America's Real AI Arms Race

• Below is the latest edition of w3w, my free newsletter about emerging technology, including AI, Bitcoin, Ethereum, and more . If you would like to receive it in your inbox every Sunday, please subscribe here.
• I used several AI apps to assist production of this edition of w3w. Final edit 100% by me. For fuller detail, see the newsletter's commit history on GitHub.

American AI companies' scramble for access to scarce chips and power is heating up:

• OpenAI partnered with Broadcom to build its first custom processor for AI use cases. (Reuters)
• Google's parent Alphabet announced a new $24 billion round of investment in AI infrastructure, most of it in India. (Wall Street Journal)
• Meta struck a pair of AI-related deals. Along with Blue Owl, it raised $30 billion to build a new data center in Louisiana. The social giant also tapped Arm Holdings to supply chips to handle AI recommendations across Facebook and Instagram. The deal includes a $1.5 billion investment to build a new AI data center in Texas.
• Analysts say the U.S. should be adding about 80 gigawatts of new power generation capacity a year to keep pace with the growth of AI and other tech applications. But the country has recently been building less than 65 gigawatts a year. As a result, AI data centers are increasingly building on-site generation as a workaround to the existing U.S. power grid. (Wall Street Journal)
• Salesforce announced it will invest $15 billion over five years to build a hub on its San Francisco campus to incubate adoption of AI. (Reuters)

To me, these sorts of headlines echo something Ken Auletta described in Googled, his excellent book about the search giant's early history.

Auletta recounts that founders Larry Page and Sergey Brin would often distinguish between "electron problems" and "atom problems" their company might face. Back in the aughts, those terms were a useful shorthand for things that could be fixed in software versus things related to computer hardware or other limitations of the physical world.

We should remember that same framework very much applies to AI now. And if you want to know where the broader AI boom's limitations might ultimately lie, perhaps start with the "atom problems."

Week in Review: Oct. 12–18, 2025

• China is surpassing the U.S. in producing powerful and popular open-source AI tools. So says the Washington Post's Kevin Schaul, based on an analysis of trends on LMArena, a site that uses blind tests to discover which AI outputs users prefer.
• California enacted new AI rules requiring companies to disclose when users are interacting with companion chatbots. The law also includes requirements to report mental health risks starting in 2026. (The Verge)
• Tempo, an on-chain payments platform backed by Stripe and Paradigm, raised a new $500 million venture investment round valuing the company at $5 billion overall. The deal is one of the biggest in the blockchain space in the last few years. (Fortune)
• Decrypt's André Beganski reports that recent volatility in crypto prices has sparked debate in the industry about whether decentralized finance platforms (DeFi) should implement market circuit breakers like the stock market to prevent runaway panics. Skeptics say such a system might be implausible to create on-chain versus on conventional stock exchanges.
• PayPal's stablecoin partner Paxos mistakenly issued 300 trillion (with a "t") of the dollar-pegged token PYUSD Wednesday. The incident had limited impact on crypto markets as Paxos quickly "burned" the excess tokens, erasing them from its blockchain network. But that didn't stop widespread speculation about how such a "fat-finger" mistake could be made in the first place. As part of its normal operations, it seems Paxos had originally intended to mint 300 million (with an "m") PYUSD tokens. Yikes. (Barron's)

Market Snapshot

A quick look at some major indicators as of Friday's close on Wall Street:

Looking Ahead

• Non-USD stablecoins are gaining momentum. Three major Japanese banks are planning to jointly issue a new stablecoin pegged to the yen... Separately, adoption of euro-pegged stablecoins has increased exponentially this year following Europe's implementation of a new regulatory framework for cryptoassets, according to a new report from Chainalysis.
• BTQ Technologies said it has successfully demonstrated a new quantum-safe method for verifying Bitcoin transactions. If their solution scales as hoped from here, it could solve a major security issue facing Bitcoin from the emergence of powerful next-generation computers. (Morningstar)
• MarketWatch's Jurica Dujmovic just wrote a normie-friendly explainer of tokenization, which he notes is now being applied to everything from real estate to U.S. Treasurys... Meanwhile, Britain's Financial Conduct Authority proposed rules that would enable asset managers to tokenize shares in their investment funds on public blockchains like Ethereum, with an eye toward attracting younger investors.

Odds & Ends

• Beyond Meat's sales and stock-market valuation have collapsed. The plant-based food company's stock closed below 65 cents a share on Friday, compared to an all-time high of $235. Since the pandemic in particular, Beyond has been beset by increased competition from Impossible Foods and a shift in consumer tastes toward animal protein. If Beyond's share price remains below $1 much longer, Nasdaq may even de-list the company. Ouch.
• Bodycam footage of police chasing a rogue inflatable jack-o-lantern in suburban Cleveland went viral.

That's it for now. Thanks for reading the newsletter today! If you want to know more about w3w's history and (ahem) the author, that info is available here.

Please note, I regularly use several AI apps to assist production of w3w. But the final edit is 100% by me. For fuller detail, see the newsletter's commit history on GitHub.

If you need to reach me directly, please email peter[at]w3w[dot]media.

Best wishes for a healthy and productive week ahead. 😊