Which AI startups are actually profitable?
In our AI infrastructure market deck, you will find everything you need to understand the market
SUMMARY
Which AI startups are actually profitable? The strongest cases are ElevenLabs, Gamma, Midjourney, OpenArt and Mercor, with Photoroom also likely profitable; Anthropic has crossed into adjusted quarterly operating profit but has not yet proved that level is durable.
The profitable group is much smaller than the ARR leaderboards suggest. Private AI companies report net income, adjusted operating profit, free cash flow, divisional profit and gross revenue almost interchangeably, so the headline number alone can be badly misleading.
ElevenLabs has the strongest reported absolute profit figure in the group, with Forbes putting 2025 profit at roughly $116 million. That is a much harder profitability signal than simply saying the company has reached break-even.
Gamma and Midjourney show two different ways to build profitable AI businesses. Gamma stayed small while selling a focused application on top of outside models; Midjourney financed its own specialized model development from customer revenue without conventional venture funding.
OpenArt is the most extreme operating-efficiency case. More than $70 million of ARR with roughly 20 employees suggests that owning the workflow and distribution layer can be extraordinarily valuable even when the company does not train every underlying model itself.
Mercor is profitable in cash terms, but its revenue needs a mental haircut before comparison with SaaS companies. Roughly 60% to 70% of its gross revenue reportedly flows to contractors, so the $2 billion-plus annualized figure overstates the revenue Mercor actually retains.
Freshness matters almost as much as the metric. Photoroom has repeated profitability evidence across several years, while Hebbia was explicitly profitable at a much smaller scale and has not disclosed an equally fresh profit figure after expanding.
Anthropic is the most important borderline case because it has shown positive adjusted operating income at frontier-model scale. The caveat is obvious: stock compensation is excluded, compute commitments remain enormous, and one quarter does not prove a stable earnings model.
xAI and Harvey are useful controls. Both have substantial revenue, yet xAI posted multi-billion-dollar operating losses and Harvey remains unprofitable, which makes it clear that fast ARR growth and actual profitability are still very different things.
The recurring pattern among profitable AI startups is disciplined scope: narrow products, small teams, clear customer value and less exposure to the full cost of the frontier-model race. Companies trying to own models, infrastructure, distribution and multiple product categories at once face a much harsher cost structure.
Falling inference costs should push more AI startups toward profitability, but those savings will not all stay with the companies. Competitors get cheaper models too, so the durable winners still need distribution, workflow depth, switching costs, proprietary data, brand or specialized technology beyond cheap access to intelligence.
This market map, featured in our AI infrastructure market deck, highlights top companies and startups in the AI infrastructure market
Which AI startups are actually profitable today?
A surprisingly small group of major AI startups can make a convincing profitability claim today: ElevenLabs, Gamma, Midjourney, OpenArt and Mercor have the strongest evidence, while Photoroom also belongs in the conversation and Anthropic has just crossed an important quarterly threshold.
The answer gets messy because private AI companies disclose whatever metric suits them. One company says it is profitable and means net income. Another means adjusted EBITDA. Another means free cash flow. A fourth may simply mean that one product line covers its direct costs. Those claims sound similar in a headline, but financially they can describe very different businesses.
We therefore looked for evidence that normal operations are already generating more cash or accounting profit than they consume. The strongest cases also need recent confirmation. A startup that was profitable two years ago can easily have returned to losses after doubling headcount or committing billions to computing capacity.
ElevenLabs has the clearest large-scale profit figure we found. Forbes reports that the voice-AI company earned about $116 million in profit in 2025. Gamma says it has been profitable since 2023 and crossed $100 million in ARR while still profitable. Forbes' latest profile of Midjourney describes the image-generation company as profitable after roughly $300 million of 2024 revenue, with no traditional outside funding.
OpenArt reached more than $70 million ARR with roughly 20 employees, and its CEO says the company was profitable when it raised its Series A. Mercor is free-cash-flow positive according to The Information, although its famous $2 billion annualized revenue figure is gross revenue before large contractor payouts.
Then we reach the borderline cases. Photoroom has repeatedly been described as profitable, though its latest detailed profit figure remains private. Hebbia was explicitly profitable when it had roughly $13 million ARR, but the business has grown several times larger since then without a fresh profitability disclosure. Anthropic reported positive adjusted operating income in its latest completed quarter, an extraordinary development for a frontier AI lab, but one profitable adjusted quarter gives us much less certainty than several years of positive company-level earnings.
