SpaceX Wants Cognition AI: Smart Deal or An Overreach?

SpaceX has barely finished its $60 billion acquisition of Cursor, yet Elon Musk’s company is already being linked to another AI coding startup. This time, the target is Cognition, the company behind Devin, an autonomous AI software engineer valued at $26 billion just three months ago. The strategic logic is easy to understand. The financial logic is much harder.

Let's break down what SpaceX reportedly wants from Cognition, how Devin could fit alongside Cursor and Grok, and why a compute partnership may create more value than a full takeover.

What Happened Between SpaceX and Cognition AI?

According to Bloomberg, SpaceX approached Cognition about a possible acquisition. Cognition did not engage with the approach.

The report added that discussions about working together may still be continuing. One possibility is that Cognition could use SpaceX’s AI computing infrastructure to train and operate its models.

Cognition CEO Scott Wu disputed the takeover report. He wrote that Cognition was “not for sale” and that the companies had not been in talks. However, his response did not specifically address the report about a possible computing partnership, according to TechCrunch.

There is no announced offer, agreed price, active negotiation or acquisition agreement. What investors have is a reported approach that Cognition apparently rejected, along with the possibility of a commercial partnership.

The more useful question is therefore not whether SpaceX is buying Cognition today. It is why SpaceX would want Cognition after already spending $60 billion on Cursor.

What Is Cognition AI and Why Is Devin Valuable?

Cognition was founded in 2023 and became famous for Devin, which it describes as an autonomous AI software engineer.

A normal AI coding assistant helps a developer while the developer is working. It might suggest code, explain a function or identify an error. Devin is designed to go further. It can receive a task, examine a codebase, plan the work, write and test the code, fix errors and submit the finished work for human review.

Think of the difference this way. A traditional coding assistant is a power tool in an engineer’s hand. Devin is closer to a junior engineering team that can be assigned work and checked later.

Cognition also owns Windsurf, an AI coding environment that it acquired in 2025. At the time, Windsurf had $82 million in annual recurring revenue, more than 350 enterprise customers and hundreds of thousands of daily users, according to Cognition’s acquisition announcement.

That gives Cognition two complementary products:

Cognition’s commercial growth explains why SpaceX may be interested.

In May 2026, Cognition said it had raised more than $1 billion at a $26 billion valuation. Its annualised revenue run rate had reached $492 million and enterprise usage had increased more than 10 times since the beginning of the year, according to the company’s funding announcement.

By August, Bloomberg reported that Cognition’s annualised revenue was approaching $1 billion. The company was reportedly considering another funding round at a valuation of at least $40 billion.

That would mean Cognition nearly doubled its revenue run rate in less than three months. The number remains privately reported and has not been audited publicly, but the direction is clear: Cognition is growing unusually quickly.

Why Cognition AI Stands Out in the AI Coding Market

Cognition’s most valuable asset may not be a particular model or coding benchmark. Its bigger advantage is that it has learned how to sell autonomous AI into large and heavily regulated organisations.

Its publicly named customers include Citi, Goldman Sachs, Dell, Mercedes-Benz, Santander, the US Army and the US Navy.

At Mercedes-Benz, Devin analysed more than 200,000 lines of COBOL code and reportedly reduced the estimated timeline for one modernisation project from eight months to eight days. Mercedes is now deploying Devin and Windsurf across engineering teams in the US, Europe and Asia, according to Cognition.

GE Aerospace says one constrained engineering team nearly doubled its output of completed user stories during a three-month deployment. The company gradually expanded adoption after testing Devin on specific workflows, rather than immediately rolling it out across the organisation, according to the GE Aerospace case study.

Cognition is also developing a government business. Its platform is used across parts of the US Army, Navy, Treasury Department and NASA’s Jet Propulsion Laboratory. Windsurf already supports high-security government environments, while Devin is progressing through the FedRAMP High authorisation process, according to Cognition’s government update.

