SpaceX and Tesla Study Potential Merger, Musk Consolidates AI Compute Footprint

By｜Blue Character Plan, Author｜Chester

One sentence from the world's richest man has once again turned the entire Silicon Valley tech world upside down.

On October 27, Bloomberg broke a piece of news capable of rewriting the history of human business: Musk's SpaceX and Tesla are studying a potential merger through a special committee.

Once the deal is finalized, a super-giant spanning heaven and earth, with a combined market value exceeding $3 trillion, will officially surface.

But this merger also carries the risk of Musk "lifting a rock only to drop it on his own feet."

As everyone knows, SpaceX, which builds rockets and Starlink, is deeply tied to the U.S. military and NASA, holding a large number of sensitive defense and aerospace contracts; while Tesla, which makes electric vehicles, is a global consumer goods company that relies on global supply chains and large overseas markets.

By conventional logic, these two companies can only prosper safely if they draw a complete line between them. Now Musk wants to put a sensitive American aerospace giant and a consumer company that sells cars globally under the same household registration—this is purely asking for trouble.

Not only would it face extremely strict national security and antitrust reviews, it would also amount to actively pushing Tesla into the crosshairs of geopolitics... Could it be that Musk still thinks he doesn't have enough regulatory troubles already?

Yet even knowing the minefield lies ahead, this is a risky move Musk has no choice but to make right now.

Building cars is no longer enough for Tesla

Of course, if Tesla were viewed through the lens of a traditional automaker, it would have no motive at all to merge with SpaceX; but the Tesla of today has long since stopped positioning itself as a traditional automaker.

On October 21, Tesla announced that the Optimus humanoid robot had begun production at its Fremont, California factory. This production line was previously used to manufacture electric vehicles. Musk said Optimus output is expected to reach 20,000 units per week by the end of 2027. At the same time, Tesla is advancing the Cybercab robotaxi project and continuing to develop the FSD autonomous driving system.

These moves all point to four words that are extremely hot right now: physical AI. Beyond building cars, Tesla also wants to prove that AI can truly control vehicles, robots and other devices in the real world. And that, precisely, requires massive amounts of chips and computing power.

The Terafab project jointly advanced by Tesla and SpaceX is exactly the hardcore intersection of the two companies at this link.

According to the project plan published by SpaceX, Terafab aims to integrate advanced logic chips, memory chips, packaging and testing into a single manufacturing system, with a long-term goal of producing chips corresponding to 1 terawatt of computing power each year. According to the timetable announced by Musk, the project plans to manufacture its first test chip in May to June 2027 and strive to achieve mass production in the third quarter of 2028.

These chips will serve two different types of demand: one for Tesla vehicles and Optimus robots, and the other targeting the space environment to support SpaceX's future orbital computing infrastructure.

For this reason, from Tesla's perspective, the reason for merging with SpaceX is very simple: I need computing power, and after merging with SpaceX, everyone can better integrate resources to build factories and make chips together, producing more computing power.

But Tesla making chips to give cars and robots stronger autonomous capabilities is one thing—why is SpaceX, a rocket company, also deeply mired in this computing power frenzy and unable to extricate itself?

SpaceX, an AI company that can launch rockets

In fact, tracing back to the source, SpaceX may have become an "AI company" earlier than Tesla. At least at the official level, as early as February this year, it acquired xAI, bringing AI businesses such as Grok and data centers under its wing; from that point on, Musk had already begun saying that SpaceX is an AI company.

According to the company's published third-quarter 2026 financial report, revenue of the SI division, which houses the AI business, reached $7.56 billion, up 830% year on year; in the same period, revenue from connectivity businesses including Starlink was $5.25 billion, while aerospace business revenue was $1.77 billion. The SI division has become SpaceX's largest revenue segment.

Revenue growth in this segment is mainly driven by cloud services and computing power leasing. In addition to serving its own Grok, SpaceX also sells computing power to Anthropic, which is led by Amodei.

Don't be fooled by Musk battling his rivals in the public arena—in reality, he is the biggest rival's computing power landlord.

As data center investment increased, SpaceX's computing capacity had reached 1.7 gigawatts by the end of September. In the third quarter, the SI division added about $6.7 billion in revenue year on year, mainly from cloud service agreements, the acquisition of the AI coding tool Cursor, and growth in models and software applications such as Grok.

But the meager sums earned from leasing computing power are still small fry compared with the real money SpaceX must first spend to expand computing power.

SpaceX's third-quarter capital expenditure reached as high as $19.593 billion, of which $16.444 billion went to infrastructure construction related to the AI business. The company expects fourth-quarter capital expenditure to rise further to $35 billion to $40 billion, and by the end of 2026, computing capacity will expand to 2.5 to 3 gigawatts.

Beyond money, the supply of electrical equipment is also slowing the pace of expansion. To address the power shortfall for AI data centers, SpaceX has even begun preparing production facilities in Bastrop, Texas, to manufacture blades and guide vanes for large gas turbines itself. Musk said this could shorten gas turbine deployment times by up to 18 months.

From scrambling for chips and building its own data centers to hand-rolling generator components, SpaceX is steadily extending its reach upstream into heavy industry. Tesla needs to put AI into cars and robots, while SpaceX is expanding the infrastructure that keeps AI running, and the two companies' demand for chips, energy and computing power is becoming ever more tightly intertwined.

After computing power, the contest is over intelligence

Piecing these fragments together, Musk's ambition is finally laid bare: from Tesla's cars and robots to SpaceX's rockets, Starlink and data centers, and on to the jointly advanced Terafab, he wants to control in his own hands all the resources needed to take AI from chip manufacturing to real-world application.

Sustaining such a vast tech empire clearly makes funding pressure a reality that cannot be ignored. Financial reports show that SpaceX's capital expenditure in the first nine months of 2026 reached $48.1 billion, with free cash flow close to negative $35 billion; Tesla achieved positive free cash flow of $1.88 billion over the same period, and it too must keep investing in cars, robots and chip R&D.

The appeal of a merger lies in advancing the two sides' cooperation from individual projects to resource arrangements across the entire companies.

How much to invest in chip R&D, where to build computing facilities, and what hardware cars and robots need can all be coordinated around shared needs; once the two companies merge into one, funds and assets can also be arranged more quickly and smoothly under a single corporate structure.

Ultimately, rockets handle transport, Starlink provides connectivity, data centers provide computing power, and chips provide the underlying hardware. By putting these resources together, Musk can run two businesses at once: selling infrastructure services externally and supporting his own AI products and machines internally.

But for its own products to win the competition as well, model capability remains an area that needs strengthening.

In September, SpaceX's AI business released Grok 4.7. But on the Terminal-Bench leaderboard, which measures whether agents can independently complete complex terminal tasks, Grok 4.7 paired with Grok Build scored 37.6%, ranking 10th on the leaderboard at the time; over the same period, Opus 5 paired with Claude Code scored 53.9%, and GPT-6 Sol paired with Codex scored 49.4%.

This gap bears on Grok's competitiveness with Cursor in the AI programming and agent markets. SpaceX can rent computing power to rivals and earn infrastructure revenue; but if it still wants to win customers with its own models and developer tools, Grok will need to deliver stronger task performance.

Tesla's cars and robots also each have their own intelligence challenges to overcome. Cars need to judge dynamic changes on the road, and robots need to understand real objects and complete continuous actions. These capabilities all depend on models, data and real-world training to solve, with chips and computing power merely inputs among them.

In the final analysis, Musk is willing to take on geopolitical risk to push for a merger of the two giants, but what truly determines the ceiling of this system is how much intelligence these machines can ultimately possess.