At the other end, some of the biggest AI companies remain deeply loss-making. xAI's financial disclosures showed roughly $3.2 billion of 2025 revenue against about $6.4 billion of operating losses. Harvey has reached around $350 million ARR, yet cofounder Winston Weinberg confirmed recently that the legal-AI company is still unprofitable.
| AI startup | Profit status today | Evidence quality | Main caveat |
|---|---|---|---|
| ElevenLabs | Profitable | Very strong | Future expansion could compress margins |
| Gamma | Profitable | Very strong | Private-company reporting |
| Midjourney | Profitable | Very strong | Detailed recent profit amount undisclosed |
| OpenArt | Profitable | Strong | Founder/company disclosure |
| Mercor | Free-cash-flow positive | Strong | Gross revenue includes contractor payouts |
| Photoroom | Likely profitable | Good | Latest exact earnings undisclosed |
| Hebbia | Previously profitable | Moderate | No fresh profit confirmation |
| Anthropic | Positive adjusted operating income in latest quarter | Strong for the quarter | Stock compensation excluded; massive compute spending |
| Harvey | Unprofitable | Very strong | Still investing heavily |
| xAI | Heavily unprofitable | Very strong | Multi-billion-dollar operating losses |
If you want more recent data on this point, please see our latest AI infrastructure market report.
What does “profitable AI startup” actually mean?
For us, an AI startup is genuinely profitable when the whole business produces positive earnings or cash from normal operations; gross margins, profitable products and optimistic break-even forecasts do not clear that bar.
Private-company reporting makes this distinction unusually important.
Net income gives us the cleanest answer because compensation, infrastructure, R&D and the rest of the company's normal expenses ultimately flow through the accounts. ElevenLabs stands out because Forbes reported an actual profit figure of about $116 million for 2025 rather than merely quoting an ARR number.
Operating income comes next. If a company can pay for the staff, compute, sales, administration and model development required to run its business and still produce operating profit, the economics are clearly working at that moment. Anthropic's recent positive adjusted operating income therefore deserves to be taken seriously.
Free cash flow is similarly useful. Mercor reportedly generates positive free cash flow, which tells us the company can fund its ongoing operation from the cash it produces.
Adjusted EBITDA requires more caution. The word “adjusted” can hide stock compensation, restructuring expenses or other costs that shareholders eventually pay for. A positive adjusted figure can still be useful, especially when it marks a huge swing from losses, but we would never rank it alongside audited net income without explaining the difference.
The same caution applies when executives say a division is profitable. Scale AI, for example, has said its data business is profitable. That tells us something about that operation while leaving the consolidated company's overall profitability less certain.
| Metric | What it tells us | How we treat it |
|---|---|---|
| Net income | Whole company earned an accounting profit | Strongest evidence |
| Operating income | Core operations produced a profit | Strong evidence |
| Free cash flow | Business generated surplus cash | Strong evidence |
| Adjusted EBITDA | Profit before selected costs | Useful but conditional |
| Gross margin | Product revenue exceeds direct cost | Far from enough |
| Profitable division | One part of the company makes money | Does not prove company-wide profit |
As this chart shows, and as featured in our AI infrastructure market deck, search interest in AI infrastructure has risen sharply
How can an AI startup make billions in revenue and still lose money?
AI companies can lose billions while revenue explodes because model training, inference, infrastructure and hiring can grow almost as quickly as sales, and sometimes faster.
xAI gives us the cleanest example.
Financial disclosures covering its AI operations showed approximately $3.2 billion of revenue in 2025. Costs and expenses reached about $9.6 billion, producing an operating loss of roughly $6.4 billion. Research and development alone exceeded $5 billion.
Put differently, xAI lost almost $2 in operations for every $1 of revenue it generated that year.
Harvey shows the same problem at a much smaller scale. The legal-AI startup has reached around $350 million ARR and serves more than 200,000 lawyers, according to a recent interview with cofounder Winston Weinberg. Weinberg still says Harvey is unprofitable. The company now employs roughly 1,200 people as it pushes into more countries, customers and legal workflows.
The numbers explain why ARR rankings can be misleading these days. A software company growing from $100 million to $300 million ARR can simultaneously become less profitable if its workforce, inference bill and sales organization expand even faster.
Frontier models make this especially brutal because computing expenses arrive in two places. The company must pay to serve today's customers, then spend heavily again to train the models needed to stay competitive.
