What are the top generative AI startups by revenue today?

The ranking of top startups in the generative AI market by revenue

Below is a table ranking the startups of this market by their current revenue scale. You can find our methodology at the end of this page.

If you want a deeper understanding of the market and its current dynamics, get our report on the generative AI market.

Which generative AI startups make the most money today?

Anthropic ranks first among generative AI startups, with a revenue run-rate above $65 billion, followed by OpenAI at more than $40 billion and Cursor / Anysphere above $4 billion. The gap between these three companies and the rest of the market is enormous, although several other startups have already crossed the billion-dollar mark on ARR or annualized revenue.

The ranking also shows how quickly the market has spread beyond general-purpose AI models. Coding tools, inference infrastructure, AI search, voice generation and enterprise applications are now generating hundreds of millions of dollars in annualized sales. Chinese companies have also established a substantial presence near the top.

One important distinction is that these figures measure different things. Some are annualized revenue run-rates, others are ARR, recognized segment revenue or bookings. We preserve those differences rather than treating every figure as achieved annual revenue.

Rank	Startup	Reported figure	Metric and period
1	Anthropic	>$65B	Revenue Run-Rate, End-Jul 2026
2	OpenAI	>$40B	Revenue Run-Rate, Aug 2026
3	Cursor / Anysphere	>$4B	Revenue Run-Rate, Early Jun 2026
4	Z.ai / Zhipu AI	$1.8B	ARR, Sep 2026
5	Fireworks AI	>$1B	Revenue Run-Rate, Jul 2026
6	Together AI	>$1.15B	Bookings, Latest quarter 2026
7	Moonshot AI / Kimi	>$1B	ARR, Aug 2026
8	Cognition	$900M	Revenue Run-Rate, Aug 2026
9	Perplexity	>$750M	Revenue Run-Rate, Aug 2026
10	SenseTime	CNY 2.33B (~$328M)	Segment Revenue, H1 2026
11	Replit	~$525M	Revenue Run-Rate, Apr 2026
12	ElevenLabs	>$500M	ARR, May 2026
13	Lovable	>$500M	Revenue Run-Rate, Jun 2026
14	Manus	~$400–500M	Revenue Run-Rate, 2026
15	DeepSeek	~$400–500M	Revenue Run-Rate, 2026
16	Harvey	>$400M	ARR, Sep 2026
17	Mistral AI	>$400M	ARR, Feb 2026
18	fal	~$400M	Revenue Run-Rate, Feb 2026
19	Glean	$300M	ARR, May 2026
20	Suno	$300M	ARR, Feb 2026

The positions reflect both commercial scale and the quality of the financial evidence. That is why a company with a larger headline figure can rank below one with a smaller but more directly comparable revenue metric.

Are Anthropic and OpenAI already in a league of their own?

Anthropic and OpenAI are operating at a completely different revenue scale from the other generative AI startups, with reported annualized revenue above $65 billion and $40 billion respectively.

Reuters reported that Anthropic's revenue run-rate increased from approximately $9 billion at the end of 2025 to $47 billion in May 2026 and more than $65 billion by the end of July. That represents more than sevenfold growth in roughly seven months. The acceleration has been closely linked to Claude's adoption by developers and enterprise customers, particularly for software engineering and increasingly complex workplace tasks.

OpenAI has also experienced substantial growth. Bloomberg reported that its annualized revenue exceeded $40 billion in August 2026, approximately double its $20 billion run-rate at the end of 2025. ChatGPT subscriptions, enterprise contracts, coding products and advertising have all contributed to the expansion. Subsequent October reporting placed OpenAI's run-rate closer to $50 billion, after some higher estimates had circulated.

The comparison becomes particularly striking when we look beyond the two companies. Cursor, ranked third, had a run-rate above $4 billion. Using those reported milestones, Anthropic's figure was more than 16 times Cursor's, while OpenAI's was more than 10 times higher.

These numbers annualize recent sales activity rather than representing audited full-year revenue. The lead is still extraordinary, though. The foundation-model business has produced two companies operating at a scale that most software startups take decades to reach.

Is Cursor already a multi-billion-dollar AI business?

Cursor had already crossed a $4 billion annualized revenue run-rate by early June 2026, making it one of the largest commercial AI application businesses in the world.

