qinbafrank

qinbafrank

83,910 Twitter followers

Follow

Investor in Crypto、TMT、AI ，跟踪最前沿科技趋势、野生宏观政经观察、研究全球资本流动性、周期趋势投资。记录个人学习和思考，经常出错常态掉坑爬坑。Runner🏃

Posts

Let's discuss RKLB's latest ATM (Auction, Purchase, and Transfer) stock offering plan. RKLB signed a new stock allocation agreement yesterday to sell common stock via ATM, with a maximum total of $3 billion. Since March 2025, RKLB has conducted four ATM stock offerings, raising nearly $4 billion to $5 billion if all are completed. This demonstrates a continuous stream of funds raised from the market. There's been much discussion, and here are my personal thoughts: 1. The timing is excellent. With SpaceX's IPO approaching, the space economy sector will receive significant attention in the near future. Raising a large sum of money during a favorable market is a very wise move from a business perspective, allowing them to strengthen their balance sheet while valuations are high and the market is bullish. 2. Why raise so much money? The company's filings with the SEC are very standard, stating the funds will be used for growth, acquisitions, and future operating expenses. In my opinion, there are two main points: 1) To reserve ammunition for future mergers and acquisitions and vertical integration. RKLB has a history of numerous acquisitions, giving the company extensive experience in M&A and integration. These past acquisitions have enhanced the company's competitiveness in many dimensions. The company recently completed the acquisition of Mynaric, strengthening its laser optical communication capabilities and establishing a European business footprint; it also signed an acquisition agreement for Motiv, aiming to in-house manufacture satellite components with limited supply or high costs, such as solar array drive components and precision motion control. This indicates that management's intention is not simply to cover losses, but to continue expanding along the end-to-end space infrastructure platform of "launch + spacecraft + components + national security". 2) Increasing Capital Buffer for Neutron Launch and Defense Projects In the first quarter, the company simultaneously disclosed 31 new Electron/HASTE contracts and 5 new Neutron launch contracts, exceeding the total number of launch orders for 2025; it also mentioned projects such as Space Based Interceptor, MACH-TB, Neutron first flight hardware integration, and Archimedes engine qualification. These projects share common characteristics: high upfront investment, working capital, inventory, and engineering redundancy all require significant cash reserves. Sufficient financial reserves give the company more confidence when bidding for and delivering large projects. Defense and constellation clients consider a supplier's long-term delivery capabilities, and financial strength is a crucial factor. 3. Of course, shareholders will feel uncomfortable in the short term. Especially yesterday's issuance of shares, roughly corresponding to about 4% of common stock, meant our holdings were diluted by 4% at once. This led to some rather emotional comments, and the stock price suffered in the short term. Previous share issuances have also caused stock price corrections. However, these corrections may also be short-term buying opportunities, and I personally will not reduce my holdings at this point. Back at the end of last year, x.com/qinbafrank/status/200053...…就聊到的逻辑，spacex能带动一波太空经济浪潮，而且之前这里x.com/qinbafrank/status/204110...…也聊到，太空经济局已经形成自循环：低成本发射 → massive satellites → rich data/services → new business models (such as space data centers, zero-gravity manufacturing) → more investment and applications. RKLB is not a second SpaceX; it has its own unique competitive advantages. These competitive advantages have been built up over many years with significant time and capital. From my personal perspective, after SpaceX's IPO and subsequent rally (which will also drive up the overall sector), that might be the selling point for the space economy. I will likely reduce my holdings around that time. This is just my personal opinion for reference. twitter.com/qinbafrank/status/...

