Which generative AI startups are making the most money today?

Are we ranking generative AI startups by revenue or profit?

For generative AI startups today, “making the most money” is best measured by revenue first, with profit treated separately because the two rankings look very different.

Private AI companies make this harder than it sounds. Anthropic and OpenAI usually talk about annualized revenue run rates, while companies such as ElevenLabs or Harvey often disclose ARR. These numbers tell us roughly how fast the business is running now, but they do not necessarily equal revenue already booked over the previous twelve months.

That distinction can create huge gaps in a market growing this quickly. Anthropic’s latest reported run rate is above $65 billion, yet Bloomberg reported preliminary second-quarter revenue of more than $11.5 billion. OpenAI’s annualized figure is above $40 billion, while its actual completed-quarter revenue has been much lower than that annualized pace.

We will therefore rank companies using the freshest credible revenue pace available, while distinguishing between actual revenue, ARR and annualized run rate whenever that changes how we should read the number.

We also exclude businesses that have stopped being independent startups. Cursor and xAI would otherwise appear high in several comparisons, but both are now owned by SpaceX. Their numbers remain useful for understanding where the money is flowing, but putting them in a current startup ranking would make the list inconsistent.

Why is generative AI revenue suddenly getting so big?

Generative AI revenue is growing much faster now because companies have moved from testing AI to paying for it across coding, research, customer service, legal work and internal workflows.

The biggest jump has happened at Anthropic. Its annualized revenue pace was around $9 billion at the end of 2025, reached roughly $47 billion by late spring and then passed $65 billion by the end of July, according to figures reported by Bloomberg and Axios. That is more than seven times the year-end level in roughly seven months.

OpenAI has grown quickly as well. Its annualized revenue has passed $40 billion, roughly double the level reported at the end of 2025. Enterprise revenue has also become much more important. OpenAI said earlier this year that businesses represented more than 40% of revenue and expected enterprise to catch consumer revenue before year-end; subsequent internal updates indicated that crossover happened earlier.

A second layer of companies has now reached hundreds of millions of dollars in annualized revenue. Cognition reported $492 million in May and expects to cross $1 billion later this year. Perplexity has moved well into the hundreds of millions. Lovable and ElevenLabs have each crossed $500 million. Glean and Suno have reached about $300 million.

The market has moved beyond a handful of chatbot subscriptions. Companies are spending serious money on AI that performs recurring work.

Which generative AI startup makes the most revenue today?

Anthropic is currently the highest-revenue independent generative AI startup we can identify, ahead of OpenAI by a large margin.

The latest reported Anthropic figure is above $65 billion in annualized revenue. OpenAI is above $40 billion. After that, the ranking collapses into the hundreds of millions: Cognition reported $492 million in May and is aiming for $1 billion, Perplexity is already much larger than it was at the start of the year, while Lovable and ElevenLabs have each crossed $500 million.

The scale difference is enormous. Even if Cognition reaches $1 billion, Anthropic would still be running at more than 60 times that level.

That is the clearest answer to the title. The highest-revenue generative AI startups today are dominated by two frontier-model companies, and every independent application startup remains far behind them.

Company	Latest credible revenue pace	Metric
Anthropic	$65B+	Annualized revenue run rate
OpenAI	$40B+	Annualized revenue run rate
Cognition	$492M disclosed, targeting $1B	Annualized revenue run rate
Perplexity	$750M+ reported	Annualized revenue
Lovable	$500M+	Annualized revenue run rate
ElevenLabs	$500M+	ARR

Is Anthropic really making more revenue than OpenAI right now?

Yes. Anthropic appears to be making more revenue than OpenAI today, even after adjusting for the accounting difference that makes Anthropic’s headline number look somewhat larger.

The latest comparison is $65 billion-plus of annualized revenue for Anthropic against more than $40 billion for OpenAI. At first glance, that puts Anthropic roughly 60% ahead.

Axios recently dug into how those numbers are calculated and found an important difference. When Claude is sold through some cloud partners, Anthropic books the full amount paid by the customer as revenue and records the partner’s share as an expense. OpenAI records only its own share of certain partner sales.

The effect is meaningful but limited. A source familiar with Anthropic’s financials told Axios that switching from gross to net reporting would reduce Anthropic’s revenue by roughly 6% to 10%. At the top end of that adjustment, a $65 billion run rate would become roughly $58.5 billion. That would still leave Anthropic almost $19 billion above OpenAI’s $40 billion figure.

