Is Cognition really worth $48B today?

Did Cognition really reach a $48 billion valuation?

Yes, Cognition reached a $48 billion valuation following a financing round of more than $2 billion, making the company behind Devin one of the world's most valuable AI software startups.

The Series E financing was announced on September 8, 2026, led by Andreessen Horowitz and Accel, alongside existing investors including Founders Fund, General Catalyst, and Avenir. Cognition had been valued at $26 billion just over three months earlier and $10.2 billion in September 2025. Its valuation increased roughly 4.7 times in one year.

The latest round was particularly striking because Cognition attracted almost $10 billion in investor interest, according to Bloomberg's reporting ahead of the transaction. Investors were competing to participate, even after the company's valuation had already increased dramatically. Rounds this large show where investor money is going, which we follow closely in our report on AI coding assistants.

The $48 billion figure represents the valuation agreed during a private financing round. It does not establish what every shareholder could receive in a sale, especially if the new shares include preferential rights.

Has Cognition's valuation grown faster than its revenue?

Cognition's valuation initially rose almost exactly as fast as its reported revenue between May and early September 2026. Its subsequent revenue milestone makes the comparison even more interesting: revenue is now growing faster than the valuation implied by those financing rounds.

In May 2026, Cognition was valued at $26 billion with $492 million in annualized revenue. By its September 8 financing announcement, those figures had reached $48 billion and almost $900 million.

That represented an 84.6% valuation increase and approximately 82.9% revenue growth in just over three months. The implied revenue multiple barely moved, from 52.8 to 53.3 times annualized revenue.

Then, on September 25, Cognition confirmed it had crossed $1 billion in annualized revenue. Assuming its last financing valuation remained the reference point, the multiple fell to 48 times. Compared with May, annualized revenue had more than doubled, while valuation increased by around 85%.

Looking further back, Devin reached approximately $73 million in annual recurring revenue in 2025, before Cognition acquired Windsurf and its reported $82 million revenue base. The combined business subsequently grew to $1 billion in annualized revenue.

Those figures include the acquisition, so the entire increase cannot be attributed to Devin's organic growth. Even allowing for Windsurf, though, the expansion has been exceptional.

Funding milestone	Valuation	Annualized revenue	Revenue multiple
September 2025	$10.2B	Not publicly comparable	N/A
May 2026	$26B	$492M	52.8x
September 8, 2026	$48B	~$900M	53.3x
September 25, 2026	$48B reference	$1B	48.0x
Change since May	+84.6%	+103.3%	-4.8x
At a hypothetical 40x multiple	$40B	$1B	40x
At a hypothetical 30x multiple	$30B	$1B	30x
At a hypothetical 20x multiple	$20B	$1B	20x

How much revenue is Cognition actually generating today?

Cognition has crossed $1 billion in annualized revenue, equivalent to approximately $83 million per month if measured using a monthly revenue run rate. That is an enormous commercial milestone for a company founded in 2024.

The Information reported that Cognition's annualized revenue had more than tripled since the beginning of 2026. The company subsequently reported almost $900 million when announcing its September financing, before confirming the $1 billion milestone on September 25.

The pace of expansion is especially unusual. Cognition added approximately $508 million to its annualized revenue base between May and late September. On a simple monthly equivalent, that represents roughly $42 million more revenue than its May pace.

But the distinction between revenue and revenue run rate is important here. The $1 billion figure extrapolates recent sales activity into a full year. It does not mean Cognition has already collected or recognized $1 billion over the preceding 12 months.

Cognition also has not released audited financial statements detailing its historical revenue, customer retention, or the split between Devin and Windsurf. Its commercial momentum is clear; the underlying revenue quality is much harder to evaluate. Without audited numbers from private companies, we build our own market size from the ground up in our AI coding assistant market report.

Did buying Windsurf make Cognition a much bigger company?

The Windsurf acquisition transformed Cognition's business almost overnight, giving the company an AI code editor, an established customer base, and a much larger opportunity to sell Devin.