Dealroom reported that Cursor's run-rate increased from $2 billion in February to $3 billion in late April and more than $4 billion in early June. That is a doubling in approximately four months.

The customer mix is particularly interesting. Around 75% of Cursor's reported run-rate came from enterprise customers, equivalent to more than $3 billion in annualized business-to-business revenue at the June milestone. Its enterprise business reportedly tripled in Q1 2026 compared with Q4 2025.

AI coding tools have clearly moved beyond individual developers experimenting with productivity software. Companies are now paying substantial amounts to integrate AI directly into engineering workflows.

There is also an important ownership development. Cursor was acquired by SpaceX on August 14, 2026, in a transaction based on an implied equity value of $60 billion. Cursor remains at rank three because the acquisition falls within the period covered by our startup inclusion rule.

Are Chinese generative AI startups catching up with US rivals?

Chinese generative AI startups are already producing billion-dollar commercial metrics, with Z.ai, Moonshot AI and DeepSeek demonstrating that the market is no longer dominated by American companies alone.

Z.ai, also known as Zhipu AI, reported $1.8 billion in ARR during an investor call in September 2026. The company also increased its year-end ARR guidance to $3 billion, although that remains a target rather than achieved revenue.

Moonshot AI, the company behind Kimi, reported ARR above $1 billion in August, according to Bloomberg. Its annualized revenue had reportedly increased from $300 million in June, following the release of its Kimi K3 model in July.

DeepSeek was estimated to have reached a $400 million to $500 million revenue run-rate in 2026, primarily through paid API access to its models. Its commercial model is particularly interesting because the popular chatbot application remains free, while developers and businesses pay to use the underlying models.

SenseTime adds another dimension to the Chinese market. Its generative AI business generated CNY 2.33 billion in recognized segment revenue during the first half of 2026, according to its interim financial results. That is a six-month figure rather than ARR or an annualized estimate.

China now has sizable commercial businesses across foundation models, API access and enterprise AI. Their figures are not directly interchangeable, but the market is clearly more competitive than a simple OpenAI-versus-Anthropic comparison suggests.

Which AI coding startups are actually making money?

AI coding is already one of the strongest commercial categories in generative AI, with Cursor, Cognition, Replit and Lovable all reporting hundreds of millions or billions of dollars in annualized revenue.

Cursor's position is particularly striking, but Cognition has also grown quickly through its Devin AI engineering agent. The company reported a $900 million revenue run-rate in August 2026, up from $492 million in May.

Replit reached an estimated $525 million annualized revenue run-rate in April 2026, according to Sacra. Its business has expanded from a browser-based coding environment into an AI platform where users can create and deploy complete applications.

Lovable crossed $500 million in annualized revenue in June. Its product is aimed at people who want to build working applications through natural-language instructions, including users without traditional programming experience.

Base44, now part of Wix, also crossed $200 million ARR in August. Emergent reached a $120 million run-rate in July, according to its founder, while CodeRabbit reported ARR above $50 million in August.

The interesting part is how different these businesses are becoming. Cursor sells deeply into professional software teams, Devin automates engineering tasks, and Lovable and Replit increasingly compete for people who want software built without writing much code themselves.

Overall rank	Startup	Reported figure	Metric and period
3	Cursor / Anysphere	>$4B	Revenue Run-Rate, Early Jun 2026
8	Cognition	$900M	Revenue Run-Rate, Aug 2026
11	Replit	~$525M	Revenue Run-Rate, Apr 2026
13	Lovable	>$500M	Revenue Run-Rate, Jun 2026
27	Base44	>$200M	ARR, Aug 2026
31	Emergent	$120M	Revenue Run-Rate, Jul 2026
48	CodeRabbit	>$50M	ARR, Aug 2026

Are AI model infrastructure companies already billion-dollar businesses?

AI infrastructure has become a major revenue category in its own right, with Fireworks AI reporting more than $1 billion in annualized revenue and Together AI reporting more than $1.15 billion in annual bookings.

Fireworks runs infrastructure for serving AI models. In its July 2026 funding announcement, the company explicitly disclosed that it had surpassed $1 billion in annualized revenue run-rate while processing more than 40 trillion tokens per day.

Together AI operates in a similar part of the market, helping businesses train and run AI workloads using open models. Its July funding announcement disclosed annual bookings above $1.15 billion in the latest quarter.