Completed-quarter revenue points the same way. Bloomberg reported more than $11.5 billion of preliminary second-quarter revenue for Anthropic, up from $4.73 billion in the previous quarter. OpenAI’s comparable reported revenue remained substantially lower.

So we should not take $65 billion versus $40 billion as an exact apples-to-apples score. The useful conclusion is simpler: Anthropic’s lead appears real, and the accounting difference is too small to overturn it.

If you want more recent data on this point, please see our latest generative AI market report.

How did Anthropic get this far ahead so quickly?

Anthropic pulled ahead largely because Claude has become deeply embedded in business work, with coding providing one of the strongest ways into large enterprise accounts.

Anthropic said earlier this year that Claude Code had passed a $2.5 billion revenue run rate, more than double its level at the start of the year. Business subscriptions to Claude Code had quadrupled, and enterprise customers represented more than half of Claude Code revenue.

The customer-spending data is even more revealing. Two years ago, Anthropic had roughly a dozen customers spending more than $1 million a year. The company now says that number exceeds 500. Eight of the Fortune 10 use Claude.

The second-quarter figures show what happened when those accounts expanded. Preliminary revenue jumped from $4.73 billion in the first quarter to more than $11.5 billion in the second, according to documents reviewed by Bloomberg. That is growth of more than 140% in a single quarter from an already multibillion-dollar base.

Anthropic says companies often begin with one product such as Claude Code, an API integration or Claude for Work, then expand into other parts of the platform. The numbers suggest that this land-and-expand motion has become a major revenue engine.

Coding got Anthropic into many companies. Once Claude was inside, those customers started spending much more.

Is OpenAI still making most of its money from ChatGPT subscriptions?

No. OpenAI now makes a much larger share of its money from businesses, APIs, coding tools and other enterprise products than the old “ChatGPT subscription company” image suggests.

OpenAI said in April that enterprise customers already represented more than 40% of revenue. At the time, the company expected business revenue to catch consumer revenue around the end of the year. Later internal updates indicated that enterprise had moved ahead earlier than expected.

Codex has helped. OpenAI said it reached 3 million weekly active users earlier this year, while its APIs were processing more than 15 billion tokens per minute. Those are very different economics from selling a fixed monthly ChatGPT plan to an individual user.

OpenAI has also started adding new monetization lines. Advertising inside ChatGPT has reportedly reached roughly a $1 billion annualized pace, giving the company another source of revenue alongside consumer subscriptions, enterprise seats, API usage and coding.

OpenAI’s $40 billion-plus annualized revenue should therefore be read as the output of a broad AI platform rather than one blockbuster consumer subscription.

How far behind are generative AI application startups?

Generative AI application startups are still far behind the frontier labs: the strongest independent apps are around $500 million to $1 billion in annualized revenue, while Anthropic and OpenAI are measured in tens of billions.

Cognition is probably closest to breaking through the $1 billion barrier. It disclosed a $492 million annualized run rate in May, up from $37 million one year earlier, and has said it expects to cross $1 billion later this year.

Lovable has passed $500 million. ElevenLabs has also moved above $500 million. Glean reached $300 million, while Suno reported $300 million of annualized revenue after tripling from about $100 million only a few months earlier. Harvey ended 2025 at $190 million and has continued growing rapidly.

Even when we put several of the biggest independent application businesses together, the total remains tiny beside Anthropic. Roughly ten of the most successful AI application startups appear to produce only several billion dollars of combined annualized revenue.

That concentration is easy to miss because application startups dominate funding announcements and product launches. Revenue tells a much harsher story: most of the money is still being captured by the companies supplying the underlying models.

Which generative AI application startups are making the most money now?

Cognition, Perplexity, Lovable and ElevenLabs currently sit near the top of the independent generative AI application market by revenue.

Cognition has the strongest clean disclosure among the leading coding startups: $492 million of annualized revenue in May, compared with $37 million one year earlier. Bloomberg subsequently reported that a potential new financing could be based on Cognition reaching a $1 billion run rate.

Perplexity has also become one of the largest standalone AI applications. Recent reporting puts annualized revenue above $750 million, driven increasingly by enterprise products and agents rather than only consumer search subscriptions.

Lovable crossed $500 million of annualized revenue in June after passing $400 million only a few months earlier. ElevenLabs also crossed $500 million ARR after ending 2025 near $330 million and adding roughly $100 million of new ARR during the first quarter.