When Cognition acquired Windsurf in July 2025, Devin had approximately $73 million in annual recurring revenue, while Windsurf brought around $82 million. Together, the two businesses represented roughly $155 million before subsequent growth.

Windsurf also came with more than 350 enterprise customers and hundreds of thousands of daily active users, according to Cognition's acquisition announcement.

The timing was unusual. Google had just recruited key Windsurf leaders through a separate licensing and talent agreement, leaving Cognition to acquire the remaining product, intellectual property, brand, and business.

The combination addressed a weakness in Cognition's original positioning. Devin was designed to execute engineering tasks autonomously, while Windsurf operated directly inside developers' everyday workflows.

With both products, Cognition could reach customers who wanted assistance while writing code and teams looking to delegate entire engineering assignments.

This also makes Cognition's later growth more interesting. Its approximately $155 million combined starting revenue base expanded to $492 million by May 2026, nearly $900 million by early September, and $1 billion later that month.

The acquisition alone cannot explain that increase. However, Cognition has not disclosed enough product-level financial information to determine precisely how much came from Devin, Windsurf, or customers using both.

Is Devin's growth coming from real customers or just AI hype?

Devin's growth is backed by substantial enterprise usage, including deployments at major financial institutions, manufacturers, and technology companies. Cognition has moved well beyond selling an experimental coding assistant to early adopters.

Its customer roster includes NVIDIA, GE Aerospace, Citi, Goldman Sachs, Mercedes-Benz, BNY, and Santander.

These customers are particularly interesting because software engineering inside large banks and industrial companies is difficult to automate. Their systems include legacy applications, strict security requirements, internal dependencies, and complicated approval processes.

Cognition's November 2025 performance review reported that Devin had helped produce hundreds of thousands of merged pull requests across customer environments. A merged pull request is a more useful measure of adoption than simply counting trial accounts, even if the size and complexity of those contributions vary enormously.

The company also reportedly had approximately 50 customers on track to spend more than $1 million annually, according to The Information's August 2026 reporting. That represents at least $50 million in potential annual customer spending from this group alone.

Customer concentration is worth watching, though. The same investigation found that one unnamed customer represented approximately 10% of Cognition's business.

At a $1 billion annualized revenue level, that proportion would correspond to around $100 million from a single customer, assuming the concentration remained unchanged. It may not have.

Large enterprise contracts can produce rapid growth, but they also create exposure when a major customer changes its budget or switches providers. Cognition has yet to disclose how consistently customers renew, expand their spending, and use Devin across their engineering organizations.

Can Devin actually replace software engineers?

Devin can complete meaningful software-engineering tasks independently, but the evidence still falls short of showing that Cognition can replace experienced engineers across their full range of responsibilities.

When Devin launched in March 2024, Cognition reported solving 13.86% of issues in its SWE-bench evaluation, compared with a previous unassisted baseline of 1.96%. The result attracted enormous attention because the benchmark involved real software repositories.

The product has improved substantially since then. In November 2025, Cognition reported that the share of Devin-generated pull requests successfully merged had increased from 34% to 67%. Merged pull requests are a good test of what agents really do, and we sort proven uses from promises in our market report about AI coding assistants.

Its own engineers were also using the agent more frequently. In February 2026, Cognition reported merging 659 Devin-generated pull requests in one week, compared with a previous weekly record of 154. That represents a 4.3-fold increase.

Still, a pull request can range from a small dependency update to a complicated architectural modification. Counting merged contributions does not tell us how much engineering judgment or human review was required.

Devin performs particularly well on repetitive tasks such as migrations, dependency upgrades, bug investigations, and vulnerability fixes. Product architecture, ambiguous requirements, and decisions involving multiple teams remain much harder to delegate completely.

Our view is that Devin can reduce the amount of routine engineering labor companies need, sometimes dramatically. Replacing the person responsible for deciding what to build, reviewing the result, and managing the consequences is a different challenge altogether.

Is Cognition's own AI model actually better than Claude or GPT?

Cognition's SWE-1.7 model is competitive with leading AI coding models, but the company's own published benchmarks show that Claude Opus 4.8 and GPT-5.5 still outperform it on important tests.

