Which AI companies have the highest revenue per employee?
In our AI infrastructure market deck, you will find everything you need to understand the market
SUMMARY
Anthropic has the strongest claim to the highest revenue per employee among major AI companies, with Cursor, OpenAI, Nvidia and Midjourney forming the next tier of extreme outliers.
The headline ratios are astonishing, but they are not all equally clean. Nvidia’s roughly $5.1 million per employee is based on audited annual revenue and a reliable workforce count, while private-company figures often mix annualized run rates with headcount snapshots from different months.
Anthropic’s lead is large enough to survive a lot of skepticism. Even if its workforce were materially above the roughly 4,020-person estimate used here, or its reported revenue were adjusted downward for cloud-partner accounting, it would still sit near the top.
Cursor may be the more revealing application-company example. Before its SpaceX acquisition, it had reached roughly $4 billion of annualized revenue with a workforce still measured in the hundreds, showing how much revenue a self-service AI product can support without a giant organization.
Midjourney remains the purest tiny-team case. The old 11-employee story is outdated, but roughly $300 million of actual 2024 revenue against about 60 employees still puts it around $5 million per person.
The biggest mistake is to read revenue per employee as profitability. Frontier-model companies can look spectacular on labor leverage while spending enormous sums on GPUs, cloud infrastructure and model training.
That is why ElevenLabs is more impressive than its roughly $1.1 million ARR per employee might suggest. Its ratio is lower than Anthropic’s or Cursor’s, but it also has reported profit, which makes the underlying economics more tangible.
AI coding appears structurally well suited to very high revenue per employee because customers can justify high software spend against six-figure engineering salaries. Cursor, Cognition and Lovable all show the same pattern, although Lovable also shows how quickly the ratio can fall once sales, support and management hiring catches up.
Business model matters as much as product quality. Scale AI looks less efficient on this metric partly because its production system has historically relied on much more human labor, including a large contributor network that is not captured by normal employee counts.
The durable winners will probably be products with expensive use cases, self-service distribution and little human implementation. If Anthropic, Cursor-like application companies and Midjourney-style products can keep these ratios after hypergrowth fades, AI will have pushed the labor-productivity ceiling far beyond traditional software.
This market map, featured in our AI infrastructure market deck, highlights top companies and startups in the AI infrastructure market
What does revenue per employee actually tell us about an AI company?
Revenue per employee tells us how much commercial output an AI company supports with its own workforce, and today the best AI companies are reaching levels that would have looked absurd in software a few years ago.
The calculation itself is simple: divide annual revenue by employee count. For private AI companies, we often have to use annual recurring revenue or an annualized revenue run rate instead because audited annual accounts are unavailable. That makes the result less clean, but the order of magnitude is still useful.
Take Nvidia. The company generated $215.9 billion of fiscal-year revenue with roughly 42,000 employees, giving us about $5.1 million per employee. Palantir generated $4.5 billion in 2025 with 4,429 employees, or just over $1 million per person. Midjourney reported $300 million of 2024 revenue and Forbes currently lists only 60 employees, which works out to roughly $5 million per employee even though the revenue and headcount dates do not perfectly match.
These numbers measure labor leverage rather than overall business efficiency. AI companies can keep employee counts unusually low because GPUs, cloud infrastructure and outside AI models perform work that would otherwise require people. A company spending billions on compute can therefore look spectacular on revenue per employee while still having weak margins.
That distinction will matter throughout the ranking.
Why is ranking AI companies by revenue per employee so messy right now?
There is no perfectly accurate AI revenue-per-employee leaderboard today because private-company revenue and headcount are moving too quickly, and companies do not report revenue in the same way.
Anthropic shows how quickly a ratio can become stale. Its annualized revenue was about $9 billion at the end of 2025, reached roughly $47 billion by May and passed $65 billion by the end of July, according to Bloomberg reporting. A headcount estimate from the beginning of the year can therefore be paired with revenue that became several times larger only months later.