If you want more recent data on this point, please see our latest AI infrastructure market report.
How did Gamma reach $100 million ARR and stay profitable?
Gamma has built one of the cleanest profitable AI application businesses we found because it reached $100 million ARR with a tiny team, modest previous funding and no need to finance its own frontier model.
Gamma's own account says the company has been profitable since 2023. When it crossed $100 million ARR, it had raised only $23 million before its latest financing and employed around 50 people. Forbes later listed 75 employees, which still leaves the company at well above $1 million of recurring revenue per employee.
Gamma's early history makes the outcome more interesting. By late 2022, the original product had about 60,000 users and a little more than one year of runway remaining. The team rebuilt Gamma around generative AI, introduced paid usage once customers started exhausting free AI credits and reached positive economics without building the massive organization normally associated with a $100 million software business.
The company now says more than 100 million people have tried Gamma, while hundreds of thousands pay for the service. Its latest $68 million financing valued the company at $2.1 billion, but Gamma raised that money after reaching profitability. Part of the round even provided liquidity to earlier employees.
Gamma sells an expensive-looking outcome — finished presentations, documents and websites — while outsourcing much of the underlying intelligence to model providers. Its own costs are concentrated in product development, distribution, inference and a relatively small workforce.
The company has also resisted hiring simply because its valuation and revenue would allow it. Founder Grant Lee has explicitly argued that Gamma could have employed roughly 200 people rather than 50.
That staffing discipline is one of the clearest patterns across profitable AI startups. OpenArt reached roughly $70 million ARR with about 20 people, while Midjourney generated hundreds of millions in annual revenue with a team dramatically smaller than a conventional software company at the same scale. Harvey, by comparison, has roughly 1,200 employees at around $350 million ARR and remains unprofitable.
A second pattern is product focus. Gamma helps people make presentations, documents and websites. Midjourney generates creative imagery. ElevenLabs handles voices and conversational audio. Photoroom handles product photography. OpenArt gives creators one place to use many visual models. Mercor connects AI companies with skilled human experts.
These companies are selling narrow products with a clear reason to pay, and most have avoided the huge fixed cost base that comes with trying to own every part of the AI stack.
This chart, included in our AI infrastructure market deck, shows annual VC investment in AI infrastructure startups
Is Midjourney the clearest profitable AI model startup?
Midjourney is the cleanest proof that a company can train its own major generative models and still make money, because it reached hundreds of millions in revenue without conventional venture funding.
Forbes' latest company profile says Midjourney generated around $300 million of revenue in fiscal 2024, based on PitchBook data, and remains profitable. The same profile lists roughly 60 employees and confirms that Midjourney has avoided outside financing.
More recent industry estimates put its revenue higher, though we would rather anchor our conclusion to the better-supported $300 million figure than pretend a newer estimate is audited fact.
Midjourney's capital history is what gives us confidence in the profitability claim. A startup can exaggerate EBITDA adjustments for a while. It cannot easily operate for years, employ dozens of people, train expensive image and video models and serve millions of users without outside funding unless customers are paying enough to finance the operation.
Midjourney also benefited from an unusual distribution model. Early growth happened largely inside Discord, which saved the startup from having to build a full social and distribution layer at the beginning. Customers paid subscriptions directly. The company did not need hundreds of enterprise account executives to create demand.
Its technical ambition was also narrower than the mission pursued by OpenAI, Anthropic or xAI. Midjourney focused first on image generation and later expanded into video. Training those models is expensive, but the company does not need to finance every frontier simultaneously across coding, reasoning, voice, search, autonomous agents, scientific work and general multimodal intelligence.
Midjourney gives us a useful counterexample to the idea that proprietary model development is inherently unprofitable. Specialized models can already produce strong economics.
If you want more recent data on this point, please see our latest AI infrastructure market report.
Is ElevenLabs the most profitable major generative-AI startup?
ElevenLabs has the strongest reported profit figure among the large private generative-AI startups we examined: Forbes says the company made roughly $116 million in profit in 2025.
That puts ElevenLabs in a different league from startups that simply say they have reached break-even.
The voice-AI company began with speech generation, voice cloning and dubbing, then expanded into conversational agents and enterprise use cases. Forbes' latest profile says customer-service voice agents now generate more than half the business.
That enterprise shift is valuable because synthetic voice can replace or automate an existing labor-intensive expense. A company already paying people to handle thousands of customer calls has a clear budget against which ElevenLabs can price its technology.
