#383: Is SpaceX’s Domination Of Transporting Mass To Orbit Just Getting Started?, & More
1. Is SpaceX’s Domination Of Transporting Mass To Orbit Just Getting Started?
This year, SpaceX has launched roughly 447,000 kgs into orbit with just 43 launches, accounting for ~80% of all spacecraft upmass and nearly 4x that of all other space orbit launches combined. Annualizing performance from the first half of the year, Starship could launch a total of 900,000 kgs into orbit.[1]
Now the question is if SpaceX’s “supply will create its own demand”[2] for future launches. Low-earth orbit satellites last ~5 years,[3] so SpaceX must have launch capability to replenish the constellation as it grows. If Starship were to launch every other day, SpaceX could create and maintain a constellation of 73,000 satellites in orbit, nearly twice as many as its initial goal of 42,000 satellites.
We would not be surprised if SpaceX scales the number of satellites in orbit beyond its current plan to meet both consumer and government demand.
2. Financial Institutions Are Making Strategic Decisions Around Stablecoins
Last week, Visa expanded[4] a pilot program to integrate stablecoins into its payment network. Building on its work[5] with Crypto.com—which leveraged Circle’s USDC to circumvent slow currency conversion processes and avoid international wire fees—Visa is hoping that USDC[6] will accelerate processing and settlement times with two merchant acquirers, Worldpay and Nuvei.
Public blockchain infrastructure enables the direct transfer of value over the internet without traditional intermediaries. Likely to disintermediate its own legacy card network, Visa’s integrations with blockchain payment infrastructure should limit the damage to its payments empire as it harnesses the technology available in the new payments world.
Visa’s moves suggest that the underlying demand for stablecoins is strong, even though supply has dropped 32% since peaking at $180 billion in 2022[7] following the collapse of algorithmic stablecoin UST and the bankruptcies of crypto-native lenders like Celsius and BlockFi who leveraged stablecoins in their yield products. Following 2022’s leverage cycle unwind, those seeking yield have found higher interest rates in short-term treasuries than in decentralized finance.
Stablecoins are gaining significant traction internationally. Circle recently forged[8] a partnership with Mercado Pago, Mercado Libre’s digital wallet provider in Latin America, to offer its Chilean customers access to USDC. As a hedge against hyperinflation, the demand for stablecoins in countries like Argentina[9] and Turkey[10] also is particularly striking.
In contrast, regulatory uncertainty in the US continues to stymie the progress of stablecoins. Recently, after a bill[11] specifying a regulatory framework for stablecoins advanced in the House of Representatives, top Democrats derailed it.
3. Meta Continues To Advance Open-Source AI
Building upon its library of open-source AI models, Meta recently released[12] an open-source expert AI coding assistant, Code LLaMa,[13] the performance of which is roughly equal to OpenAI’s GPT 3.5. To achieve those results, Meta fine-tuned its newest open-source large language model, LLaMa 2, on a code-specific dataset, and then reinforced it with synthetic data generated by the model itself.
Not only is Code LLaMa available free for commercial deployment, but the computational inference costs to use it are much lower than competitive models. As a proxy for computational operating cost, the largest Code LLaMa model’s 33 billion parameters are roughly 20% the reported size of GPT-3.5 and 5% the size of GPT-4.[14] Nonetheless, the highest performing publicly available version of Code LLaMa scored competitively on a standard AI coding benchmark, answering 54% of questions on the first try compared to GPT-3.5 at 48% and GPT-4 at 67%.[15]
Meta is pushing an open-source AI strategy and could be poised to disrupt companies seeking to monetize large language models. Why pay $10 per month for Github CoPilot[16] when a capable coding assistant like Code LLaMa runs locally for free? While allowing startups to experiment with and optimize around Meta’s AI architecture, Meta’s licensing terms prevent mega-cap tech companies from co-opting its models. Ultimately, the strategy should benefit Meta, as it fine-tunes its open-source models—trained on publicly available data—with proprietary data. Moreover, the open-source community likely will build tools around Meta’s architecture, creating a virtuous performance cycle for its in-house systems.
[1] Bryce Tech. 2023. “Global Orbital Space Launches Q2 2023.”
[2] See Say, J. P. 1803 (1971). “A Treatise on Political Economy, Or, the Production, Distribution, and Consumption of Wealth.” August M. Kelley, Publishers. See also The Investopedia Team. 2022. “Say’s Law of Markets: Theory and Implications Explained.” Investopedia.
[3] Pultarova, T. and Howell, E. “Starlink satellites: Everything you need to know about the controversial internet megaconstellation.” Space.com.
[4] Visa. 2023a. “Visa Expands Stablecoin Settlement Capabilities to Merchant Acquirers.”
[5] Visa. 2023b. “By settling in USDC, Crypto.com is setting a new course.”
[6] USDC, or USD Coin, is a digital stablecoin pegged to the United States dollar.
[7] See CoinGecko. Stablecoins by Market Capitalization. Data as of 09/09/23. See also Federal Reserve Bank of New York. 2023. “Runs On Stablecoins.” Liberty Street Economics.
[8] Circle. 2023. “Circle Teams Up with Mercado Pago to Introduce USDC to Chile Customers.”
[9] Engler, A. 2022. “Argentines Take Refuge in Stablecoins After Economy Minister Resignation.” Coindesk.
[10] Szalay, E. and Yilmaz, U. 2023. “Turkish Investors Looking for Haven Turn to Stablecoin Tether.” Bloomberg.
[11] Lang, H. 2023. “US congressional committee advances stablecoin bill.” Reuters.
[12] Meta AI. 2023. “Introducing Code Llama, an AI Tool for Coding.”
[13] Meta AI. 2023. “Introducing Code Llama, an AI Tool for Coding.”
[14] Brown, T.B. et al. 2020. “Language Models are Few-Shot Learners.” arXiv. Patel, D. and Wong, G. 2023. “GPT-4 Architecture, Infrastructure, Training Dataset, Costs, Vision, MoE.” SemiAnalysis.
[15] Meta AI. 2023. “Introducing Code Llama, an AI Tool for Coding.”
[16] GitHub Docs. 2023. “About billing for GitHub Copilot.”