Let's discuss RKLB's latest ATM (Auction, Purchase, and Transfer) stock offering plan. RKLB signed a new stock allocation agreement yesterday to sell common stock via ATM, with a maximum total of $3 billion. Since March 2025, RKLB has conducted four ATM stock offerings, raising nearly $4 billion to $5 billion if all are completed. This demonstrates a continuous stream of funds raised from the market. There's been much discussion, and here are my personal thoughts: 1. The timing is excellent. With SpaceX's IPO approaching, the space economy sector will receive significant attention in the near future. Raising a large sum of money during a favorable market is a very wise move from a business perspective, allowing them to strengthen their balance sheet while valuations are high and the market is bullish. 2. Why raise so much money? The company's filings with the SEC are very standard, stating the funds will be used for growth, acquisitions, and future operating expenses. In my opinion, there are two main points: 1) To reserve ammunition for future mergers and acquisitions and vertical integration. RKLB has a history of numerous acquisitions, giving the company extensive experience in M&A and integration. These past acquisitions have enhanced the company's competitiveness in many dimensions. The company recently completed the acquisition of Mynaric, strengthening its laser optical communication capabilities and establishing a European business footprint; it also signed an acquisition agreement for Motiv, aiming to in-house manufacture satellite components with limited supply or high costs, such as solar array drive components and precision motion control. This indicates that management's intention is not simply to cover losses, but to continue expanding along the end-to-end space infrastructure platform of "launch + spacecraft + components + national security". 2) Increasing Capital Buffer for Neutron Launch and Defense Projects In the first quarter, the company simultaneously disclosed 31 new Electron/HASTE contracts and 5 new Neutron launch contracts, exceeding the total number of launch orders for 2025; it also mentioned projects such as Space Based Interceptor, MACH-TB, Neutron first flight hardware integration, and Archimedes engine qualification. These projects share common characteristics: high upfront investment, working capital, inventory, and engineering redundancy all require significant cash reserves. Sufficient financial reserves give the company more confidence when bidding for and delivering large projects. Defense and constellation clients consider a supplier's long-term delivery capabilities, and financial strength is a crucial factor. 3. Of course, shareholders will feel uncomfortable in the short term. Especially yesterday's issuance of shares, roughly corresponding to about 4% of common stock, means our holdings were diluted by 4% at once. This has led to some rather emotional comments, and the stock price has suffered in the short term. Previous share issuances have also caused stock price corrections. However, these corrections may also be short-term buying opportunities. I personally will not reduce my holdings during this pilot program. Back at the end of last year, x.com/qinbafrank/status/200053...…就聊到的逻辑，spacex能带动一波太空经济浪潮，而且之前这里x.com/qinbafrank/status/204110...…也聊到，太空经济局已经形成自循环：低成本发射 → massive satellites → rich data/services → new business models (such as space data centers, zero-gravity manufacturing) → more investment and applications. RKLB is not a second SpaceX; it has its own unique competitive advantages. These competitive advantages have been built up over many years with significant time and capital. From my personal perspective, after SpaceX's IPO and subsequent rally (which will also drive up the overall sector), that might be the selling point for the space economy. I will likely reduce my holdings around that time. This is just my personal opinion for reference. twitter.com/qinbafrank/status/...

Why did Nvidia suddenly separate edge computing in its latest quarterly earnings report? Nvidia's latest earnings report explicitly states that the company is switching to a new reporting framework, disclosing information across two market platforms: data centers and edge computing. Edge computing includes data processing devices used for both gentle and physical AI, such as PCs, game consoles, workstations, AI-RAN base stations, robots, and automobiles. Nvidia also disclosed that Edge Computing revenue for the latest quarter was $6.4 billion, a 10% increase quarter-over-quarter and a 29% increase year-over-year; data center revenue for the same quarter was $75.2 billion, a 92% increase year-over-year. These figures are crucial: edge computing is not currently Nvidia's main revenue engine, accounting for less than 8% of total revenue; however, it has been placed on par with data centers as a "second platform." Why separate a segment with such a small share and make it grow alongside the core business? My understanding is as follows: 1. Nvidia is actively reshaping its narrative: from "selling data center GPUs" to "a full-stack AI operating system." In the past, Nvidia's valuation narrative mainly focused on its cloud AI factory. However, this new classification is equivalent to Nvidia re-dividing its business into two major worlds. This isn't an accounting technique issue, but a valuation framework issue. Previously, Nvidia's edge-related businesses were scattered across Gaming, Professional Visualization, Automotive, and OEM. By consolidating these into Edge Computing, it's essentially telling investors: these aren't fragmented businesses, but a second growth curve within the same AI era. 2. It wants to prove that the CUDA moat extends beyond the data center to the physical world. What Nvidia really wants to sell isn't a single GPU, but a platform from cloud to edge to robotics: CUDA + GPU + networking + Isaac + Omniverse + Drive + Jetson + RTX + AI-RAN. If this system only remains in the cloud, Nvidia's ceiling is its data center capital expenditure. However, if Nvidia enters the automotive, robotics, factory, edge server, AI PC, and AI base station sectors, its logic shifts from a "data center chip company" to a "general-purpose computing platform company for the AI era." Nvidia's financial reports also highlight edge computing in areas such as RTX local agentic AI, autonomous driving, Cosmos, Isaac GR00T, industrial software, and AI-RAN. This indicates that it wants to prove one thing: AI is not just about answering questions in the cloud; it also needs to see, understand, move, operate, and make decisions in the real world. 3. Alleviating market concerns about the "capital expenditure cycle of cloud vendors" Nvidia's biggest problem now is not insufficient growth, but market concerns: what will happen to Nvidia's growth rate if Microsoft, Google, Amazon, and Meta slow down their AI capital expenditures? Therefore, Nvidia needs to tell the market: my next phase is not just about hyperscalers; I also have enterprise AI, industrial AI, robotics, automotive, AI PCs, and AI-RAN. This is why it further breaks down its Data Center into hyperscale and ACIE, while also listing Edge Computing separately. It's presenting investors with a new roadmap: First growth curve: Cloud AI factory. Second growth curve: Enterprise and industrial AI. Third growth curve: Physical AI and edge AI. 4. Defining the investment narrative of "Physical AI" in advance. Nvidia has been emphasizing physical AI for the past two years. Physical AI is not just ordinary chatbots, but AI that can interact with the physical world, such as autonomous driving, robots, factory automation, warehouse robots, AI cameras, medical robots, drones, and smart grid inspection. Nvidia management stated in the earnings call that many industrial companies must place computing in context, where actions need to occur, and cannot rely entirely on the cloud; for example, chip factories cannot have all real-time control going to the cloud and back. Management also emphasized that the next wave is physical AI, with a large number of autonomous and robotic systems entering the physical world in the future. This is the core message behind Nvidia's separate listing of Edge Computing: It aims to transform "physical AI" from a long-term narrative into a trackable revenue stream. twitter.com/qinbafrank/status/...