The reported profit also appears large enough to survive normal measurement noise. We are talking about more than $100 million of annual profit rather than a startup briefly crossing zero after cutting a few discretionary expenses.
ElevenLabs' headcount has grown substantially and Forbes now lists about 450 employees. Even so, the company has created a much larger amount of revenue and profit per worker than a traditional software vendor would normally produce at this stage.
There is more risk ahead than Gamma faces. ElevenLabs is moving beyond voice into music, video and broader generative-media tools while investing heavily in infrastructure. Every additional modality pushes the company closer to the expensive model race.
As of now, ElevenLabs probably deserves the top spot if “most profitable” means the clearest large absolute profit among prominent independent generative-AI startups.
This chart, included in our AI infrastructure market deck, shows why CoreWeave is winning in AI infrastructure
Are OpenArt and Mercor really profitable, or do their revenue numbers flatter them?
OpenArt and Mercor both appear genuinely profitable today, but their headline revenue figures need very different interpretations before we compare them with software companies such as Gamma.
OpenArt is the simpler case. CEO Coco Mao said the company reached more than $70 million ARR with roughly 20 employees, grew revenue about sevenfold during 2025 and was profitable when it announced a $30 million Series A. Canaan, which led the round, separately confirmed the roughly $70 million ARR and 20-person scale.
That works out to around $3.5 million of ARR per employee.
OpenArt can remain tiny partly because it aggregates AI models rather than trying to train the best model in every creative category itself. Users can access image and video systems from multiple providers through one workflow. OpenArt captures value in the interface, creation tools, distribution and user experience while letting other companies absorb much of the frontier research bill.
Mercor has completely different economics.
The company reportedly surpassed a $2 billion gross annualized revenue run rate in June, according to The Information, after generating $614 million of gross revenue during the first half of the year.
Yet Mercor passes roughly 60% to 70% of its topline revenue to the contractors performing work for customers. On a $2 billion gross run rate, we estimate that Mercor itself keeps somewhere around $600 million to $800 million before its own operating costs.
That is still a very large business. We simply should not compare $2 billion of Mercor gross revenue with $2 billion of high-margin SaaS ARR as though they were equivalent.
The more important piece of information is that The Information reports Mercor is free-cash-flow positive. Sacra also reports that the company produced about $6 million of profit during the first half of 2025 before its latest acceleration.
OpenArt gets to profitability through extreme software efficiency. Mercor gets there with a variable-cost marketplace whose contractor expenses move with customer demand.
If you want more recent data on this point, please see our latest AI infrastructure market report.
Do Photoroom and Hebbia still belong on the profitable AI startup list?
Photoroom still deserves to be called a profitable AI startup, while Hebbia should now be described more carefully as a company with confirmed historical profitability and unclear current profit status.
Photoroom has one of the longer profitability records in European generative AI. Sifted reported that the Paris-based image-editing company was profitable while annual recurring revenue was around €50 million, and later coverage continued to describe it as one of the rare profitable GenAI companies as the organization expanded.
The product's economics make sense. Photoroom automates commercial photography and image editing for merchants, marketplaces and online sellers. Those customers already spend money creating product imagery, so the software replaces a recognizable cost rather than asking businesses to invent a new AI budget.
Photoroom has also spent years reducing inference costs. For a tool generating and editing large quantities of images, a small reduction in the cost of every operation compounds rapidly across millions of users.
We have slightly less confidence in its current margin because the company has grown its team and model ambitions without publishing a fresh annual net-income figure.
Hebbia needs an even bigger caveat.
TechCrunch reported that Hebbia was profitable around the time it reached approximately $13 million ARR and raised its $130 million Series B. Since then, the enterprise research platform has grown dramatically. Sacra now estimates about $48 million ARR, up from roughly $30 million at the end of 2025 and around $17 million at the end of 2024.
That trajectory shows the product is working commercially, but we have no equally recent confirmation that Hebbia stayed profitable while revenue, hiring and expansion accelerated.
So our current classification is deliberately asymmetric: Photoroom remains likely profitable because profitability has been reported repeatedly through a longer operating period. Hebbia has proven that its model can reach profitability, but we would want a newer financial disclosure before putting it in the same confidence tier as Gamma or ElevenLabs.
This chart, included in our AI infrastructure market deck, shows annual funding in AI infrastructure startups
Has Anthropic finally made frontier AI profitable?