These relationships matter because SpaceXAI’s biggest weakness has not been access to GPUs. It has been converting AI technology into a trusted enterprise product.

Why SpaceX May Want Cognition AI After Buying Cursor

At first glance, buying both companies looks unnecessary. Cursor and Cognition both sell tools that help engineers create software. However, they enter the customer’s workflow through different doors.

Cursor is mainly a developer-first product. It sits inside the coding environment and helps an engineer write, understand and edit software faster.

Devin is closer to an autonomous worker. It receives a project and continues working in the cloud even when the human engineer is doing something else.

Cognition also has Windsurf, which competes more directly with Cursor. But its strongest differentiation comes from Devin, enterprise deployment support, legacy-modernisation expertise and regulated-industry relationships.

The possible SpaceXAI product stack would look like this:

SpaceX has been building the factory, the engine and the workbench. Cognition could provide an autonomous workforce and an enterprise sales team that knows how to get through procurement, security and compliance reviews. That is the strategic case.

Where Could the SpaceX-Cognition Synergies Come From?

1. SpaceX could sell Cognition computing capacity

SpaceX ended Q2 2026 with 1.4 gigawatts of nameplate AI compute capacity. Management expects capacity to exceed 2 gigawatts by the end of 2026 and move closer to 10 gigawatts than 5 gigawatts by the end of 2027.

The company has already signed cloud-service agreements with Google, Anthropic and other customers. It reported $14.1 billion of contracted AI cloud sales during Q2, according to SpaceX’s Q2 earnings release.

Cognition’s products require considerable computing capacity because its agents can work for hours, operate virtual machines, examine entire codebases and repeatedly call different AI models.

A compute agreement would give Cognition access to SpaceX’s infrastructure while giving SpaceXAI another fast-growing customer. Importantly, SpaceX does not need to buy Cognition to capture this benefit.

The following is an illustrative sensitivity model. Cognition has not disclosed its actual infrastructure expenses.

Even the high scenario would not financially justify paying $40 billion or more for Cognition. It would, however, make a long-term compute partnership attractive because SpaceX could earn infrastructure revenue without paying an acquisition premium.

Buying Cognition merely to secure its computing spend would be like buying a customer to increase sales.

2. Cognition could improve Grok’s software-engineering capabilities

Cursor has already worked with SpaceXAI on Grok 4.5 and Grok 4.6. Cursor contributed coding workflows, technical knowledge and large volumes of software-related data while using SpaceX’s Colossus infrastructure.

Cognition could add a different form of technical feedback.

Every Devin session contains the original task, planning process, actions taken, testing steps, errors, corrections and final output. That makes autonomous-agent traces potentially useful for understanding how models behave across longer engineering tasks.

However, SpaceX would not automatically receive permission to train Grok on Cognition customers’ private code.

Cognition’s enterprise terms state that customer data cannot be used to train or optimise an AI model without the customer’s prior written consent. Ownership does not make contractual and privacy restrictions disappear.

Therefore, the data synergy could be meaningful for Cognition’s own internal data, synthetic workloads and customers who opt in. It should not be valued as unrestricted access to every enterprise codebase using Devin.

3. SpaceX could cross-sell AI products to Cognition customers

Cognition already has relationships with banks, carmakers, aerospace companies, technology firms and government agencies.

SpaceX could potentially sell these customers:

• Grok enterprise models
• SpaceXAI computing capacity
• Cursor subscriptions
• Devin deployments
• AI security and software-modernisation tools

In the other direction, SpaceX could introduce Devin to Starlink enterprise customers, Starshield customers and organisations already buying SpaceXAI infrastructure.

SpaceX management has said it expects to benefit from a combined enterprise sales capability after integrating Cursor. Cognition would make that enterprise channel wider.

Still, customer logos should not be mistaken for guaranteed cross-selling. A bank that trusts Cognition for software modernisation will not automatically buy Grok or move its computing workloads to SpaceX.

