle Market defines the middle market as companies with $10 million to $1 billion in annual revenue. There are roughly 200,000 of them, against nearly 31.7 million small businesses, and they account for about one third of private sector GDP and employment.
| Segment | Annual revenue | Typical profile |
|---|---|---|
| Lower middle market | $10M to $50M | Founder or family led, lean back office, first professional managers |
| Core middle market | $50M to $500M | Multi-site operations, a real ERP, a small IT team, often PE-backed |
| Upper middle market | $500M to $1B | Multi-entity, acquisitive, enterprise-grade complexity on a mid-market budget |
Four facts make this segment the right target for national AI investment:
- It carries the jobs. About 48 million Americans work in middle-market companies, per NCMM's Year-End 2025 Middle Market Indicator.
- It grows through storms. NCMM reports these firms added 2.2 million jobs through the 2007 to 2010 financial crisis, a period when the U.S. economy as a whole lost millions of jobs.
- It is growing now. More than three quarters of middle-market firms reported revenue growth in 2025, with year-over-year growth reaching 11.7%, per NCMM data summarized by CNL. NCMM notes the core and upper tiers, $50 million and up, are driving both revenue and hiring gains.
- It makes the physical economy run. These are the regional manufacturers, distributors, logistics operators, healthcare groups, engineering firms and specialty insurers that sit inside every supply chain. They are rarely household names. When they stall, everything downstream feels it.
This is the part of the economy that is most mission critical and least covered.
It does not fit the small business story or the big company story, so policymakers, investors and vendors routinely skip past it. That is a mistake we can no longer afford with AI.
03 · The adoption gap
Integrated at the edges, stuck in the middle
Mid-market companies have tried AI. Few have scaled it. What stops them is the systems and data they run on.
The Census Bureau's Business Trends and Outlook Survey put national business AI use at 19.8% as of May 2026. Use climbs with size: 32% of firms with 100 to 249 employees and 37% of firms with 250 or more. A Census working paper finds that weighting by employment instead of firm count lifts overall adoption to 32%, and that heavy use concentrates in large firms in information, finance and professional services. By industry, information leads at 39.7% and retail trails at about 14%.
"Using AI" is not the same as AI doing the work
Be careful with the headline number. In most surveys, "using AI" can mean anything from a rebuilt pricing engine to a few seats of a chat assistant. Rolling out a copilot counts the same as rewiring a core process.
The Census definition itself got broader. In November 2025 the Bureau changed its question from AI use "in producing goods or services" to use in any business function, per a Federal Reserve note. That captures more activity, but it also means a firm drafting emails with AI now counts. Among firms that use AI, the most common function is sales and marketing, at 52%, per the Census working paper as summarized by Cambium.
When researchers ask about results instead of use, the numbers fall off a cliff. McKinsey's State of AI 2025 survey shows the ladder:
- Regularly use AI in at least one function88%
- Report any enterprise-level EBIT impact (most under 5% of EBIT)39%
- Have begun scaling AI across the enterprise~33%
- Report AI fully scaled7%
- High performers: 5%+ of EBIT from AI, plus significant value~6%
That survey skews toward large organizations with more resources than a typical mid-market company. If only about 6% of them get meaningful profit from AI, the share in the middle is unlikely to be higher. An MIT Project NANDA report from August 2025 went further, finding that 95% of enterprise generative AI pilots showed no rapid P&L impact. That figure is contested and defines failure narrowly, but it points the same way.
McKinsey also found that redesigning workflows is the change most tied to EBIT impact, yet only about 21% of generative AI users have redesigned any workflows, per a summary of the report.
Buying a tool is adoption. Changing how the work gets done is transformation. Most of the market, and most of the middle, has done the first and not the second.
What the mid-market surveys show
| Finding | Figure | Source |
|---|---|---|
| AI-using mid-market firms with AI integrated into operations | 86% | RSM Middle Market AI Survey 2026 |
| Of those, AI fully embedded across core processes | 36% | RSM |
| Pilots rated only moderate or limited success | 51% | RSM |
| Top reasons pilots underperformed: data quality / integration | 53% / 47% | RSM |
| Mid-market firms with fully governed, integrated data | 16% | Kaufman Rossin via MarketScale |
| Manufacturers stuck in AI pilot phase | 73% | Kaufman Rossin |
| Manufacturers citing legacy integration as the top barrier (vs 41% mid-market average) | 55% | Kaufman Rossin |
Taken together, the numbers point one way. Mid-market companies have bought the tools. Their people are using them. What they cannot do is connect AI to the systems and data that actually run the business. That is where the value is, and that is exactly where they are blocked.