Other companies are building substantial businesses around the infrastructure required to use generative AI. fal, which focuses on generative media infrastructure, has an estimated $400 million revenue run-rate. OpenRouter, which offers access to multiple AI models through a unified interface, has an estimated $160 million run-rate.

There is plenty of money to be made without building the underlying foundation model. Inference capacity, model routing, APIs and deployment tools have become valuable products in their own right, particularly as more companies use several models rather than committing to one provider.

Does Together AI make more than Fireworks AI?

Together AI reports a larger headline figure than Fireworks AI, but Fireworks ranks higher because its figure represents revenue run-rate while Together's represents bookings.

Together AI crossed $1.15 billion in annual bookings in the latest quarter of 2026. Fireworks AI had surpassed $1 billion in annualized revenue run-rate in July.

Bookings represent the value of customer commitments or contracts recorded during a period. They do not necessarily represent revenue already recognized or generated from completed services.

Fireworks therefore offers stronger evidence of current revenue generation, even though its headline number is slightly lower.

Neither number tells us how profitable the business is, and the comparison does not establish which company has more realized revenue. What we can say is that independent inference infrastructure has already become a billion-dollar commercial category.

Are enterprise AI startups becoming major businesses?

Enterprise generative AI has become a substantial commercial market, with Glean, Harvey, Cohere and Sierra all reporting ARR between $200 million and more than $400 million.

Glean reached $300 million ARR in May 2026, just 15 months after crossing $100 million. The company sells an enterprise search and knowledge platform that connects information across different workplace applications.

Harvey, which builds AI tools for legal professionals, crossed $400 million ARR in September. The company reported that more than 80% of the Am Law 100 law firms were using its platform.

Cohere reached $240 million ARR in 2025, according to an investor memo reported by TechCrunch. Its business focuses on enterprise foundation models and AI platforms, with an emphasis on private deployments and business-specific applications.

Sierra, meanwhile, had reached $200 million ARR by May 2026. Its conversational AI platform helps large companies automate customer-service interactions.

These businesses sell into budgets that already exist: legal services, customer support, knowledge management and enterprise software. They can charge for specific workflows and measurable results, without having to compete directly with the biggest foundation-model developers.

Are AI agents generating real revenue yet?

AI agents are already generating substantial commercial revenue, with Perplexity, Manus and several specialized agent platforms reporting hundreds of millions of dollars in annualized sales.

Perplexity had surpassed $750 million in annualized revenue by August 2026, according to a report by The Information cited by Reuters. Its growth has increasingly involved Perplexity Computer, a product designed to automate complex computer-based tasks for professionals.

Manus had reached an estimated $400 million to $500 million revenue run-rate in 2026, according to The Information. Its business is based on subscriptions to general-purpose AI agents capable of carrying out multi-step tasks.

There is an unusual ownership story here. Manus was acquired by Meta in December 2025, but Chinese authorities subsequently ordered the transaction to be unwound. Its original investors later bought back the shares, and the unwinding was completed in 2026.

Genspark, with an estimated $250 million run-rate, and Venice AI, with approximately $110 million, add further examples of consumer-facing AI search and agent businesses finding paying customers.

The shift is toward selling completed work: multi-step automation, professional information retrieval and tasks that would otherwise consume employee time. The revenue figures are becoming sizable, although the reliance on third-party estimates remains higher in parts of the agent market.

Is generative video already a big business?

Generative video has developed into a substantial commercial category, with Midjourney, HeyGen and Synthesia demonstrating that AI-generated creative products can reach hundreds of millions of dollars in revenue or ARR.

HeyGen crossed $200 million ARR in June 2026, doubling its ARR in approximately eight months. The company serves businesses and individuals creating AI-generated videos, avatars and multilingual content.

Midjourney has an estimated $300 million in annual revenue, although its figure relates to 2024 and is considerably older than HeyGen's disclosure.

Synthesia had an estimated $150 million ARR as of May 2026. Runway, which focuses on generative video tools for creators and professional production teams, had a revenue run-rate above $90 million in June 2025.

The smaller businesses are also worth following. Stability AI had estimated annual revenue of approximately $50 million in 2024, Mirage / Captions generated $28.4 million in trailing twelve-month in-app revenue for the period ending March 2026, and Luma AI had an estimated $8 million run-rate in December 2024.