Cognition released SWE-1.7 in July 2026, building on Kimi K2.7 and using additional reinforcement learning designed for long-running software-engineering tasks.

On FrontierCode 1.1 Main, SWE-1.7 achieved 42.3%, compared with 46.5% for Claude Opus 4.8 and 43.0% for GPT-5.5.

On Terminal-Bench 2.1, SWE-1.7 reached 81.5%, compared with 86.9% for Opus 4.8 and 84.2% for GPT-5.5.

The improvement over Cognition's previous model was nevertheless enormous. SWE-1.6 had achieved just 9.4% on FrontierCode and 39.7% on Terminal-Bench.

These are Cognition-published benchmark results, so they should be interpreted within the disclosed testing setup rather than as a universal ranking of coding products.

The company's achievement is particularly relevant to its economics. Cognition is training specialized models that can handle substantial engineering workloads while potentially costing less than relying entirely on third-party frontier models.

It doesn't necessarily need to build the world's best general-purpose model. It needs models good enough for the work its customers actually delegate, at a cost that allows the company to make money.

Is Cognition already falling behind Claude Code, Codex, and Cursor?

Cognition remains a fast-growing competitor, but Claude Code, OpenAI Codex, and Cursor make the AI coding market much more dangerous than its recent funding rounds might suggest.

Anthropic said Claude Code had surpassed $2.5 billion in annualized revenue by February 2026. That is more than twice Cognition's latest $1 billion run rate, even though Cognition has expanded rapidly since the beginning of the year.

Cursor has grown even faster in absolute revenue terms. According to Sacra, Cursor reached approximately $4 billion in annualized revenue in May 2026, up from roughly $1.2 billion at the end of 2025. The Information also reported an approximately $4 billion run rate in August.

Cursor was therefore generating around four times Cognition's latest disclosed annualized revenue.

The competitive environment changed again when SpaceX completed its $60 billion acquisition of Cursor in August 2026. Cursor subsequently emphasized access to SpaceX's large computing infrastructure, providing resources to train models and lower operating costs.

OpenAI is pursuing a similar opportunity through Codex, while Microsoft can distribute GitHub Copilot to organizations already using GitHub and its enterprise software.

Cognition's growth shows that these competitors have not prevented it from attracting customers. But the difference in scale is hard to ignore, and the biggest AI model providers have obvious distribution and infrastructure advantages.

There is also a broader concern. The Information calculated that Anthropic and OpenAI accounted for approximately 89% of the combined application and model-sales revenue among the AI-native companies it examined, up 4.5 percentage points over nine months.

That calculation covers a broader market than coding alone. Still, it illustrates how quickly the largest AI companies can capture spending.

Cognition's best defense is probably not winning a general coding-assistant popularity contest. It is becoming indispensable for expensive engineering work that enterprises cannot easily hand over to a cheaper alternative.

Are companies saving real money by using Devin?

Devin has demonstrated substantial cost and time savings in enterprise modernization projects, although Cognition has not published enough independent customer data to establish how typical those results are.

In one case discussed in Cognition's November 2025 performance review, a bank completed certain data-migration tasks in three to four hours using Devin, compared with 30 to 40 hours for human engineers.

Another customer reportedly completed Java repository migrations 14 times faster.

The strongest examples involve outdated enterprise systems. In a September 2026 AWS announcement, Mercedes-Benz described using Devin to analyze more than 200,000 lines of COBOL, reducing an estimated eight-month modernization project to eight days.

Another automotive company reportedly migrated a 25,000-line COBOL workflow to AWS Lambda at an estimated 73% lower cost.

These results explain why Cognition has increasingly focused on enterprise modernization. Companies spend heavily maintaining and updating old applications, and successful automation can save much more than a software subscription costs.

However, these examples are largely vendor-reported case studies. Cognition has not released a representative breakdown showing average savings, failed projects, customer retention, or engineering hours saved across its full customer base.

Can Cognition make money while running such expensive AI models?