That makes this a GDP problem as much as a technology one. The economy's productivity gains from AI are a function of how much work AI actually touches. If a third of private output is running AI at the edges instead of the core, a third of the potential dividend stays on the table.
04 · What transformation means
Inside a 30-year-old company
In the middle market, transformation means rewiring how a long-lived, profitable, complicated business runs, without stopping it. A chatbot on the website doesn't count.
Consider a composite drawn from the companies we see every week. (This is illustrative rather than a portrait of any one client.)
- The company
- A regional industrial distributor, founded in the mid-1990s. About 1,200 employees, $400 million in revenue, a dozen branches, two acquisitions still not fully integrated. PE-backed since 2021.
- The stack
- An on-premise ERP installed in 2004 and heavily customized, a separate warehouse system, pricing living in spreadsheets, and a server room that someone still has to walk into.
- The team
- An IT group of 25 that spends most of its week on tickets, patches and vendor calls. One person understands the ERP customizations, and they are eligible to retire.
- The pressure
- The sponsor wants margin expansion and a clean exit story. Customers expect real-time inventory and quoting. A competitor just started quoting in minutes instead of days.
- The instinct
- Leadership knows it has to change. It also remembers the last ERP project that ran over budget, and it knows the business cannot afford three weeks without order entry.
That last fear is rational. When one publicly traded manufacturer went live with a new ERP in North America in November 2025, it estimated the disruption cost roughly $30 million in fourth quarter net sales and about $22 million in adjusted EBITDA, per the company's own results release. A mid-market firm cannot absorb a hit like that.
So real transformation for this company means four things at once:
- A strategy for where AI changes the economics of the business.
- A data foundation that finally connects the core systems.
- Incremental modernization of the legacy platform rather than a big-bang replacement.
- People who are trained and trusted to run the new way of working.
It has to happen in months, not years, and it has to pay for itself along the way.
05 · The five traps
What holds the middle back
Ambition isn't the problem. Mid-market leaders are caught in five traps that reinforce each other.
01The legacy trap
Most mid-market cores were built for a different era: on-premise hardware, ERPs that have not had a meaningful upgrade in 15 or 20 years, and years of custom code that nobody wants to touch. Every AI use case that matters, from pricing to forecasting to service, needs clean, connected data from those systems. Only 16% of mid-market firms have it, per Kaufman Rossin. And the traditional fix, a big-bang replacement, is expensive and risky. In our experience, a full ERP migration for a mid-market company typically runs from $500,000 to several million dollars before counting the disruption, as the cutover example in the previous chapter shows.
02Too small for the giants
The global systems integrators and strategy houses were built to serve the Fortune 500. Their economics depend on large, multi-year programs staffed by deep pyramids of consultants. A $400 million distributor is not their priority account, and when it does get their attention, it gets a proposal sized for a company five times larger and a B team to deliver it.
03Too big for the boutiques
The other option is the local shop or the solo specialist. They are often talented and affordable, but they tend to do one thing: a strategy deck, a software build, or an ERP configuration. A mid-market transformation needs all three joined together, plus change management. When strategy, engineering and implementation come from different vendors, the seams are where projects fail.
04The talent trap
Most mid-market companies are not magnet employers for AI talent. They compete with big tech, the largest banks and well-funded startups for the same people, and the price keeps rising: PwC's 2026 Global AI Jobs Barometer puts the average wage premium for AI skills at 62%, up from 57% a year earlier, with AI-skill job postings growing 69% against 9% for the overall market. In RSM's 2025 survey, 39% of mid-market leaders who felt unprepared for AI named lack of in-house expertise as their top issue.