Corporate video, AI avatars, advertising content and professional creative tools are already supporting real sales. But the figures come from different years and business models, making this a harder category to rank cleanly than the largest foundation-model companies.

Overall rank	Startup	Reported figure	Metric and period
22	Midjourney	~$300M	Annual Revenue, 2024
26	HeyGen	>$200M	ARR, Jun 2026
30	Synthesia	~$150M	ARR, May 2026
40	Runway	>$90M	Revenue Run-Rate, Jun 2025
54	Stability AI	~$50M	Annual Revenue, 2024
60	Mirage / Captions	$28.4M	TTM In-App Revenue, 365 days ending Mar 2026
67	Luma AI	~$8M	Revenue Run-Rate, Dec 2024

Are AI voice startups making serious money?

AI voice is already a substantial revenue category, led by ElevenLabs with more than $500 million ARR and supported by a growing group of companies selling voice agents and speech-generation infrastructure.

ElevenLabs disclosed that it ended 2025 with $350 million ARR and surpassed $500 million during the first four months of 2026. The company attributed much of its growth to enterprise deployments of voice agents in customer support, sales, recruitment and marketing.

Retell AI is another interesting example. Sacra estimates that the company reached an $80 million annualized revenue run-rate in August 2026, up from approximately $45 million at the end of 2025.

Parloa, which focuses on enterprise customer-service automation, has an estimated $52 million ARR. PolyAI reported £17.9 million in fiscal-year revenue for FY2026, based on its group accounts filed at Companies House.

Smaller companies are also generating meaningful revenue. Vapi reported $10 million ARR in June 2025, while Wispr Flow, Fireflies.ai and Synthflow AI have estimated annual revenues of approximately $10 million, $10.9 million and $1.1 million respectively.

Voice has two attractive markets at once: generating realistic speech and handling conversations that previously required human agents. ElevenLabs has pulled well ahead, but there is already a meaningful second group of companies selling to businesses.

Overall rank	Startup	Reported figure	Metric and period
12	ElevenLabs	>$500M	ARR, May 2026
43	Retell AI	~$80M	Revenue Run-Rate, Aug 2026
52	Parloa	~$52M	ARR, 2025
61	PolyAI	£17.9M	Fiscal-Year Revenue, FY2026
64	Vapi	$10M	ARR, Jun 2025
65	Wispr Flow	~$10M	Annual Revenue, 2025
66	Fireflies.ai	~$10.9M	Annual Revenue, 2024
73	Synthflow AI	~$1.1M	Annual Revenue, 2025

Is AI music already a $300M business?

Suno has already reached $300 million ARR, making generative music one of the more commercially developed creative AI categories.

In February 2026, Suno's co-founder and CEO Mikey Shulman disclosed that the company had reached $300 million ARR and approximately two million paying subscribers.

The progression is particularly interesting. Just three months earlier, Suno had told The Wall Street Journal that annual revenue had reached $200 million.

That is a remarkable amount of subscription revenue for a product built around generating songs. People are willing to pay for music creation even when the result is a track they may listen to only a few times.

Copyright remains the major complication. Licensing agreements with music publishers and record labels directly affect Suno's ability to operate and commercialize its models. Its settlement and licensing agreement with Warner Music Group is an important example of how the industry is trying to resolve that problem.

Are healthcare AI startups becoming major businesses?

Healthcare generative AI is already producing several substantial businesses, particularly in medical research, clinical documentation and physician workflow automation.

OpenEvidence reached an estimated $300 million revenue run-rate in July 2026, according to The Information. Its platform helps doctors research medical information and answer clinical questions, with pharmaceutical advertising forming an important part of its business model.

Abridge, which focuses on AI-powered clinical documentation, has an estimated $100 million ARR. Ambience Healthcare has an estimated $30 million ARR, while Freed has an estimated $19 million ARR.

These companies make money in quite different ways. OpenEvidence monetizes medical information and clinician attention, while Abridge and Ambience focus more directly on automating documentation and administrative work.

The strongest opportunities may be in activities that already consume substantial clinical time, such as documenting appointments and navigating medical records. Healthcare AI does not need to replace a doctor to become a large software business.

Are legal AI startups making serious money?

Legal AI is already one of the strongest specialized enterprise categories, with Harvey exceeding $400 million ARR and EvenUp reaching $100 million ARR.