Glean reached $300 million in May. Its number deserves a small caveat because some customers pay based on consumption, so part of the reported “ARR” behaves more like an annualized usage run rate.

Suno has reached roughly $300 million in annualized revenue as well. Forbes reported that it had only been around $100 million a few months earlier.

Harvey, Sierra and Synthesia sit below that group based on their most recently disclosed figures, although several are growing quickly enough that the order can change within a quarter.

Startup	Latest credible figure	Main product
Cognition	$492M disclosed; targeting $1B	AI coding agents
Perplexity	$750M+ reported	AI search and agents
Lovable	$500M+	AI software creation
ElevenLabs	$500M+	Voice and audio AI
Glean	~$300M	Enterprise AI search and agents
Suno	~$300M	AI music
Harvey	$190M at year-end, higher now	Legal AI
Sierra	$150M+ disclosed	Customer-service agents
Synthesia	$100M+ disclosed	AI video and training

Has AI coding become the richest generative AI application market?

Yes. AI coding currently produces more high-revenue generative AI startups than any other application category we can identify.

Cognition reported $492 million of annualized revenue in May and expects to cross $1 billion. Lovable has already passed $500 million. Replit has also grown from a relatively small business into one tracking toward a much larger revenue base, although its latest figures are less cleanly disclosed.

The category looks even bigger once we include products that no longer qualify as independent startups. Cursor had crossed $2 billion in annualized revenue earlier this year and was reportedly nearing $4 billion by the time SpaceX acquired it. Anthropic’s Claude Code separately passed a $2.5 billion run rate.

That puts several billion dollars of annualized revenue into AI coding before we even count OpenAI Codex, GitHub Copilot or Google’s developer products.

The reason is fairly practical. Companies already spend enormous amounts on software engineers, and coding agents can be used every day across migrations, testing, debugging, documentation and routine development work. A tool that saves hundreds of engineering hours can support much larger spending than a consumer image generator used a few times a week.

Coding has become the clearest example of AI turning directly into an operating budget.

If you want more recent data on this point, please see our latest generative AI market report.

Can Cognition actually become a $1 billion AI software company?

Yes. Cognition is close enough to $1 billion in annualized revenue that the target now looks plausible rather than promotional.

Cognition reported $492 million of annualized revenue in May. One year earlier, the figure was only $37 million. That is roughly 13 times growth in twelve months.

The company also said enterprise usage of Devin had been growing around 50% month over month for six consecutive months. Customers include Goldman Sachs, Mercedes-Benz, NASA and government organizations, which gives the business a much broader base than a developer tool paid for mostly by individuals.

Bloomberg later reported that Cognition was discussing another funding round based on reaching a $1 billion annualized revenue run rate. The company itself has said it expects to cross $1 billion later this year.

We still should not promote a future target into current revenue. Today, the firmest public number is $492 million, with much higher figures implied by subsequent fundraising discussions.

Still, a company that moved from $37 million to $492 million in one year does not need an implausible final step to reach $1 billion.

Is Perplexity still mostly an AI search engine?

Perplexity is now making enough money from agents and business users that describing it simply as an AI search engine misses where the company is going.

Perplexity’s annualized revenue has reportedly moved above $750 million, putting it ahead of most independent generative AI application startups.

Earlier growth came largely from subscriptions around its answer engine. The newer acceleration has coincided with Perplexity pushing deeper into agents, enterprise products and tools designed to perform tasks rather than only retrieve information.

That changes the revenue ceiling. A search subscription usually gives a company a fairly fixed amount per customer each month. Agents can generate more usage as customers hand them additional work, especially inside companies where the same system can be deployed across many employees.

Perplexity is starting to resemble a general AI work platform with search as its entry point.

The distinction helps explain why its revenue has climbed so quickly even while Google, OpenAI and Anthropic all offer increasingly capable search-like products.

Is ElevenLabs now the biggest generative AI startup outside text and coding?

ElevenLabs is currently one of the largest independent generative AI companies outside text and coding, with more than $500 million in ARR.

The company ended 2025 near $330 million ARR. It then added around $100 million of net new ARR during the first quarter, reaching roughly $450 million, and crossed $500 million shortly afterward.

That growth has come while ElevenLabs expands beyond synthetic narration. Voice agents are now being used for customer support, sales, recruitment and other business interactions. The company has also moved into music and broader multimodal products.

This gives ElevenLabs several ways to grow spending inside the same customer. A company may begin by generating voice content and later use the technology for thousands of customer calls or automated conversations.

