The quickest route to achieving 1 terawatt of computational power might be hiding in a place few are discussing: a swarm of robots.
On August 13, Morgan Stanley analyst Adam Jonas, in a recent report dissecting four key narratives for SpaceX this year, introduced a market-overlooked thesis. By 2040, the distributed computing power of a robot swarm could help SpaceX reach its 1-terawatt computing goal faster than any traditional data center.
2.2 Billion Robots: An Alternative Path to 1 Terawatt
Morgan Stanley’s global robotics model estimates that by 2040, the total number of robots in various forms will reach approximately 2.2 billion. The key assumption is that each robot is equipped with 500 watts of computing power. The 2.2 billion units would collectively generate about 1.1 terawatts of processing capability. This is based on Elon Musk’s indication that the Tesla AI5 chip, designed for the Optimus humanoid robot, consumes 250 watts. Morgan Stanley believes this figure will rise over time as computing demands increase and the robots' form factors expand.
Consequently, 2.2 billion robots multiplied by 500 watts equals 1.1 terawatts, all of which could theoretically be connected via Starlink. This is what Morgan Stanley calls a "distributed inference cloud"—a network of fixed and mobile edge nodes, linked to large-scale data center clusters, forming a hybrid computing architecture.
For comparison, Morgan Stanley’s own model for SpaceX predicts a computing scale of about 380 gigawatts by 2040, far below 1 terawatt. Musk has publicly discussed plans to build a "Terafab," a terawatt-scale computing factory, but external discussions have almost exclusively focused on ground-based or orbital data centers, with scant attention paid to this alternative.
Morgan Stanley notes that Tesla investors are not unfamiliar with the "distributed inference cloud" concept. Musk has repeatedly discussed using idle vehicles and Megapods, modular AI computing boxes deployed at Supercharger stations, to build a massive distributed computing network. As the collaboration between SpaceX and Tesla deepens, SpaceX investors also need to become acquainted with this computing deployment logic.
Grok: From Follower to Frontier
Beyond the computing narrative, Morgan Stanley is also closely watching SpaceX's AI model progress. Grok 4.6 has achieved a score of 61 on the Artificial Intelligence Index, approaching the current highest score of 63, and surpassing the 56 points for Grok 4.5 released in mid-July and the 38 points for Grok 4.3 from April. More importantly, its cost is comparable to China's Kimi K3, placing it on the value-for-money Pareto frontier.
Musk subsequently stated on X that "Grok 4.7 is much better than Grok 4.6 and should be ready in 3-4 weeks." Grok 4.7 is based on a new 2.1 trillion-parameter base model, while Grok 5 is expected to be released by the end of the year with a parameter count ranging from 6 trillion to 10 trillion.
Morgan Stanley points out that model capability is only one aspect; the real importance lies in actual usage. During the second-quarter earnings call, Musk mentioned that after the release of Grok 4.5, which served as the default model for Cursor, token usage tripled in a short period. However, Morgan Stanley also notes that OpenRouter, an industry-wide tracking platform, is currently unable to capture usage data from X and Cursor—precisely the two main sources of Grok's usage. Morgan Stanley expects that as the acquisition of Cursor is completed in the coming weeks, relevant data will be disclosed in more detail during earnings reports.
Morgan Stanley believes most investors still view SpaceX AI as a "moderately successful emerging cloud service provider." However, the firm argues that SpaceX does not need to achieve Musk's stated goal of 10 gigawatts of computing power by the end of 2027 to realize a significant valuation re-rating. Stripping out Morgan Stanley's DCF-implied valuation of $127 per share for the "Space + Connectivity" business from the current stock price of around $140, the remaining implied valuation for the AI business is only about 1.5 times the 2028 EV/Sales, which is below the average for similar emerging cloud service providers. "Proving that SpaceX can maintain its position at the AI frontier and convert that into real usage is key to achieving a significant re-rating toward the $300 price target," Morgan Stanley wrote.
Computing Pricing: Conservative Forecasts Face Upside Risks
Morgan Stanley also sees potential upside signals for SpaceX's valuation from CoreWeave's earnings report. Due to tight demand and customers migrating to inference, CoreWeave raised prices across all its SKUs by approximately 25%, and has signed a contract for NVIDIA Ampere chips extending into 2029.
In its SpaceX model, Morgan Stanley predicts a weighted average computing price of around $31 per watt in the fourth quarter of 2026. However, influenced by a large amount of industry capacity coming online, it forecasts a compression to about $17 per watt by the fourth quarter of 2027, when the computing scale is only about 5 gigawatts. But Morgan Stanley also conducted a stress test: if the $31 per watt price persists throughout 2027, it would generate approximately $42 billion in incremental revenue, with total revenue growth roughly tripling compared to the current model's prediction of a doubling. If the price reaches $50 per watt, revenue growth would approach 4.5 times—all this under the premise that the computing scale is only about half of Musk's target.
Morgan Stanley also cites data from Compute Desk, noting that Blackwell GPU rental prices have risen by about 30% year-to-date to approximately $5.25 per GPU hour, which translates to about $20-25 per watt.
Starship's 14th Flight: The Most Important Catalyst Since the IPO
Morgan Stanley considers the 14th flight of Starship to be one of the most significant stock catalysts for SpaceX since its IPO, aside from its earnings reports. The 14th flight is expected to be the first attempt to capture the Starship upper stage/ship. The 21st booster is reported to have completed cryogenic testing around July 20 and is currently undergoing engine installation, with a static fire test expected around August 20. Based on the recent pace of approximately two weeks between the static fire of the V3 booster and its flight, the hardware timeline points to a September launch.
On the regulatory front, SpaceX still needs to obtain FAA approval for the ship's tower catch. The existing Part 450 operating license covers launches and booster catches, but the ship catch requires a license modification. Referring to the precedent of the 5th flight, which was the first booster catch, new information was submitted in mid-August 2024, but approval was not granted until the day before the flight in mid-October. Morgan Stanley maintains its baseline expectation of early September.
"If SpaceX can successfully 'catch the ship' on the 14th flight... we believe this will represent one of the most important milestones in spaceflight history," Morgan Stanley wrote. From an investor's perspective, a "catch" would mean a significant reduction in the cost of orbital "real estate," opening up new growth opportunities for its space, connectivity, and AI businesses. If the 14th flight is successful, the 15th flight could attempt a simultaneous catch of the ship and the booster, which is a more difficult maneuver.
Comments