Employee figures have the same problem. Revelio Labs currently estimates Anthropic at about 4,020 employees based on its March workforce snapshot. Anthropic has continued hiring aggressively since then, with more than 1,000 active job postings recorded during 2026. Dividing $65 billion by 4,020 gives $16.2 million per employee, but we should treat that as an indication of scale rather than a precise present-day ratio.
Revenue definitions create another headache. Axios recently reported that Anthropic recognizes some Claude sales through cloud partners on a gross basis and later records the partner's share as an expense. OpenAI reportedly treats comparable revenue differently. That accounting choice makes Anthropic's top line look larger even when the underlying economics are closer.
So we can rank companies in broad tiers with reasonable confidence. Pretending that Anthropic is precisely at $16.17 million while OpenAI sits at exactly $7.82 million would give the numbers more precision than the evidence deserves.
As this chart shows, and as featured in our AI infrastructure market deck, search interest in AI infrastructure has risen sharply
Which AI companies have the highest revenue per employee right now?
Anthropic has the strongest claim to the revenue-per-employee lead today, with Cursor, OpenAI, Nvidia and Midjourney forming the next group of extraordinary outliers.
Anthropic passed $65 billion in annualized revenue by the end of July. Against Revelio Labs' roughly 4,020-person workforce estimate, that produces around $16.2 million per employee. Even if Anthropic had already grown to 5,000 employees, the ratio would still be $13 million.
Cursor reached roughly $4 billion in annualized revenue before its acquisition by SpaceX. Forbes had described the company as having around 400 employees earlier in the year. That pairing gives an extraordinary $10 million per employee, although both headcount and revenue continued changing quickly around the transaction.
OpenAI is generating roughly $40 billion at an annualized rate. Forbes listed 5,000 employees in the spring, which would give $8 million per employee. OpenAI has almost certainly hired beyond that figure, so the current ratio should be lower.
Nvidia gives us the cleanest large-company benchmark at around $5.1 million on audited annual revenue. Midjourney lands at roughly $5 million using its $300 million 2024 revenue against Forbes' current 60-person workforce.
The gap below those companies is substantial. ElevenLabs sits around $1.1 million based on more than $500 million ARR and 450 employees. Lovable's latest picture is now closer to $1.3 million per employee after its workforce expanded sharply toward 450 while annualized revenue approached $600 million. Earlier snapshots above $2.5 million per employee are already outdated.
| Company | Latest useful revenue measure | Relevant headcount | Approx. revenue per employee |
|---|---|---|---|
| Anthropic | $65B+ annualized | ~4,020* | ~$16.2M* |
| Cursor | ~$4B annualized | ~400* | ~$10M* |
| OpenAI | ~$40B annualized | ~5,000* | ~$8M* |
| Nvidia | $215.9B actual revenue | ~42,000 | ~$5.1M |
| Midjourney | $300M actual 2024 revenue | ~60* | ~$5.0M* |
| ElevenLabs | $500M+ ARR | ~450 | ~$1.1M |
| Lovable | ~$600M annualized | ~450 | ~$1.3M |
*Private-company revenue and headcount snapshots do not perfectly coincide, so these ratios should be read as approximate.
Is Anthropic really number one in revenue per employee?
Anthropic is the most convincing number one, even after we account for imperfect headcount data and its more generous treatment of some cloud-partner revenue.
The scale of the revenue jump is hard to explain away. Anthropic went from around $9 billion of annualized revenue at the end of 2025 to $47 billion in May and more than $65 billion by the end of July. Revenue increased more than sevenfold in roughly seven months.
Headcount did grow, but nowhere near that quickly. Revelio Labs estimates that Anthropic had 3,173 employees at the end of 2025 and about 4,020 by March. Even giving Anthropic substantially more employees than that today leaves the company far above normal software productivity. At 6,000 employees, for example, $65 billion still works out to $10.8 million per person.
The biggest qualification comes from accounting. Axios found that Anthropic records the full value of some Claude sales made through cloud partners and then puts the partners' share through expenses, whereas OpenAI handles comparable arrangements differently. That inflates Anthropic's reported revenue relative to a net presentation.