The trajectory is unusually strong for a company outside the coding boom. ElevenLabs added roughly $170 million of ARR between the end of last year and early May alone.

If you want more recent data on this point, please see our latest generative AI market report.

Can a narrow vertical AI startup make hundreds of millions in revenue?

Yes. Harvey shows that a generative AI startup focused on one expensive profession can already reach hundreds of millions of dollars in recurring revenue.

Harvey reached $100 million ARR in 2025 and ended the year around $190 million. More recent reporting indicates it has continued growing rapidly, although the freshest clean company-disclosed milestone remains lower than some third-party estimates circulating today.

The interesting part is the customer economics. Harvey sells into law firms and corporate legal departments where experienced professionals can cost hundreds of dollars an hour. If the software removes hours of research, drafting or document review, a law firm can justify paying far more per user than a consumer would pay for a general chatbot.

Harvey has also expanded across large law firms and major corporate legal teams rather than depending on one or two giant contracts.

That gives us an important benchmark for vertical AI. A startup does not need hundreds of millions of users if every customer workflow is expensive enough.

Legal AI may be the cleanest example so far.

Are creative generative AI startups actually making serious revenue?

Yes. Creative generative AI has produced real $100 million-plus businesses, although the category still trails coding and enterprise AI at the very top.

Suno is the standout recent example. Its annualized revenue rose from roughly $100 million in October to about $300 million by February, according to Forbes. The company also said it had crossed 2 million paying subscribers.

That means monthly revenue moved from roughly $8 million to around $25 million in only a few months.

Midjourney has been another unusually strong creative AI business. Historical estimates put revenue in the hundreds of millions, and the company became profitable without following the usual venture-funded growth model. Fresh numbers are less transparent, so we would not place Midjourney precisely in the current ranking.

Synthesia provides a more enterprise-heavy version of creative AI. The company crossed $100 million ARR through AI-generated training and corporate video and is now pushing into interactive training sessions where employees can practice conversations with AI avatars.

Creative AI is producing substantial businesses. What we have not seen yet is a standalone image, video or music startup reaching anything close to the tens of billions now generated by the leading model labs.

Do AI startup ARR figures make these companies look bigger than they really are?

Sometimes. ARR and annualized run rates are useful in this market, but they can make very fast-growing AI companies appear much larger than the revenue they have actually collected.

Anthropic gives us a clean example. Its annualized pace passed $65 billion at the end of July. That number says that revenue around that period, extended across twelve months, would exceed $65 billion. Anthropic had obviously not already booked $65 billion over the preceding year.

The gap becomes especially dramatic when revenue is accelerating every month. Cognition reported $492 million annualized in May after producing only $37 million at the same point a year earlier. If the latest month is much stronger than the previous eleven, annualizing that one month gives a better picture of where the business is heading but a worse picture of what it has already earned.

Glean adds another wrinkle. The company describes its top line as ARR, yet some customers pay based on consumption. TechCrunch correctly noted that this makes part of the $300 million closer to annualized usage than fixed recurring subscription revenue.

We still use these figures because trailing twelve-month revenue can be hopelessly stale for an AI startup growing fivefold or tenfold in a year. They just need to be read as current speed rather than money already in the bank.

Which generative AI startups are actually profitable now?

Anthropic appears to have reached operating profitability for at least one recent quarter, while OpenAI remains heavily loss-making despite generating far more revenue than almost every software company on the planet.

Bloomberg reported that Anthropic produced more than $11.5 billion of preliminary second-quarter revenue and positive adjusted operating income. The Financial Times had earlier reported projections for roughly $10.9 billion of quarterly revenue and about $559 million of operating profit.

That is a significant change from the usual frontier-model economics. Training and serving large models consume enormous amounts of compute, and most leading labs have spent aggressively to keep growing.

OpenAI is still on the other side of that curve. Its revenue is above $40 billion on an annualized basis, but the company continues to spend heavily on model development, infrastructure and compute capacity. Its path to sustained profitability remains much longer.

Smaller generative AI companies can look healthier. Midjourney has historically been profitable while generating hundreds of millions of dollars in revenue. Harvey, despite strong growth, has continued investing heavily and has not reached comparable profitability.

Anthropic currently has the most interesting combination of revenue scale and improving operating economics. OpenAI remains enormous, but scale alone has not translated into profit yet.

If you want more recent data on this point, please see our latest generative AI market report.