4. Devin could improve SpaceX’s internal engineering productivity

SpaceX operates highly complex software across rockets, satellites, Starlink networks, data centres and AI models. Devin could help with testing, code review, legacy systems, security fixes and routine engineering tasks.

But internal productivity cannot be the main financial justification for a $40 billion to $60 billion acquisition.

Consider a transparent example:

Even if the model were expanded to 10,000 engineers, the implied productivity value would be around $200 million a year. Capitalising that benefit at 15 times would create approximately $3 billion of value.

These are hypothetical inputs, not SpaceX workforce data. The point is the order of magnitude. Faster internal software development would be valuable, but nowhere near enough to support a $50 billion purchase price.

Cognition itself warns that engineering hours saved are not the same as business value. Its productivity estimator has been tested across enterprise deployments, but individual estimates remain noisy and the underlying data relies partly on self-reported human estimates.

Key Acquisition Risk: Cursor and Windsurf Would Compete

Cognition does not bring only Devin. It also owns Windsurf. If SpaceX acquired Cognition, it would own Cursor and Windsurf, two products competing for broadly similar developer workflows.

SpaceX would need to decide whether to:

1. Maintain Cursor and Windsurf as separate brands.
2. Merge Windsurf into Cursor.
3. Keep Windsurf for government and enterprise deployments.
4. Use Devin as the autonomous layer across both products.
5. Discontinue overlapping features.

Every choice creates risk. Maintaining both products increases duplicated engineering, sales and marketing expenses. Combining them could alienate customers and employees. Discontinuing one product could destroy part of the value SpaceX just paid to acquire.

The cleanest strategic architecture would probably be to keep Cursor as the main developer interface, retain Devin as the autonomous cloud agent and use Windsurf’s government and enterprise capabilities selectively. But that is an analytical possibility, not an announced SpaceX plan.

The Model-Neutrality Problem: Could SpaceX Ownership Hurt Cognition AI's Model Neutrality?

Cognition promotes Devin as a model-independent platform.

Instead of forcing customers to use one AI model, Cognition evaluates models from different providers and selects the model that offers the best performance and cost for a particular task. This can include Cognition’s own models and models developed by other AI laboratories.

That independence is part of the product’s appeal. Enterprises do not want to rebuild their workflows whenever a different model becomes the market leader.

SpaceX ownership could weaken that position.

Customers may worry that Cognition will gradually prioritise Grok, even when another model is better for a particular task. OpenAI and Anthropic may also reconsider the commercial terms under which their models are supplied to a company owned by a direct competitor.

SpaceX could preserve Cognition’s open model-routing system. But it would need to demonstrate that model selection remains genuinely neutral. Otherwise, the acquisition could remove one of the reasons customers chose Cognition in the first place.

Could Cognition Strengthen SpaceX’s Government Business?

This is one of the less obvious but potentially valuable synergies.

SpaceX already has major US government relationships through launch services, Starshield and national-security satellite contracts. Cognition has built a separate position in government software modernisation.

Combining the two could create an end-to-end government technology offering covering:

• Secure satellite connectivity
• Data infrastructure
• AI computing capacity
• Software modernisation
• Cybersecurity remediation
• Autonomous engineering agents

However, this advantage also creates more regulatory and procurement scrutiny. Government customers would need to review ownership changes, data controls, model access and potential conflicts between commercial and defence workloads.

Cognition’s security authorisations and government relationships are valuable partly because they took time to build. They should not be treated as automatically transferable revenue.

How Expensive Is Cognition AI?

Cognition’s latest confirmed funding round valued it at $26 billion in May 2026, when its reported revenue run rate was $492 million.

By August, its revenue run rate was reportedly approaching $1 billion and investors were discussing a valuation of at least $40 billion.

Interestingly, the multiple fell between May and the reported August financing talks because revenue grew faster than the valuation.