Harvey has grown particularly quickly by selling AI tools for legal research, document review, drafting and increasingly complex legal workflows. In September 2026, the company reported more than $400 million ARR and adoption across over 80% of the Am Law 100.

EvenUp, which applies generative AI to personal injury law and legal case preparation, has an estimated $100 million ARR.

The commercial appeal is straightforward. Lawyers spend a lot of time reviewing documents, preparing arguments and handling large volumes of information. Firms already pay heavily for specialized software and information services, so a tool that meaningfully reduces this workload has an obvious customer.

Harvey's growth also demonstrates that a specialized AI application can become a major business without building its own frontier foundation model. Legal expertise, workflow integration and customer relationships can be valuable advantages of their own.

How many generative AI startups have already crossed $100 million?

At least 39 ranked generative AI startups have reached a reported commercial-scale figure of $100 million or more, although these figures include ARR, revenue run-rate, bookings and partial-period revenue.

The cohort extends from Anthropic at rank one through Moveworks at rank 39. The next company, Runway, has a reported revenue run-rate above $90 million.

What stands out is the variety. The $100 million group includes foundation-model developers, coding tools, AI infrastructure, enterprise search, video generation, healthcare applications, legal software and consumer AI products.

Some companies have achieved this scale through subscriptions, while others rely on enterprise contracts, consumption-based pricing or API usage.

The number needs to be interpreted carefully. Together AI's figure is bookings, MiniMax reports six months of recognized revenue, and several other companies report ARR or annualized revenue. Consequently, 39 companies with $100 million-scale metrics should not be confused with 39 businesses that have each recognized $100 million in annual sales.

Still, it is a sizable group. Generative AI now has a broad commercial market beneath its two biggest foundation-model companies, rather than a handful of isolated success stories.

Which generative AI startups report actual revenue rather than ARR?

Several generative AI startups now disclose recognized revenue in financial statements or interim results, offering a more direct view of achieved sales than annualized estimates.

SenseTime reported CNY 2.33 billion in generative AI segment revenue during H1 2026. This is a six-month figure covering its generative AI business rather than the company's entire revenue.

MiniMax reported approximately $117 million in total revenue for H1 2026, compared with $30.4 million during the corresponding period in 2025. Its open platform and other AI-based enterprise services generated $73.9 million of that amount.

PolyAI reported £17.9 million in fiscal-year revenue for the year ending January 2026, according to its Companies House group accounts.

Upstage reported KRW 24.8 billion in fiscal-year revenue for FY2025, while Sarvam AI reported INR 291 million for FY2025.

These conventional financial disclosures are particularly useful. They reveal sales recognized during a defined accounting period, although the business perimeter still varies from one company to another.

A six-month revenue figure, an annual fiscal-year figure and an ARR estimate answer different financial questions. Mixing them without explanation would make the ranking look more precise than it really is.

Are generative AI startups profitable yet?

Generative AI startups can reach impressive revenue figures while still losing substantial amounts of money, as MiniMax's financial results demonstrate.

MiniMax reported $116.6 million in recognized revenue during H1 2026, representing 283.1% year-over-year growth. Yet the company also reported an adjusted net loss of $293 million during the same six-month period.

Its gross profit was $20.8 million, equivalent to a gross margin of 17.9%. That is a relatively low margin for a software business, reflecting the substantial costs associated with running AI models and serving customers.

The company has been investing heavily in infrastructure, model development and inference capacity while expanding its commercial platform.

The economics are tough. Strong demand can coexist with expensive computing infrastructure, heavy model-development costs and weak margins. Revenue growth alone does not establish whether an AI business can eventually generate attractive profits.

MiniMax is a useful case because it publishes more detailed financial information than many private competitors. As other AI companies release income statements, we will get a clearer picture of how much of their rapid sales growth translates into sustainable earnings.

Can we trust the revenue numbers these AI startups report?

The strongest generative AI revenue figures generally come from company disclosures and reputable financial reporting, but the underlying metrics still require careful interpretation.

Looking at the 20 highest-ranked startups, ten figures are company-disclosed, seven come from credible reported disclosures, one comes from company financial results and two are third-party estimates.

That is a relatively strong evidence base for a market dominated by private companies. The bigger problem is what the figures actually measure.

Anthropic, OpenAI and Cursor report revenue run-rates. Z.ai, Moonshot and ElevenLabs report ARR. Together AI reports bookings, while SenseTime reports recognized revenue from a specific business segment.