Anthropic has shown that frontier AI can produce positive adjusted operating income for a quarter, which is the strongest evidence yet that the economics of a major general-purpose model lab can eventually work.
Recent reporting says Anthropic's latest completed quarter generated more than $11.5 billion of revenue and positive adjusted operating income. Earlier investor forecasts had pointed to roughly $10.9 billion of quarterly revenue and $559 million of adjusted operating profit, so the eventual revenue performance came in even stronger.
The company has kept growing since then. Its annualized revenue run rate exceeded $65 billion by the end of July, according to Reuters, up from about $47 billion in May and roughly $9 billion at the end of 2025.
That means Anthropic increased its annualized revenue pace by more than seven times in around seven months.
Claude Code appears to be a major part of that acceleration, particularly among developers and enterprise customers. Anthropic has become much more commercially concentrated on coding and professional usage than the broad consumer footprint associated with ChatGPT, which may help explain why high-value workloads are scaling so quickly.
The profitability metric is meaningful because the earlier $559 million forecast included the cost of training new models. Anthropic was therefore expected to cover one of the largest expenses that critics often exclude when discussing AI economics.
Stock-based compensation remained excluded, however, and the company's future compute commitments are enormous.
Reuters has reported that Anthropic agreed to pay SpaceX roughly $1.25 billion per month for computing capacity under one arrangement. It also has major infrastructure agreements involving Amazon, Google and other providers. Recent reporting puts Anthropic's broader infrastructure commitments comfortably into the tens of billions of dollars.
Its latest gross-margin estimates also remain far below mature software levels. Reuters Breakingviews cited a PitchBook estimate of roughly 44%, illustrating how much revenue still flows straight back into computing and infrastructure.
We are willing to say something much stronger about Anthropic today than we could earlier: the company has demonstrated that a frontier lab can cross into adjusted operating profitability while still growing explosively.
Several more quarters through another major training cycle would tell us whether that profit is durable.
If you want more recent data on this point, please see our latest AI infrastructure market report.
Could falling AI inference prices make many more startups profitable?
Cheaper inference should push many AI application startups toward profitability, but it will not automatically create good businesses because competitors receive nearly the same cost reduction.
Model providers repeatedly reduce API prices or deliver more capability for the same dollar. Open-weight models also give startups alternatives to the leading commercial APIs. Hardware improves. Quantization, caching, routing and smaller specialized models reduce the amount of expensive frontier inference required per customer request.
For an application charging a fixed subscription, that can create enormous operating leverage.
Imagine an AI service that collects $20 from a user and originally spends $8 on inference. If optimization eventually reduces that inference bill to $3 while the subscription price remains unchanged, five additional dollars flow toward gross profit before other costs.
Gamma and Photoroom have already demonstrated how aggressively infrastructure optimization can matter.
But competitors get cheaper models too.
If producing an AI presentation, image, voice or website becomes dramatically cheaper, rivals can lower prices. OpenAI or Google can bundle equivalent functionality into existing subscriptions. New startups can enter with almost no model-development budget.
The companies most likely to keep the savings are those with something beyond cheap model access: strong distribution, proprietary workflow, brand, customer data, switching costs or specialized models.
This chart, included in our AI infrastructure market deck, compares the main business model options for AI cloud infrastructure providers
So which AI startups are actually profitable?
There are genuinely profitable AI startups today, and our strongest picks are ElevenLabs, Gamma, Midjourney, OpenArt and Mercor; Photoroom probably belongs with them, while Anthropic has reached profitability on an adjusted quarterly basis but still needs to prove it can stay there.
ElevenLabs has the strongest reported absolute profit figure, with Forbes putting 2025 profit at roughly $116 million.
Gamma gives us the cleanest example of a profitable AI application company. It crossed $100 million ARR after being profitable for years and did so with remarkably little capital and headcount.
Midjourney is the strongest self-funded model-company case. Around $300 million of 2024 revenue, continued profitability and no traditional venture financing make the underlying economics difficult to dismiss.
OpenArt is smaller but arguably even more extreme operationally: more than $70 million ARR, roughly 20 employees and a company-confirmed profit.
Mercor is also generating cash, although readers need to adjust mentally for its accounting. Its $2 billion-plus annualized figure represents gross customer spend; 60% to 70% reportedly flows to contractors. Free-cash-flow profitability is still real profitability.