Still, the accounting difference would have to remove a huge share of Anthropic's reported run rate to push the company anywhere near ordinary software-company levels. Even a 40% haircut to $65 billion would leave $39 billion. Against 4,020 employees, we would still get roughly $9.7 million per person.
Anthropic's exact ratio is debatable. Its place in the top tier really is not.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows annual VC investment in AI infrastructure startups
Is Cursor actually more efficient than OpenAI?
Cursor probably generated more revenue per employee than OpenAI before joining SpaceX, and the gap may have been substantial.
Cursor's annualized revenue went from roughly $100 million in early 2025 to $2 billion in February 2026, $3 billion in late April and $4 billion by early June. Forbes described Cursor as having around 400 employees in March. If headcount stayed anywhere near that range while revenue reached $4 billion, the company was producing roughly $10 million per employee.
The growth path is almost more striking than the endpoint. Around a year earlier, Cursor had crossed $100 million in annualized revenue with only about 20 employees and no traditional sales team. It then hired hundreds of people, moved aggressively into enterprise sales and still managed to increase revenue much faster than its workforce.
OpenAI operates at a much larger scale. Its annualized revenue has recently reached about $40 billion, up from $20 billion at the end of 2025. Forbes listed approximately 5,000 employees in the spring. That snapshot gives $8 million per employee, although OpenAI's current workforce is likely larger.
The two businesses also carry very different costs. Cursor historically depended heavily on models from Anthropic and OpenAI. A large piece of the productive work therefore happened inside another company's infrastructure. OpenAI builds frontier models, trains them and runs an enormous global consumer and enterprise platform.
Cursor's raw labor productivity still deserves attention. Before the SpaceX acquisition, very few companies in any industry had ever approached $4 billion of revenue run rate with a workforce measured in only hundreds.
Is Midjourney still the ultimate tiny-team AI company?
Midjourney remains the cleanest example of a genuinely tiny AI company producing hundreds of millions in revenue, although the famous 11-employee figure is badly outdated.
Forbes now lists Midjourney at about 60 employees and says the company generated $300 million in 2024 revenue. Pairing those figures gives approximately $5 million per employee. Because the revenue number is older than the employee count, this calculation is actually conservative if Midjourney has continued growing.
The old story was even more extreme. Midjourney was widely reported at roughly $200 million of revenue with a core team of around 11 people, producing an eye-catching figure above $18 million per employee. That snapshot still circulates online, but using it today would make the company look far smaller than it is.
What makes Midjourney especially interesting is how little conventional organization sits behind the revenue. The company grew through a self-service subscription product, initially centered on Discord, without building a large enterprise sales force. It also avoided venture funding and is profitable, according to Forbes and PitchBook data.
That makes Midjourney a cleaner efficiency case than many frontier-model companies. Anthropic produces far more revenue per employee on current run-rate figures, but Anthropic also carries enormous compute commitments. Midjourney achieved its result with a simpler product, a much smaller organization and no outside equity financing.
If the question is which AI company best proves that a tiny team can build a huge business, Midjourney still has the strongest case.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows why CoreWeave is winning in AI infrastructure
Are AI coding companies becoming the new revenue-per-employee champions?
AI coding companies are producing some of the highest revenue per employee in the industry because customers will pay a lot for software that directly replaces or accelerates expensive engineering work.
Cursor is the extreme example. At roughly $4 billion of annualized revenue and around 400 employees in the closest usable workforce snapshot, the implied ratio reaches about $10 million per person.
Cognition is moving in the same direction. The company behind Devin reported a $492 million revenue run rate in May, up from only $37 million a year earlier. Forbes listed approximately 250 employees in the spring, putting that snapshot around $2 million per employee. Sacra now estimates Cognition near $900 million of annualized revenue after the Windsurf acquisition, although we would not mix that newer estimate with the older headcount and present the result as a precise current ratio.
Lovable looked even more extreme earlier in the year. It reached $400 million ARR with 146 employees, or roughly $2.7 million per person. The company then hired very quickly. Recent reporting puts annualized revenue around $600 million and workforce plans or headcount near 450, bringing the current figure closer to $1.3 million.
