According to BlockBeats, on August 14th, Hyperliquid announced several new features for HIP-4: Deployers can now add named results after issue creation, with the initial balance of new results matching the outstanding catch-all balance; template instantiation now requires setting `deployerFeeScale`—similar to the cost multiplier in HIP-3, which deployers can configure between 0 and 10 to charge for results; a new `shortString` type hint has been added for template-side names, facilitating interface field parsing; billing details have been pruned from L1 status, allowing dependent node APIs or applications reading pre-compiled data to index the required data themselves; multiple template sets have been added to the testnet, with template IDs ordered by sequence number suffixes, and all except the largest sequence number in each set are considered deprecated templates.

After the next network upgrade, the results market for validator deployments will implement a fee mechanism, with the average transaction fee in the results market being half that of non-result spot transactions. The HIP-4 mainnet launch will adopt a conservative phased specification: each deployer will initially be limited to 100 concurrent results and a maximum of 500 results deployments per day; once the technology is stable, the deployer limit is expected to be rapidly increased to 1000 concurrent results and 5000 deployments per day, and a further limit-raising scheme will be designed based on feedback.

According to BlockBeats, on August 14th, Morgan Stanley analyst Adam Jonas made a previously overlooked prediction in his latest research report: by 2040, the distributed computing power of robot swarms may help SpaceX reach its 1 terawatt computing power target faster than any data center. Morgan Stanley estimates that by 2040, the total number of robots of various forms will reach approximately 2.2 billion. Assuming each robot carries 500 watts of computing power, this would amount to approximately 1.1 terawatts, theoretically all of which could be connected to the Starlink network.

Musk previously revealed that the Tesla AI5 chip designed for Optimus consumes 250 watts, a figure Morgan Stanley believes will increase over time. This is what Morgan Stanley calls a "distributed inference cloud"—a network of fixed and mobile edge nodes connecting to large-scale data center clusters to form a hybrid computing architecture. Morgan Stanley points out that Tesla investors are no strangers to "distributed inference cloud," as Musk has repeatedly discussed using idle vehicles and Megapods deployed at Supercharger stations to build a large-scale distributed computing network. As SpaceX and Tesla deepen their collaboration, SpaceX investors also need to familiarize themselves with this computing deployment logic.

Beyond the computing power narrative, Morgan Stanley also tracked SpaceX's AI model progress—Grok 4.6 has reached 61 points on the Artificial Analysis Intelligence Index, close to the current highest of 63, and its cost is comparable to Kimi K3. Musk stated that Grok 4.7 will be ready in 3 to 4 weeks, and Grok 5 is expected to be released by the end of the year. Regarding computing power pricing, the weighted average computing power price is predicted to be approximately $31/watt in the fourth quarter of 2026, but stress tests show that if this price persists throughout 2027, it will generate approximately $42 billion in incremental revenue, representing a nearly threefold increase in revenue.

Morgan Stanley also emphasized that the 14th Starship flight is one of the most important stock price catalysts for SpaceX since its IPO. Achieving the first successful launch tower capture would represent one of the most significant milestones in spaceflight history and open up new growth opportunities for its space, connectivity, and AI businesses. Morgan Stanley gave SpaceX a target price of $300, believing that proving SpaceX's ability to maintain its leading position in AI models and translate them into real-world usage is key to achieving a significant valuation revaluation.