05The we've-always-done-it-this-way tax
Thirty years of success builds habits. The processes work, the people know them, and every past technology project left scar tissue. So the business keeps paying a quiet tax: manual reconciliations, spreadsheet workarounds, decisions made on last month's numbers. Leaders know the tax is rising. They just have not seen a safe way to stop paying it.
Each trap makes the others worse. Legacy systems make AI harder, which makes outside help more necessary, which runs into firms that are either too big or too narrow, which leaves the work to a stretched internal team that cannot hire its way out. Breaking the loop requires a different kind of partner.
06 · The economics
The middle is the multiplier
Every serious forecast of AI's economic impact assumes broad adoption. The middle market is where that assumption is most likely to break.
The Penn Wharton Budget Model estimates AI will raise U.S. productivity and GDP by 1.5% by 2035, nearly 3% by 2055 and 3.7% by 2075. It also estimates that about 40% of current GDP could be substantially affected by generative AI. Other estimates run higher; a widely cited Goldman Sachs analysis projected a 7% lift to global GDP over ten years, per a review by the International Center for Law & Economics.
1.5%
Projected lift to U.S. GDP from AI by 2035, a conservative forecast.
Penn Wharton
~$450B
Added output per year implied by that lift on a roughly $30 trillion economy.
Current estimate
~$150B
The share that depends on the middle market, if gains track share of output.
Current estimate
Now do the arithmetic on the conservative number. On an economy of roughly $30 trillion, a 1.5% lift is on the order of $450 billion a year in added output. If gains track share of output, about a third of that, roughly $150 billion a year, depends on middle-market companies actually getting AI into their core operations. That is our back-of-envelope estimate, not a published forecast, and it is deliberately cautious.
The exact figure matters less than the dependency. The forecasts are built on adoption curves, and the middle market is where the curve is flattest relative to its weight. If the middle stays in pilot mode, the national number misses, no matter how much the largest companies spend.
AI in the middle protects and grows jobs
The fear is that AI in the middle market means layoffs. The evidence points the other way. PwC found that headcount in the most AI-exposed sectors grew 52% since 2018 versus 36% in the least exposed, and that companies most able to use AI are expanding hiring faster than their peers, per its 2026 barometer and Metaintro's summary.
The real threat to middle-market jobs is not AI. It is a competitor that adopts AI first, quotes faster, prices smarter and takes the account.
Forty-eight million jobs are safer inside companies that modernize than inside companies that wait.
07 · The new model
AI-native services, built for the middle
The firms that finally serve the middle market well will be smaller, more senior, platform-driven, and built from day one around AI doing a large share of the work.
The reason is economic. The legacy consulting model sells hours, and its margin comes from a pyramid of junior staff doing repetitive analysis, documentation, testing and code. AI now does much of that work. That breaks the pyramid, and it changes who can afford real transformation.
The evidence on legacy modernization is already public. Amazon reported that an AI code transformation agent migrated tens of thousands of internal applications to a newer Java version, saving more than 4,500 years of development work and producing $260 million in annual cost savings, per its December 2024 announcement. A smaller firm, Novacomp, reported upgrading 10,000 lines of Java in minutes against an estimated three weeks for a senior architect, per an AWS case study. The 20-year-old custom code that traps a mid-market company is exactly the kind of work that is getting dramatically cheaper.
| Legacy transformation model | AI-native services model | |
|---|---|---|
| Team shape | Deep pyramid, junior-heavy | Small senior pods supervising AI agents |
| Unit of value | Hours billed | Outcomes shipped |
| Typical timeline | Multi-year programs | Weeks to first value, months to transformation |
| Legacy modernization | Manual rewrite or big-bang replacement | AI-assisted discovery, documentation and incremental migration |
| Strategy and build | Separate firms or separate teams | One team from strategy through production |
| Knowledge left behind | Slide decks | Working software, documented systems, trained people |
| Right-sized for | Fortune 500 | $50M to $1B companies |
This is the model we are building at Current, and we are not the only ones. Our approach pairs senior strategists and engineers with a software factory and supervised AI agents, so a small team can take a mid-market client from a view of the future to production systems without the bulk of a traditional firm. The goal is enterprise-grade transformation at a price and pace the middle can actually absorb.
The category matters more than any single firm. The middle market needs a whole new generation of these partners, fast.