Are businesses now spending more on generative AI than consumers?

At the leading AI startups, business spending is increasingly driving the largest revenue pools.

Anthropic is the clearest case. The company says more than 500 customers now spend over $1 million a year, while enterprise customers generate more than half of Claude Code revenue. Eight of the Fortune 10 use Claude.

OpenAI has moved in the same direction. Enterprise represented more than 40% of its revenue earlier this year and was expected to overtake consumer revenue much later; the crossover appears to have arrived ahead of schedule.

The application leaderboard is even more business-heavy. Cognition sells Devin into companies such as Goldman Sachs and Mercedes-Benz. Harvey sells legal AI to law firms and corporate legal teams. Glean sells enterprise search and agents. Sierra sells customer-service agents. ElevenLabs increasingly earns money from companies deploying voice systems at scale.

Consumer AI can still create large businesses. Suno has 2 million paying subscribers, and ChatGPT itself remains one of the largest consumer subscription products in technology.

The fastest path to very large revenue, though, increasingly runs through company budgets. A business will spend far more when an AI product is doing work that previously required employees, contractors or conventional software.

Does today’s generative AI revenue ranking tell us who will win the market?

No. Today’s generative AI revenue ranking tells us who has monetized fastest so far, while the long-term winner could still change as model quality, pricing and distribution move around.

Anthropic’s rise shows how quickly that can happen. OpenAI looked overwhelmingly dominant not long ago. Anthropic now appears to generate more revenue, helped by Claude’s strength in coding and enterprise work.

OpenAI still has several major advantages: enormous consumer distribution, a huge developer ecosystem, rapidly growing enterprise sales, Codex and a broad API business. A $40 billion-plus revenue run rate also gives it plenty of room to fund new products and compete aggressively.

Application startups face a different problem. Cognition, Perplexity, Lovable, ElevenLabs and Harvey often rely on models supplied by the same frontier labs that can move into their markets. Their best protection comes from owning the customer workflow, proprietary context, distribution or data deeply enough that better foundation models make their products stronger.

Revenue gives us the strongest evidence available that customers find a product useful enough to pay for it now. It still cannot tell us which company will hold the strongest position five years from today.

If you want more recent data on this point, please see our latest generative AI market report.

So which generative AI startups are making the most money today?

Anthropic is the clear revenue leader today, OpenAI is second, and the gap between those two companies and every independent generative AI application startup remains enormous.

Anthropic’s latest annualized revenue figure is above $65 billion. OpenAI has crossed $40 billion. Recent reporting on accounting methods suggests Anthropic’s gross treatment of some cloud sales may inflate the comparison by roughly 6% to 10%, but that adjustment still leaves Anthropic comfortably ahead.

Below the frontier labs, Cognition and Perplexity are the strongest independent application contenders. Cognition has disclosed $492 million of annualized revenue and expects to pass $1 billion. Perplexity has reportedly crossed $750 million. Lovable and ElevenLabs are already above $500 million, while Glean and Suno sit around $300 million.

Coding stands out most clearly across the application layer. Cognition, Lovable and Replit have all grown rapidly, while Cursor reached several billion dollars of annualized revenue before its acquisition and Claude Code passed $2.5 billion inside Anthropic. No other AI application category has produced that many businesses at that scale.

Profit gives us a different picture. Anthropic appears to have reached positive operating income in a recent quarter, while OpenAI continues to lose heavily despite its huge top line. Smaller companies such as Midjourney have shown that generative AI can also produce profitable businesses without tens of billions of dollars in revenue.

As of now, the answer is quite sharp: Anthropic makes the most revenue among independent generative AI startups, OpenAI follows, and nobody else is close. The more interesting race is happening underneath them, where coding, agents, voice and vertical AI are turning a growing group of startups into $300 million-to-$1 billion businesses.

Current tier	Company	Best current revenue evidence	What we can say confidently
1	Anthropic	$65B+ annualized	Clear revenue leader
2	OpenAI	$40B+ annualized	Clear second place
Next group	Perplexity	$750M+ reported	One of the largest independent AI apps
Next group	Cognition	$492M disclosed, targeting $1B	Could soon become a $1B application business
Next group	Lovable	$500M+	Major AI coding winner
Next group	ElevenLabs	$500M+	Leading independent voice AI company
Following group	Glean	~$300M	Large enterprise AI business
Following group	Suno	~$300M	Leading consumer creative AI startup

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.