Good Morning, Asia. Here’s what’s making news in the markets:
Welcome to Asia Morning Briefing, a daily summary of top stories during U.S. hours and an overview of market moves and analysis. For a detailed overview of U.S. markets, see CoinDesk’s Crypto Daybook Americas.
Mistral AI recently offered a rare benchmark in the Artificial Intelligence industry’s environmental disclosure, detailing the footprint of its flagship large language model, Mistral Large 2.
Over 18 months, training and operating this model generated 20.4 kilotonnes of CO₂-equivalent emissions, consumed 281,000 cubic meters of water, and depleted 660 kilograms of antimony-equivalent materials, Mistral’s report said. Notably, a single 400-token response from its chatbot, Le Chat, uses just 1.14 grams of CO₂, 45 mL of water, and 0.16 milligrams of mineral resources.
But how does this compare to bitcoin’s carbon footprint? After all, bitcoin’s energy use has been the subject of significant debate and is often cited when establishing bans on bitcoin mining in jurisdictions.
That makes AI inference seem downright frugal compared to Bitcoin’s proof-of-work engine. On average, one Bitcoin transaction emits between 600 and 700 kilograms of CO₂, consumes more than 17,000 liters of water, and generates over 130 grams of electronic waste.
Zooming out, the entire Bitcoin network emitted roughly 48 million tonnes of CO₂ in 2023, according to the Cambridge Centre for Alternative Finance. It also consumed over 2 billion liters of water and produced more than 20,000 tonnes of e-waste.
However, the Cambridge Centre’s numbers, although peer-reviewed, have been the source of considerable criticism and require important caveats.
First, Bitcoin’s electricity mix is not monolithic.
According to a survey of miners conducted by BTC Investment fund Batcoinz as of March 2023, Hydropower (23.1%), wind (13.9%), and solar (5%) collectively account for more than 40% of Bitcoin’s energy consumption. The difference between the numbers is because surveys done by Batcoinz include off-grid generation.
Nuclear energy, often considered carbon-neutral, accounts for another 7.9%. Gas and coal together represent 44%, but Bitcoin’s energy profile is more diversified than critics often assume.
Second, LLMs may benefit from a cleaner grid by default. For example, nuclear energy comprises over 22% of the European Union’s electricity generation, which reduces the CO₂ emissions associated with model training and inference in EU-based data centers such as Mistral’s.
That advantage isn’t due to model architecture, it’s grid geography. A U.S.-based training run drawing from coal-heavy regions would present a very different environmental profile.
So while the marginal footprint of using an LLM is vastly smaller than processing a BTC transaction, both operate within infrastructure landscapes that significantly shape their true environmental impact.
Training frontier models like GPT-4 or Gemini can still require millions of GPU-hours and heavy water consumption, depending on location. Still, Bitcoin’s design, mining every 10 minutes regardless of demand, results in a fixed energy cost that scales with time, not usage.
In contrast, AI’s marginal cost scales with the frequency of model usage. That distinction makes the emissions from a chatbot reply easier to amortize than those from a block reward.
As global scrutiny increases over the environmental costs of computation, transparency initiatives like Mistral’s, provide important reference points.
While proof-of-work is energy-intensive, the Bitcoin blockchain’s halving mechanism steadily reduces the rate at which new coins are created, encouraging miners to become more efficient over time. Its environmental footprint should be weighed against the utility it provides in securing a decentralized, global financial network.
Continued improvements in clean energy adoption and mining optimization will be key for both BTC and AI as they scale into core pillars of the digital economy.
Market Movers:
BTC: Bitcoin is trading at $119,500, struggling to maintain momentum after last week’s all-time high of $123,100, as retail-driven sell pressure on Binance has pushed Net Taker Volume below $60 million and signaled growing bearish sentiment, according to CryptoQuant.
ETH: Ether has pulled back over 3% to $3,696 after a multi-week climb toward $4,000, as technical indicators flash red and analysts question whether the rally can continue without a broader correction, despite ongoing institutional accumulation.
Gold: Gold prices rose nearly 1% on Tuesday, with spot gold reaching a five-week high of $3,430.41 amid ongoing trade uncertainty and falling US bond yields, which continue to draw investor interest.
Nikkei 225: Asia-Pacific markets opened higher after U.S. President Donald Trump announced a “massive Deal” with Japan, lifting tariffs to 15% on Japanese exports, with the Nikkei 225 rising 1.71% at the open.
S&P 500: US stocks closed mixed Tuesday, but the S&P 500 edged slightly higher to a record 6,309.62 as investors weighed earnings reports