That decline is useful evidence, not a disappointment. Lovable shows what happens when an AI startup moves from a viral product into a larger enterprise organization: sales, support, infrastructure, recruiting and management start catching up with revenue.
Coding still has unusually good economics. Paying an AI company several thousand dollars per developer can be easy to justify when a software engineer costs a company well into six figures each year. That gives Cursor, Cognition and similar products room to charge meaningful prices without requiring armies of employees to deliver the service.
Is ElevenLabs more efficient than its revenue-per-employee ranking suggests?
ElevenLabs currently generates around $1.1 million of ARR per employee, but its profitability makes that number more impressive than several larger headline ratios.
ElevenLabs ended 2025 at roughly $350 million ARR and passed $500 million during the first four months of 2026, according to the company. Forbes lists about 450 employees, putting ARR per employee a little above $1.1 million.
That sounds modest beside Anthropic or Cursor. The economics underneath it are unusually strong. Forbes reported that ElevenLabs made about $116 million in profit during 2025. Voice agents now represent more than half of its business, with companies using ElevenLabs for customer support, sales, recruiting and other conversations.
ElevenLabs has also been adding local teams around the world as it moves deeper into enterprise sales. Those people lower revenue per employee even if they improve retention, customer size and profit.
This is why the ranking is not a scoreboard for business quality. A company producing $1 million per employee at strong margins can have a much better business than one producing $10 million per employee while spending most of that revenue on GPUs and cloud contracts.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows annual funding in AI infrastructure startups
Why does Scale AI look much less efficient than Anthropic or Cursor?
Scale AI's revenue per core employee is much lower because its business has historically required far more human work than a self-service AI product.
Scale said it had more than 1,000 employees toward the end of 2025, while Forbes reported just under $1 billion of revenue for that year. On that company-disclosed employee base, Scale sits around $1 million of revenue per employee.
Even that figure needs care. Scale's data business uses a much broader network of contributors and contractors who do not necessarily appear in the normal employee denominator. Scale now says it has paid more than $1 billion to contributors globally. Revenue per employee therefore makes the company look more labor-light than its full production system really is.
The business is changing. Scale says its applications operation more than doubled revenue in the second half of 2025, while data labeling still accounts for the vast majority of revenue. Management expects applications eventually to become larger.
If that transition happens, Scale could become much more productive per internal employee. Enterprise AI software requires people, but usually fewer people per dollar of revenue than managed data-labeling work.
Scale is a useful counterexample because “AI company” covers very different businesses. A model lab, coding tool and data-services operation can all sell into the same AI boom while having completely different labor economics.
How far ahead are AI companies compared with Nvidia, Meta and Palantir?
The most productive AI-native companies have moved beyond the revenue-per-employee levels of even exceptionally efficient public technology companies, although Nvidia remains the benchmark we trust most.
Nvidia generated $215.9 billion of fiscal-year revenue with roughly 42,000 employees. That comes to around $5.1 million per person. Producing that ratio at more than $200 billion of sales is extraordinary.
Meta generated roughly $201 billion of 2025 revenue with 78,865 employees, or around $2.5 million per person. Palantir generated $4.5 billion with 4,429 employees, landing just above $1 million.
Anthropic's current run-rate ratio appears more than three times Nvidia's. Cursor's last standalone snapshot was roughly twice Nvidia's. OpenAI also appears to sit above Nvidia unless its workforce has already grown far beyond the 5,000 employees Forbes reported in the spring.
That comparison gives us a useful reality check. A few million dollars per employee has already been demonstrated by audited public companies. The AI labs and coding startups are pushing the ceiling higher, but their private-company figures are much less standardized.
| Company | Approx. revenue per employee | What we are measuring |
|---|---|---|
| Anthropic | ~$16.2M* | Annualized private-company revenue |
| Cursor | ~$10M* | Annualized private-company revenue |
| OpenAI | ~$8M* | Annualized private-company revenue |
| Nvidia | ~$5.1M | Audited fiscal-year revenue |
| Meta | ~$2.5M | Audited annual revenue |
| Palantir | ~$1.0M | Audited annual revenue |
*Approximate private-company snapshot; current headcount may be higher.
