ARK • Disrupt

Last Friday, growth in the supply of bitcoin “halved” for the fourth time in history, dropping from ~1.8% at an annual rate to ~0.9%, as shown below. As a result, bitcoin’s supply growth slipped below that of gold—estimated to be ~1.7% in 2023¹—on a long-term basis.

Source: ARK Investment Management LLC, 2024, based on data from Glassnode as of April 19, 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.

Thanks to bitcoin’s mathematically metered supply growth, the reward for mining a block halves every 210,000 blocks, which takes place roughly every four years. The first halving occurred in November 2012, reducing the reward from 50 to 25 bitcoins per block. Friday’s halving reduced the block reward from 6.25 to 3.125 bitcoins per block, as shown below.

Source: ARK Investment Management LLC, 2024, based on data from Glassnode as of April 19, 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.

The halving is critical to Bitcoin’s design, highlighting its predictable monetary policy and bitcoin’s role as a scarce asset. By reducing miner rewards by 50% every four years, Bitcoin lowers the supply growth of bitcoin on a predetermined schedule, guaranteeing that the total number of bitcoins outstanding will never exceed 21 million, its predetermined cap. In sharp contrast, central bank policymakers can adjust the supply growth of government-backed currencies idiosyncratically at will.

Currently, bitcoin’s circulating supply stands at roughly 19.7 million, as illustrated below. In other words, it is only ~6.5% below the 21 million at which it will be capped.

Source: ARK Investment Management LLC, 2024, based on data from Glassnode as of April 19, 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.

Since its inception 15 years ago, Bitcoin’s halving has taken place in the context of extended bull market runs. While past is not prologue, in the 12 months following the three previous halvings, bitcoin’s price appreciation has ranged from ~285% to ~8,478.2%, averaging ~3,108%, as shown below.

Source: TradingView.com. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.

Redwood Materials, a battery recycler founded by Tesla's former CTO, J.B. Straubel, boasts² a battery recycling run rate of 20 gigawatts (GWh) per year—enough to power ~250,000 electric vehicles, more than 14-fold the US industry’s battery recycling total four years ago.

Although modest relative to what the electric vehicle (EV) market will demand at maturity, Redwood Materials is scaling recycling rapidly with a method that slashes carbon emissions by 40-70% compared to traditional recycling processes. Because 95% of battery minerals are recoverable, recycling will become critical as EV batteries reach end-of-life.³ At scale capacity, therefore, recycling should reduce the need to mine for materials significantly.

Importantly, Redwood Materials’ recycling initiative could help shift the center of the supply chain for batteries from China to the US. For more insights, the full article⁴ on advances at Redwood Materials is worth the read.

Last week,⁵ Meta released two versions of its long-awaited Llama 3 family of open-source large language models (LLMs)—a small 8-billion parameter version and a more performant 70-billion parameter version. Meta also disclosed early performance metrics on its flagship 400-billion parameter version of Llama 3, which is still in training.

Meta has trained all Llama 3 models on 15-trillion tokens of training data—7.5x the number that trained Llama 2. Importantly, Meta found that performance continued to scale with greater amounts of training data for much longer than previous research suggested was possible.

Thus far, the benchmarking for Llama 3 has been impressive. The 70-billion parameter version scores 82.0% on massive multi-task language understanding (MMLU), a general knowledge benchmark, and 81.7% on HumanEval, a Python coding benchmark. Meanwhile, the 400-billion model—which should improve more by its release date—scored 86.1% and 84.1%, respectively.⁶ Those scores suggest that Llama 3 is the first open-source model powerful enough to challenge the performance of closed models like GPT-4 Turbo, which scored 86.7% and 88.2%⁷ on MMLU and on HumanEval, respectively, as shown below.

Source: ARK Investment Management LLC, 2024, based on data from a range of external sources, of April 19, 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.