Cognition's profitability remains one of the weakest parts of its investment case. The company's enterprise gross margin was reportedly close to 50% in August 2026, far below the margins traditionally associated with successful software businesses.

The Information reported that Cognition's gross margin excludes the cost of training its proprietary models. Those research expenses therefore sit outside the direct costs used to calculate the reported margin.

A 50% gross margin means that approximately half of the relevant revenue remains after direct service-delivery costs. At a hypothetical $1 billion revenue level, applying that margin across the whole business would leave around $500 million in gross profit before research, sales, administration, and other expenses. Margins after model costs are one way we compare business models in our AI coding assistant report.

That calculation is illustrative because Cognition has not disclosed a consolidated gross margin or the precise share of revenue coming from its enterprise business.

The comparison with conventional software companies is uncomfortable. Established software businesses often aim for gross margins above 70%, giving them more room to fund research and sales while remaining profitable.

AI companies face an additional expense: every complex customer request consumes computing capacity. The more work their agents perform, the more infrastructure they need.

Cognition has responded by investing in its own models and infrastructure rather than relying completely on third-party providers. The company believes this can improve costs over time, and The Information reported that it has already become more efficient at serving models from its computing cluster.

Its April 2026 pricing changes also made the products accessible through lower-cost subscriptions, with Pro starting at $20 per month and higher tiers available for teams and enterprises.

That broadens adoption, but cheap entry-level subscriptions alone will not support a $48 billion valuation. Cognition needs enterprise workloads that generate substantial revenue without consuming most of that revenue in computing costs.

Is Cognition burning too much cash?

Cognition could burn approximately $800 million in cash during 2026, an extraordinary amount even for a company that has crossed $1 billion in annualized revenue.

According to The Information's August investigation, Cognition burned roughly $200 million during the second quarter alone. Its annual cash consumption could approach $800 million because of the high cost of leasing NVIDIA computing infrastructure.

The company's server cluster reportedly costs hundreds of millions of dollars annually and supports both customer inference and proprietary model training.

This represents a dramatic change from Cognition's earlier approach. In September 2025, management said cumulative net burn since inception had remained below $20 million.

One year later, the company was investing at an entirely different scale.

There is a financial distinction worth making. Cognition pays for computing capacity that supports future development, and some expenses relate to training models that may eventually reduce the cost of serving customers.

The Information reported that the company would be close to free-cash-flow breakeven if spending on developing its own models were excluded.

With more than $2 billion raised in its latest financing, Cognition has significant resources available. Even so, its operating costs create a much tougher financial profile than its revenue growth alone suggests.

We think the spending is defensible only if it produces a durable cost advantage or a clear lead in enterprise engineering. Otherwise, Cognition risks funding an extremely expensive technology race in which competitors can improve just as quickly.

Could Cognition's AI coding technology become much cheaper?

Cognition faces a serious risk that AI coding becomes cheaper faster than demand grows, particularly as Anthropic, OpenAI, Microsoft, and Cursor improve their competing products.

The pressure is already visible in the market's structure. Developers can choose between several capable coding agents, and companies increasingly offer coding assistance within broader AI subscriptions or existing enterprise software.

Cognition's model-development strategy provides a partial defense. By building on open-source models and applying specialized training, the company can potentially deliver high-quality coding assistance at a lower cost than using expensive frontier models for every request. Cheaper open-source models and bundled coding tools are risks we rate by likelihood in our AI coding assistant market report.

Its SWE-1.7 release demonstrated substantial performance gains over SWE-1.6, supporting the idea that specialized optimization can improve efficiency.

But competitors are pursuing similar approaches. Cursor, for example, highlighted the economic benefits of gaining access to SpaceX's computing infrastructure after its acquisition.

The bigger question is how customers decide what to pay for.

Simple code completion is relatively easy to compare across products. A successful migration of a complicated banking application is much harder to replace with a cheaper subscription.

Cognition is pushing toward these more complex workflows, where customers can evaluate the value of completed work rather than simply counting generated tokens.