This chart, included in our AI infrastructure market deck, compares the main business model options for AI cloud infrastructure providers
Does high revenue per employee mean an AI company is actually profitable?
High revenue per employee can coexist with huge losses in AI, so we should never use this metric alone to decide which company has the best economics.
Frontier AI makes the problem obvious. Anthropic can support more than $65 billion of annualized revenue with only a few thousand employees because Claude does the work on GPUs rather than through a massive human workforce. Running those GPUs is extremely expensive. Anthropic has signed enormous infrastructure agreements with cloud and specialized compute providers as it expands capacity.
OpenAI faces the same basic economics. Revenue can rise without hiring people in proportion to usage, yet every additional ChatGPT, Codex or API workload still consumes computing resources.
ElevenLabs gives us a useful contrast. The company has a much lower revenue-per-employee ratio than Anthropic, around $1.1 million on the latest clean snapshot, but Forbes reported approximately $116 million of profit for 2025. Midjourney is also described as profitable while operating with only around 60 employees.
Palantir makes the same point from the public market. Revenue per employee sits around $1 million, far below the frontier-model labs, while its GAAP operating margin has become exceptionally high.
Revenue per employee tells us who has removed the most human labor from the organization. To know who has built the better economic machine, we still need gross margins, operating expenses, compute costs and free cash flow.
If you want more recent data on this point, please see our latest AI infrastructure market report.
Are GPUs replacing employees inside AI companies?
GPUs, outside models and automation are replacing enough human work to push revenue per employee far above traditional software norms, and that shift is one of the main reasons these ratios are now possible.
Midjourney can serve millions of image requests without hiring illustrators. ElevenLabs can generate voices without employing armies of voice actors. Cursor can help customers produce more software without supplying consultants for every project. Anthropic can sell Claude to millions of users while a relatively small research and engineering organization maintains the underlying models.
Part of the missing workforce has simply moved elsewhere. Cursor has historically paid Anthropic and OpenAI for model usage. Frontier labs buy or rent enormous amounts of computing infrastructure. Scale relies on external contributors alongside its internal staff.
A simple example shows why the distinction matters. Imagine one company producing $1 billion of revenue with 1,000 employees and $100 million of infrastructure costs. Another reaches the same revenue with 100 employees while spending $700 million on outside models and compute. The second company would have ten times the revenue per employee even though its operating economics could be much worse.
Still, shifting work from employees to machines is a genuine productivity change. The cost structure has moved rather than disappeared.
That is why revenue per employee is useful for understanding AI companies, as long as we remember exactly what it measures.
This chart, featured in our AI infrastructure market deck, shows the share of revenue generated by each customer segment in the AI infrastructure market
Which AI business models can keep revenue per employee this high?
The AI companies most likely to preserve unusually high revenue per employee are products with expensive use cases, self-service distribution and very little human implementation.
Coding fits that description particularly well. Cursor grew from roughly $100 million to $4 billion of annualized revenue while moving from dozens to hundreds of employees. Cognition has also expanded revenue dramatically without building a workforce remotely comparable to an IT-services company handling the same amount of engineering work.
Midjourney shows the consumer version. Customers subscribe directly, create their own images and rarely need a salesperson or implementation team. That structure explains how the company could reach hundreds of millions in annual revenue with only dozens of employees.
Enterprise-heavy businesses face more pressure. Lovable's workforce expanded quickly as it pushed into larger customers. ElevenLabs has been opening local offices and building go-to-market teams. Scale has always needed more operational labor because of the nature of data work.
The craziest ratios should keep coming from AI products where one employee can support enormous amounts of automated customer activity. Model labs can reach spectacular headline figures too, although their compute bills make revenue per employee a weaker guide to the underlying economics.
So which AI companies really have the highest revenue per employee today?
Anthropic has the strongest claim to the highest revenue per employee among major AI companies, while Cursor produced the most extraordinary ratio among large AI applications and Midjourney remains the standout tiny-team business.