Those distinctions can change the meaning of a billion-dollar headline. A company reporting $1.15 billion in bookings has not necessarily generated $1.15 billion in revenue. Likewise, a $1 billion ARR figure represents the annualized value of recurring revenue at a particular point, rather than $1 billion already recognized over a full financial year.

Third-party estimates also deserve less confidence than direct financial disclosures. Replit and fal, for example, have substantial estimated revenue figures, but their financial evidence is less direct than a company announcement or a regulatory filing.

Our ranking keeps these differences visible. The exact order becomes less certain when companies use different metrics, but the broader picture of commercial scale is much clearer than it was even a year ago.

Which promising generative AI startups still have no comparable revenue figure?

Several important generative AI companies remain outside the numerical ranking because their available commercial metrics do not establish a comparable revenue figure.

xAI / Grok, for example, had 117 million Grok monthly active users reported in March 2026. That demonstrates significant product adoption, but does not tell us how much revenue Grok independently generates.

Canva disclosed approximately 800 million AI-tool uses per month in 2026, but its overall revenue includes a much broader design business. We cannot attribute that entire business to generative AI.

Hugging Face has an estimated $150 million in ARR for its broader machine-learning platform, but that figure cannot be treated as pure generative AI revenue.

Other examples include Factory, which reported two-times month-over-month revenue growth for six months without an absolute revenue amount, and Cartesia, which has reported adoption by more than 50,000 companies.

Weights & Biases has also been reported to derive approximately 2% of its company ARR from generative AI-related products, while Granola has reported approximately 400% year-over-year growth.

Some of these businesses may already be making considerable money. But user adoption, revenue growth percentages and total company sales are not enough to establish how much revenue belongs specifically to generative AI.

What does the growth of generative AI startups tell us about where AI spending is going?

Generative AI spending is spreading across the entire technology stack, with foundation models still representing the largest individual businesses but applications increasingly generating substantial revenue of their own.

Seven of the top 20 ranked startups operate primarily in foundation models or closely related GenAI platforms. The remaining 13 cover coding, infrastructure, search, agents, voice, legal AI, enterprise knowledge and music generation.

Coding is particularly prominent, with Cursor, Cognition, Replit and Lovable all appearing in the top 20. Businesses are increasingly paying for AI systems that produce working software rather than just answer questions.

Infrastructure is another major opportunity. Fireworks and Together AI demonstrate that there is significant commercial demand for the computing and model-serving services required to deploy generative AI at scale.

Meanwhile, ElevenLabs, Harvey, Glean and Suno show that specialized applications can build substantial businesses around particular workflows and creative activities.

There is an interesting division of spending emerging. Model developers charge for access to AI capabilities, infrastructure companies charge to run and deliver those models, and application companies charge for what customers can accomplish with them.

The boundaries overlap, of course, especially as foundation-model developers move further into applications. But AI spending is no longer concentrated in subscriptions to general-purpose chatbots. It has become a much wider software and infrastructure market.

So, who is actually winning the generative AI revenue race today?

Anthropic is the top-ranked generative AI startup by revenue scale, followed by OpenAI and Cursor / Anysphere, with Z.ai, Fireworks AI, Together AI and Moonshot AI forming the next major group.

Anthropic's reported revenue run-rate above $65 billion puts it at the top of the ranking, while OpenAI's figure above $40 billion confirms that the two largest foundation-model businesses operate at a scale far beyond most competitors.

Cursor's position above $4 billion is arguably one of the most interesting developments because it shows how quickly a specialized AI application can become a multi-billion-dollar business. Its acquisition by SpaceX does not change its position in this startup ranking.

Below the top three, the market becomes more diversified. Chinese foundation-model developers have reached substantial commercial scale, AI infrastructure companies are generating billion-dollar metrics, and specialized applications in coding, voice, legal services, enterprise software and healthcare are producing hundreds of millions in annualized revenue.

There is no single winner across all those categories. Anthropic leads on our reported revenue-scale ranking, Cursor has become a giant in coding, Fireworks has built a major inference business, and companies such as Harvey and ElevenLabs are establishing strong positions in particular markets.

The clearest conclusion is that generative AI has already produced a sizable group of commercial businesses. The next test is whether these companies can sustain their growth while keeping inference costs, customer acquisition spending and model-development expenses under control.