Its newer security products follow the same logic. Devin Security Swarm can investigate vulnerabilities and prepare fixes, extending the company's reach into work that security teams already spend considerable money completing.

We expect basic coding assistance to become much cheaper. Complex, reliable automation inside large enterprise systems should retain more pricing power, at least while the work remains difficult to reproduce.

How does Cognition's valuation compare with Cursor?

Cognition is valued at a much higher revenue multiple than Cursor's reported $60 billion acquisition price implies, even though Cursor has the larger business.

Cognition's $48 billion valuation represents approximately 48 times its latest $1 billion annualized revenue.

Cursor, meanwhile, was acquired by SpaceX for a reported $60 billion. Sacra estimated that Cursor had reached $4 billion in annualized revenue by May 2026, a figure also reported by The Information in August.

Using those figures, Cursor's acquisition price represents approximately 15 times annualized revenue.

The comparison is especially interesting because Cursor's $60 billion transaction was announced in April and completed in August, while Cognition's $48 billion round followed in September.

The two companies have different commercial mixes, and the Cursor transaction involved a strategic buyer with its own computing infrastructure and AI ambitions. We should therefore avoid treating the acquisition multiple as a perfect market benchmark.

Still, Cognition's 48-times multiple is approximately 3.2 times Cursor's implied 15-times multiple.

That is a substantial premium for a company with a smaller revenue base. Cognition's enterprise automation ambitions might eventually justify a higher multiple, but investors are already giving it considerable credit for achieving that future.

An earlier comparison produces a narrower gap. When Cursor was reportedly discussing financing at a $50 billion valuation in April, its annualized revenue had passed $2 billion, implying a multiple below 25 times. Later revenue estimates make the valuation gap even clearer.

Company or scenario	Revenue reference	Valuation	Implied multiple
Cognition, latest confirmed run rate	$1B	$48B	48x
Cursor, SpaceX acquisition	~$4B	$60B	~15x
Cursor, earlier financing discussions	More than $2B	$50B proposed	Below 25x
Cognition at Cursor's ~15x multiple	$1B	$15B	15x
Cognition at 20x revenue	$1B	$20B	20x
Cognition at 30x revenue	$1B	$30B	30x
Cognition at 40x revenue	$1B	$40B	40x
Cognition at its actual valuation	$1B	$48B	48x

How much revenue would Cognition need to justify $48 billion?

Cognition would need approximately $2.4 billion in annualized revenue to trade at 20 times sales, or $3.2 billion at 15 times, assuming its valuation stays at $48 billion.

Those targets are respectively 2.4 and 3.2 times its latest confirmed $1 billion revenue run rate.

The Information reported in August 2026 that Cognition executives had projected more than $1.5 billion in annualized revenue by the end of 2026 and $4 billion to $5 billion the following year.

These figures are management projections reported by journalists, rather than achieved financial results.

If Cognition reaches $1.5 billion, its current valuation would fall to 32 times annualized revenue. At $4 billion, the multiple would be 12 times, and at $5 billion it would be 9.6 times.

The mathematics explain why investors are prepared to pay such a high price today. They expect Cognition to become several times larger in a relatively short period.

But growing from $1 billion to $4 billion would require a fourfold revenue increase. Reaching $5 billion would require five times its current scale.

Even a company growing as quickly as Cognition faces serious execution challenges at those levels. Customer acquisition must continue, existing accounts must expand, and serving more workloads cannot consume all the additional revenue.

Could Cognition really become a $5 billion revenue business?

Cognition has a credible path toward several billion dollars in revenue, especially if autonomous agents take over a meaningful share of enterprise software maintenance. Reaching $5 billion on the reported timeline would still require extraordinary growth.

The opportunity extends beyond developers paying for an AI coding assistant.

Large companies spend heavily on application modernization, security remediation, infrastructure migration, and maintaining legacy systems. Cognition is already targeting these budgets through Devin, AWS integrations, and specialized products. What has to happen for coding agents to win modernization budgets is covered under growth drivers in our report on AI coding assistants.