Anthropic's latest reported annualized revenue exceeds $65 billion. Against Revelio Labs' roughly 4,020-person workforce estimate, that comes to about $16 million per employee. The exact figure is certainly moving and Anthropic's accounting makes comparisons imperfect, but the company would remain near the top even with a much larger employee denominator.
Cursor's last standalone numbers are nearly as striking. Roughly $4 billion of annualized revenue against around 400 employees gives an implied figure close to $10 million per person. The company's acquisition by SpaceX now makes future standalone comparisons much less useful.
OpenAI probably follows in the several-million-dollar range. Using its roughly $40 billion annualized revenue and Forbes' older 5,000-person employee snapshot gives $8 million, though current hiring means the real figure should be lower.
Midjourney sits around $5 million using $300 million of actual 2024 revenue and its more recent 60-person headcount. Nvidia also produces about $5.1 million per employee, with the advantage of audited figures and a workforce count we can trust far more.
Below that exceptional group, we currently get roughly $1 million to $2 million per employee for companies such as ElevenLabs, Lovable, Cognition on its earlier disclosed run rate, Palantir and Scale using its internal workforce.
The bigger story is how fast the ceiling moved. Around $1 million of revenue per employee once looked like an unusually efficient software company. Nvidia proved that a giant technology business could exceed $5 million. Anthropic, Cursor and OpenAI are now testing whether AI companies can live above that level for years rather than briefly passing through it during hypergrowth.
Our answer is fairly clear: Anthropic leads the measurable large AI companies, Cursor was probably the strongest AI-application outlier before its acquisition, and Midjourney still gives us the purest example of a tiny organization turning AI into hundreds of millions of dollars of revenue.
The next question is whether these ratios survive maturity. If Anthropic can stay above $10 million per employee after its workforce, infrastructure and public-company obligations expand, we will be looking at a genuinely new productivity regime rather than a temporary startup anomaly.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows how GPU cloud infrastructure technology has evolved over time
OUR METHODOLOGY
This analysis ranks AI companies by revenue per employee while treating the metric narrowly as labor leverage, not as a proxy for profitability or total business efficiency. The basic calculation is revenue divided by employee count, but private-company reporting makes the comparison much messier than the formula suggests.
We broke the comparison into the factors that can materially change the result: revenue basis, workforce size, timing, accounting treatment and business-model structure. We prioritized audited annual revenue where available. For private companies without audited public accounts, we used ARR or annualized revenue run rate and labeled the resulting ratios accordingly.
We paired revenue and workforce figures from the closest defensible periods rather than combining the most flattering numbers available. This is especially important for companies such as Anthropic, Cursor and OpenAI, where revenue and employee counts have both been changing quickly.
We also stress-tested the apparent leaders. Where rapid hiring, cloud-partner accounting, contractors and contributors, or reliance on outside models and compute could materially alter the ratio, we checked whether the ranking still held under more conservative assumptions. The goal is to identify robust tiers rather than pretend the evidence supports decimal-point precision.
Audited public companies such as Nvidia, Meta and Palantir are used as calibration points because their standardized reporting gives us a cleaner reference for what unusually high revenue per employee already looks like at scale.
Key sources used for this analysis include Axios on Anthropic and OpenAI revenue run rates, Axios on gross versus net cloud-partner revenue accounting, Revelio Labs on Anthropic workforce estimates, Forbes on Cursor's $4 billion annualized revenue, Forbes on Cursor's workforce and earlier growth, Forbes on OpenAI's workforce, Nvidia's SEC filing, Forbes on Midjourney, ElevenLabs on passing $500 million ARR, Forbes on ElevenLabs' workforce and profit, TechCrunch on Lovable's $400 million ARR and 146 employees, Lovable on its expansion, Cognition on its $492 million revenue run rate, Forbes on Cognition's workforce, Scale AI on its employee base, Scale AI on its contributor network, Forbes on Scale AI's 2025 revenue and business mix, Meta's 2025 results, and Palantir's 2025 10-K.
In our AI infrastructure market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- Which AI infrastructure startups generate the most revenue today?
- Which AI startups are actually profitable?
- Which AI startups are growing without burning billions?