But an acquisition price would probably have to exceed the price available to new investors. Scott Wu has repeatedly stressed the importance of Cognition remaining independent. A reluctant founder, rapid growth and strong access to private capital would all increase the price needed to secure control.

There is no disclosed offer. We use $40 billion, $50 billion and $60 billion only as scenario values.

What Growth Would Cognition AI Need to Justify Its Valuation?

A fast-growing company can justify a high revenue multiple if it continues growing quickly and eventually produces strong margins.

The question is how much growth would be required.

The following model assumes:

• Cognition begins with a $1 billion revenue run rate.
• SpaceX requires a 15% annual return.
• Cognition is valued at 10 times revenue in 2030.
• The measurement period is four years.
• Interim cash flows, retention costs and integration expenses are ignored.

At a $50 billion purchase price, Cognition would need to grow from around $1 billion to approximately $8.7 billion in revenue by 2030 merely to support a 15% annual return under these assumptions.

If the 2030 valuation multiple fell to eight times revenue, required revenue would rise to about $10.9 billion. If the multiple remained at 12 times, Cognition would still need approximately $7.3 billion.

This is possible only if autonomous software agents become a major enterprise software category and Cognition remains one of its dominant providers. It is not a conservative acquisition hurdle.

Acquisition or Partnership: Which Is Better for SpaceX?

Our view is that a partnership offers the better risk-adjusted outcome.

A well-structured partnership could include:

• A multi-year SpaceXAI compute commitment
• Preferred access to new SpaceXAI infrastructure
• Grok availability inside Devin and Windsurf
• Joint development of software-engineering models
• Optional co-selling to enterprise customers
• Clear restrictions on customer data use
• No requirement for Cognition to abandon rival models

This would capture much of the compute, distribution and technical benefit without forcing SpaceX to pay a control premium. The commercial agreement could also reveal whether the companies work well together before either side considers a larger transaction.

Cursor itself followed a similar path. The companies began with a compute and model-training partnership before SpaceX exercised its acquisition option.

What Should SpaceX Investors Watch Next?

The first thing to watch is not another takeover headline. It is whether the companies announce a compute or model partnership.

Investors should focus on the following questions:

• Does Cognition commit to using SpaceXAI infrastructure?
• Is there a minimum compute-spending commitment?
• Will Grok become available inside Devin or Windsurf?
• Does Cognition retain access to OpenAI and Anthropic models?
• Is customer data excluded from Grok training?
• Will Scott Wu remain independent or take a SpaceXAI role?
• Does SpaceX separately disclose Cursor’s revenue and margins?
• How much additional stock-based compensation is issued?
• Does SpaceX’s AI operating loss narrow as software revenue grows?
• Are customers renewing after ownership or partnership changes?

The answers would reveal whether SpaceX is building a genuine enterprise AI platform or simply collecting expensive AI assets.

Our View: SpaceX Should Partner With Cognition, Not Chase It

Cognition would unquestionably strengthen SpaceXAI.

It would add Devin, enterprise customers, government relationships, model-routing technology and a team that has already demonstrated rapid commercial growth. Combined with Cursor, Grok and SpaceX’s computing infrastructure, it could create one of the broadest AI software-development platforms in the market.

But strategic fit is not the same as financial value.

At a valuation of at least $40 billion, Cognition already prices in years of exceptional growth. A takeover premium could push SpaceX’s entry multiple above 50 times revenue. SpaceX would also be buying a second coding environment, accepting more dilution and integrating another founder-led company while its AI division is still evolving.

Our conclusion is therefore clear: SpaceX should aggressively pursue the partnership, but remain disciplined on ownership.

A compute, model and distribution agreement could deliver most of the strategic advantages. A full acquisition becomes financially sensible only if Cognition accepts a price close to its standalone valuation, its leadership commits to staying, and SpaceX can show a credible path toward at least $7 billion to $10 billion of Cognition revenue by 2030.

Until those conditions are visible, buying Cognition would look less like filling a missing piece and more like paying twice for the same part of the AI stack.