Data centers are truly high-quality cash flow assets. Looking at SpaceX's IPO filing, Anthropic has agreed to pay SpaceX $1.25 billion per month for access to the computing power of the Colossus data center, with the contract running until May 2029. If fully executed, this amounts to nearly $45 billion. Initially, the main computing power will be from Colossus1 (approximately 300MW, over 220,000 NVIDIA GPUs), later expanded to Colossus2 (over 1GW). However, it's estimated that only a portion of Colossus2's computing power will go to Anthropic, as xAI currently primarily runs on Colossus2. Based on previously disclosed costs, Colossus1 cost $7 billion, and Colossus2 cost over $20 billion, bringing the total cost of the two data centers to nearly $30 billion. Currently, the entirety of Colossus1 and a small portion of Colossus2 (estimated to be small; based on market rates, consuming all of Colossus1's computing power would generate approximately $1 billion per month) can bring in $1.25 billion in monthly revenue, totaling $15 billion annually—a truly high-quality cash flow. It's worth noting that reports indicate SpaceX's initial IPO prospectus disclosed a data center construction cost of only $2.7 million per megawatt (industry benchmark approximately $12.3 million/megawatt), but this figure was removed in the final version, leaving only the phrase "considerably lower than industry benchmarks." It must be said that Musk's company's engineering capabilities are absolutely top-tier.

Why is xAI renting out the Colossus1 data center to Anthropic? This tweet is probably the most insightful analysis, with the core logic being: xAI currently holds a total of about 550,000+ GPUs (based on H100-equivalent performance), while Colossus1 (220,000 units) accounts for x.com/jukan05/status…

AI demand is growing parabolically, the potential market size for CPUs is further increasing, and edge computing was reported separately for the first time. Key takeaways from Nvidia's earnings call: 1. Jensen Huang stated directly: Demand is growing parabolically. The reason is simple: intelligent agent AI has arrived. Mainstream AI has transitioned from one-time reasoning to logical reasoning, and now it has entered the "intelligent agent" stage. AI is no longer optional, but a necessity. Tokens are now profitable. In the AI era, computing power equals revenue and profit. This also coincides with yesterday's... 2. The biggest incremental information and potential of the call came from Nvidia's significant investment in its CPU business. Huang announced: "The Vera CPU opens up a completely new $200 billion market for Nvidia, a market we've never been in before." At AMD's last earnings call, Lisa Su increased the TAM (Total Potential Market Size) for CPUs by $120 billion; today, Huang has once again pushed it to $200 billion. Faced with the characteristics of "AI agents" requiring extensive use of tools, browsers, and orchestration, GPUs alone are no longer sufficient; the market demands a completely new CPU architecture. Vera will not only be sold as a companion device to Rubin GPUs but also as a standalone CPU, storage node, and security node. Huang revealed that nearly $20 billion in total revenue from standalone CPUs is expected this year, marking Nvidia's readiness to become a leading global CPU supplier. (This coincides with Chen Liwu's speech yesterday | _2024111120231_| 3. Edge computing officially enters the mainstream.) Nvidia changed its reporting framework in this earnings report, dividing its business into two major platforms: "Data Center" and "Edge Computing." The Data Center business (Q1 revenue of $75 billion) is further subdivided into "Hyperscale Cloud Vendors (Hyperscale)" and "ACIE (AI Cloud, Industrial, and Enterprise)": 1) Hyperscale ($38 billion): Accounts for approximately 50% of the data center market share, a 12% increase quarter-over-quarter. 2) ACIE ($37 billion): A significant 31% increase quarter-over-quarter, with AI cloud revenue more than doubling year-over-year and sovereign AI revenue growing by over 80% year-over-year. Edge computing business reached $6.4 billion ($6.4 billion), a 10% increase quarter-over-quarter and a 29% increase year-over-year. This is a newly separated category. Edge computing encompasses end devices for agent AI and physical AI, including PCs, game consoles, workstations, AI-RAN base stations, robots, and automobiles. 4. Mass production and shipment of Vera Rubin began in Q3. By integrating seven dedicated chips, Vera Rubin's inference throughput can reach up to 35 times that of Blackwell. Vera Rubin will be more successful than Grace Blackwell in this respect. Every leading-edge model company will fully transition to Vera Rubin from the outset. twitter.com/qinbafrank/status/...