Photoroom has enough repeated evidence for us to keep it in the profitable group, though we would like a fresher exact earnings number. Hebbia gets a more cautious label because its last clear profitability confirmation came when the business was much smaller.
Anthropic is the company to watch. Positive adjusted operating income in its latest quarter, more than $11 billion of quarterly revenue and a subsequent $65 billion-plus annualized revenue run rate have moved frontier-AI profitability from theory to something we can observe. Its huge compute obligations keep us from declaring victory after one quarter.
Meanwhile, xAI's roughly $6.4 billion 2025 operating loss and Harvey's continued losses at $350 million ARR show how little headline revenue tells us on its own.
| AI startup | Our verdict | Why we believe it | Confidence |
|---|---|---|---|
| ElevenLabs | Clearly profitable | ~$116M reported 2025 profit | Very high |
| Gamma | Clearly profitable | Profitable since 2023; $100M+ ARR | Very high |
| Midjourney | Clearly profitable | ~$300M 2024 revenue, profitable, self-funded | Very high |
| OpenArt | Clearly profitable | $70M+ ARR, ~20 staff, company says profitable | High |
| Mercor | Cash-flow profitable | Positive free cash flow despite contractor pass-through | High |
| Photoroom | Likely profitable | Repeated profitability evidence over several years | Medium-high |
| Anthropic | Quarterly adjusted profit | Latest completed quarter positive on adjusted operating basis | Medium-high |
| Hebbia | Previously profitable | Confirmed profit at smaller scale, no fresh update | Medium |
| Harvey | Unprofitable | Cofounder recently confirmed it | Very high |
| xAI | Heavily unprofitable | ~$6.4B 2025 operating loss | Very high |
OUR METHODOLOGY
We treated the question of which AI startups are actually profitable as a structured evidence-aggregation problem rather than a ranking of headline ARR. Private companies disclose different financial measures at different moments in their growth, so the same word — “profitable” — can refer to net income, operating income, free cash flow, adjusted EBITDA or even the economics of a single division.
For each company, we collected the freshest available evidence across several dimensions: the quality and scope of the profit measure, how recent the disclosure was, whether profitability appeared sustained or temporary, the underlying business model, the share of headline revenue the company actually retains, and the strength of the source behind the claim.
We gave the most weight to company disclosures, regulatory filings and high-quality financial reporting. Net income, operating income and free cash flow rank above gross margin or divisional profitability for this purpose, while adjusted metrics are treated more cautiously when they exclude costs such as stock-based compensation.
We also normalized cases where the headline figures were not economically equivalent. Mercor's gross marketplace revenue is adjusted conceptually for the large contractor pass-through; Anthropic's positive adjusted operating result is treated as evidence for one quarter rather than proof of sustained company-level earnings; and Hebbia's older profitability confirmation carries less weight because the company has expanded substantially since then.
The final classifications therefore reflect the combined weight, freshness, consistency and comparability of the evidence. Repeated profitability across several periods strengthens a case; historical profitability loses weight when the business has changed materially; and clearly loss-making companies such as xAI and Harvey provide a useful control against the idea that rapid revenue growth automatically translates into profit.
Key sources include Forbes on ElevenLabs and Forbes' deeper ElevenLabs profile; Gamma's first-hand account of reaching $100 million ARR profitably; Forbes on Midjourney; OpenArt CEO Coco Mao's Series A disclosure and Canaan's account of the round; The Information on Mercor's gross annualized revenue, contractor pass-through and free cash flow; Sifted on Photoroom's profitability, Sifted's later Photoroom coverage, and Photoroom on inference efficiency; and TechCrunch on Hebbia's earlier profitability.
For Anthropic, we used Reuters Breakingviews on its projected first positive adjusted operating quarter, Anthropic's Series H disclosure, Reuters Breakingviews on its later revenue run rate and gross-margin estimate, plus Anthropic's first-hand disclosures on its Amazon compute expansion, Google and Broadcom compute expansion, and SpaceX compute agreement. For the loss-making controls, we used SpaceX's SEC-filed S-1 for xAI's financial figures and The Times interview with Harvey cofounder Winston Weinberg.
This chart, featured in our AI infrastructure market deck, shows the share of revenue generated by each customer segment in the AI infrastructure market
Related blog posts
- Which AI startups are growing without burning billions?
- Which AI startups have the most durable revenue?
- Which AI infrastructure startups generate the most revenue today?