Its Mercedes-Benz modernization example illustrates why customers might pay substantial amounts for completed engineering work. Replacing months of expensive manual development with automated execution can create considerable savings.

Cognition's growing enterprise relationships also provide a route to larger contracts. A company spending $1 million on Devin could potentially spend significantly more if autonomous engineering becomes useful across dozens of internal teams.

However, the company has not disclosed enough information about customer expansion, retention, or usage frequency to calculate how quickly those opportunities could translate into revenue.

The competitive environment also makes the timeline harder to predict. Claude Code already surpassed $2.5 billion in annualized revenue earlier in 2026, while Cursor reached approximately $4 billion.

Those businesses demonstrate that the market can support multibillion-dollar coding products. They also show that Cognition will be fighting for spending against much larger rivals.

We see a credible multibillion-dollar opportunity for Cognition. The difficult part is reaching that scale with margins attractive enough to justify its valuation, rather than simply purchasing growth through massive computing expenditure.

What could cut Cognition's valuation in half?

Cognition could lose half its $48 billion valuation even while continuing to grow, simply because investors stop paying such an extraordinary multiple of annualized revenue.

At $1 billion in annualized revenue, a 30-times multiple implies a $30 billion valuation, approximately 37.5% below the latest financing price.

At 25 times revenue, the implied valuation falls to $25 billion, a decline of nearly 48%. A multiple of 24 times would cut the valuation exactly in half, to $24 billion.

That outcome would not require Devin to disappear or its customers to abandon the product.

The most immediate threat is a growth slowdown. Investors are pricing Cognition for several more years of rapid expansion. If revenue increases much more slowly than expected, the existing valuation becomes difficult to defend.

Competition is another risk. Anthropic, OpenAI, Microsoft, and Cursor could put pressure on pricing while offering increasingly capable autonomous engineering products.

Cognition's computing obligations add financial exposure. The company could spend approximately $800 million in cash during 2026, according to The Information, and its enterprise gross margin was reportedly around 50%.

Even if growth remains strong, investors may become less willing to pay premium software multiples for businesses that need enormous amounts of capital to train and operate their models.

There is also a clear valuation benchmark. Cursor's reported $60 billion acquisition at an estimated $4 billion revenue run rate implies roughly 15 times revenue, far below Cognition's current multiple.

We think multiple compression is a more immediate valuation risk than outright business failure. Cognition could build an increasingly successful company while investors who entered at $48 billion still lose money.

Is Cognition really worth $48 billion today?

Cognition's $48 billion valuation is still ahead of what its current financial performance can comfortably justify. We consider it an aggressive bet on exceptional future growth, backed by a real business but priced for near-flawless execution.

The commercial achievements are difficult to dispute. Cognition has crossed $1 billion in annualized revenue, attracted major enterprise customers, expanded through Windsurf, and demonstrated that autonomous coding agents can complete valuable engineering work.

Its revenue also grew almost as quickly as its valuation between May and early September 2026, then reached another milestone later that month. At the latest confirmed revenue level, its valuation represents approximately 48 times annualized revenue.

But the financial risks are equally concrete. Cognition's enterprise gross margin was reportedly around 50%, its annual cash burn could approach $800 million, and its competitors include Anthropic, OpenAI, Microsoft, and Cursor.

The comparison with Cursor is particularly revealing. SpaceX paid a reported $60 billion for a company generating approximately $4 billion in annualized revenue, implying around 15 times revenue. Cognition is valued at roughly 48 times.

We would currently place Cognition in a more defensible illustrative range of 25 to 35 times annualized revenue, equivalent to approximately $25 billion to $35 billion at its latest disclosed scale. That range is based on our assessment of the company's growth, competitive risks, and financial performance, rather than an observed market-clearing price.

A $48 billion valuation could become much easier to justify if Cognition approaches $4 billion to $5 billion in annual revenue while improving margins substantially. At that scale, the current price would represent approximately 9.6 to 12 times revenue.

For now, however, investors are paying for that future before it has arrived. Cognition is one of the most impressive growth stories in AI software, but we do not think its demonstrated fundamentals yet support a $48 billion valuation.