Today, Chen Liwu's speech boosted the CPU sector again, with ARM showing the largest increase. In fact, in early to mid-April, CPUs became a key component of proxy AI. A highlight of Chen Liwu's speech today was his first mention of a CPU-to-GPU ratio approaching 1:1, moving away from the previous 1:8. In Intel's earnings call in late April, he also mentioned a ratio approaching 1:4, representing a significant increase in the ratio and naturally stronger market expectations. The tweet from mid-April also explicitly mentioned that the CPU shortage would benefit US stocks like Intel, AMD, and ARM. A month later, all three stocks have seen substantial gains: Intel has doubled, while AMD and ARM have seen gains of around 60-70%. However, it's worth discussing that ARM had the largest increase among the three stocks today: firstly, ARM has the lowest market capitalization; secondly, the market may have realized that in the era of proxy AI, the ARM architecture actually has advantages over the x86 architecture in terms of CPU demand, as discussed in the mid-April tweet. Therefore, Nvidia's Vera CPU, and the self-developed CPUs of Amazon, Google, and Microsoft all use the ARM architecture. twitter.com/qinbafrank/status/...

Long-Term Agreements (LTAs) are changing the AI ecosystem. OpenAI's "Guaranteed Capacity" or similar enterprise-level capacity guarantee plans for B-end (enterprise customers) are essentially long-term LTAs. It's undeniable that LTAs are transforming the AI ecosystem. Previously, LTAs began in the upstream AI computing power supply chain: 1) Storage vendors required downstream chip customers to sign long-term supply agreements to lock in demand before expanding production (this is the core logic behind many investment banks' belief that it will change the valuation methods of storage vendors); 2) Nvidia signed long-term agreements with Lite and Cohr to prioritize the supply of key components such as lasers and optical interconnects in order to promote CPO (Consumer Product Ownership). 3) Large-scale model vendors offered numerous long-term computing power agreements to cloud providers to secure computing power. Now, OpenAI is extending this "long-term contract capacity lock-in" model from the upstream supply chain to the downstream enterprise end. This priority guarantee and discount lock-in mechanism, offered to enterprise customers against the backdrop of scarce AI computing power, offers significant benefits: 1) Addressing computing power bottlenecks and providing enterprises with certainty: By signing 1-3 year long-term agreements, enterprises can lock in discounted token prices and priority access rights (including cutting-edge models such as the GPT series, o series, and Agent). For enterprise customers, this means controllable budgets, lower costs, and protection from sudden rate throttling or price increases, making it suitable for large-scale deployment of AI Agents and productivity tools. 2) For large model vendors: Previously, OpenAI's enterprise revenue primarily relied on ChatGPT Enterprise + API pay-as-you-go billing. The LTA allows OpenAI to obtain predictable, high-return revenue in the long term (similar to LTAs for chip manufacturers), helping them plan for massive CapEx (capital expenditure) while securing large clients. Of course, the core significance is accelerating B2B penetration and seizing the enterprise AI market. OpenAI is heavily shifting towards the B2B market (Frontier platform, Deployment Company, Codex, etc.), and the proportion of enterprise revenue has significantly increased. LTA is a "moat" tool that binds OpenAI to leading companies, helping it compete against rivals like Anthropic and Google. Given its effectiveness, it's believed that Anthropic, Google, and others will quickly follow suit. LTA can be seen as a crucial step in commoditizing "scarce computing power," providing reassurance to companies and securing long-term cash flow and market dominance for large-scale model vendors. This also represents an important step towards further "infrastructure-izing" large-scale models. twitter.com/qinbafrank/status/...

Loading..