08 · The next 90 days
What mid-market leaders should do now
You do not need a three-year roadmap to start. You need a clear view of where AI changes your economics and one real win in production.
- 01
Name the three processes that decide your margin.
Quoting, pricing, planning, collections, service. Put a dollar value on each.
- 02
Map the data those processes depend on.
Which systems hold it, how clean it is, and who actually understands it. This is usually where the real work hides.
- 03
Pick one use case that touches the core.
Not a chatbot. Something that runs on your ERP or operational data and moves a number the CFO tracks.
- 04
Get it into production in weeks, not quarters.
Supervised, measured, reversible. Prove the path works before you scale it.
- 05
Modernize around the core, not through it.
Use AI to document, wrap and incrementally migrate legacy systems instead of betting the company on a single cutover.
- 06
Train the people who will run it.
Your team should own the new way of working, not rent it.
- 07
Choose partners who ship.
Ask for working software in the first month and senior people on the work. If the proposal is mostly slides and headcount, it was built for someone else.
09 · Conclusion
Win the middle, win the decade
The national conversation about AI is dominated by the two ends of the barbell: frontier labs and giant enterprises on one side, apps for small businesses on the other. The outcome that matters most sits in between.
The middle market is a third of private output and 48 million jobs. It is growing, it is resilient, and it is stuck. Its companies have tried AI, and they cannot get it into the systems that run their business, because they are trapped by legacy technology, underserved by the firms of the past, and outbid for talent.
That can change now. AI has made the work of transformation cheaper and faster, and a new generation of AI-native services firms can bring enterprise-grade strategy and engineering to companies that were never on the old firms' radar.
If you lead a mid-market company, the question is not whether AI will reshape your industry. It is whether you will be the one doing the reshaping. We built Current to help you be that company.
what we believe
Five convictions, stated plainly.
The middle market is the most important place for AI in the American economy.
Transforming it is how AI shows up in GDP, in wages and in jobs.
Legacy systems are a reason to move, not a reason to wait.
The old transformation model was not built for the middle, and it will not rescue it.
AI-native partners can, and the middle market deserves them.
Nathan Dionne
Co-Founder and CEO, Current Technology Corporation
10 · Sources and methodology
Sources and methodology
- National Center for the Middle Market, Middle Market 101. middlemarketcenter.org/middle-market-101
- National Center for the Middle Market, Year-End 2025 Middle Market Indicator. middlemarketcenter.org
- National Center for the Middle Market, 5 Reasons Why You Need to Know the Mighty Middle Market. middlemarketcenter.org
- CNL Securities, Why You Should Be Talking About Middle-Market Businesses. cnlsecurities.com
- Uplevered, Middle Market Definition. uplevered.com
- U.S. Census Bureau, AI use by businesses, May 2026. census.gov
- U.S. Census Bureau, The Microstructure of AI Diffusion (CES-WP-26-25). census.gov
- Federal Reserve, Monitoring AI Adoption in the U.S. Economy. federalreserve.gov
- Cambium, Census AI adoption by function. cambium.ai
- McKinsey, The State of AI in 2025. mckinsey.com
- Silicon Canals, summary of the McKinsey 2025 survey and MIT NANDA report. siliconcanals.com
- Libertify, summary of the McKinsey workflow redesign finding. libertify.com
- RSM, Middle Market AI Survey 2026. rsmus.com
- RSM, Middle Market AI Survey 2025. rsmus.com
- MarketScale, Kaufman Rossin survey on mid-market manufacturers. marketscale.com
- Tennant Company, 2025 Fourth-Quarter and Full-Year Results. tennantco.com
- Fortune, JPMorgan's $19.8 billion tech budget. fortune.com
- Penn Wharton Budget Model, The Projected Impact of Generative AI on Future Productivity Growth. wharton.upenn.edu
- International Center for Law & Economics, AI, Productivity, and Labor Markets. laweconcenter.org
- PwC, 2026 Global AI Jobs Barometer. pwc.com
- Metaintro, summary of PwC headcount findings. metaintro.com
- Amazon, Amazon Q Developer legacy transformation announcement. aboutamazon.com
- AWS, Novacomp case study. aws.amazon.com