The author is a Reuters Breakingviews columnist. The opinions expressed are her own.

By Karen Kwok

NEW YORK, April 21 (Reuters Breakingviews) - If the artificial-intelligence frenzy unleashed by ChatGPT has an ethos, it might be this: scale is all you need. Since 2022, the chatbot’s designer OpenAI rushed to secure as much computing capacity as possible, sparking an industry-wide investment arms race. Chief rival Anthropic, though, is taking a somewhat cautious approach amid the mania. Boss Dario Amodei, like his nemesis Sam Altman, has signed a gigantic deal for servers and chips with Amazon.com AMZN.O. The details reveal contrasting visions of how crazed the AI race will get.

The pact announced on Monday might seem a standard AI mega-deal: Anthropic, last valued at $380 billion, pledgedto spend $100 billion on Amazon technologies over 10 years and secured enough computing grunt to consume up to 5 gigawatts of power.

Crucially, though, the deal spans a broad mix of servers and chips, where Amazon-designed silicon has become a cheaper alternative to gold-standard Nvidia NVDA.O products in certain cases. Amodei’s company will use older Trainium2 chips, alongside third- and fourth-generation models and options for future designs as they become available.

OpenAI’s deal, which in February added $100 billion to an earlier $38 billion partnership, plans to burn through its commitment over a shorter eight-year window while aggressively locking in only next-generation Trainium hardware.

When it comes to AI, capacity is money: without plentiful and powerful chips, a company can’t field user queries or train new models. Altman’s solution is brute force, announcing a $500 billion project and pushing partners into a deluge of spending. The company’s forecast annual training budget exceeds $100 billion in 2028 and 2029, the Wall Street Journal reported, dwarfing Anthropic’s projections. In a memo to investors, the company lambasted its arch-rival as compute-constrained, according to CNBC.

Indeed, Anthropic’s measured approach with Amazon may be shaped by the scarcity of computing power. Yet Amodei has long argued that while AI will advance rapidly, the time it takes to monetize breakthroughs may lag, making a too-early over-commitment on spending fatal.

That’s reflected in Monday’s deal. As part of the agreement, Amazon will invest up to $25 billion more in the startup, but $20 billion of that is at Amodei’s discretion to draw, sources familiar with the matter told Breakingviews, which could prevent equity dilution if unnecessary. Amazon's $50 billion pledge to OpenAI features a bigger $15 billion initial injection, with the rest triggering on milestones like a public listing.

Maybe Altman will prove right, and sheer force of scale will simply win out. If he’s not, though, the implications for Nvidia – and Amazon and Google, both racing to provide cheaper alternative silicon – are big. A world more focused on efficiency is one that’s just a bit less frothy.

Follow Karen Kwok on LinkedIn and X.

CONTEXT NEWS

Technology giant Amazon.com said on April 20 that it will invest up to $25 ​billion in Anthropic, as the artificial intelligence startup commits to spending more than $100 billion over the next 10 years on Amazon's cloud technologies.

OpenAI said on February 27 that it is raising $110 billion in a funding round that would value the ChatGPT maker at $840 billion. As part of the investment, Amazon committed up to $50 billion in funding, starting with an initial $15 billion and progressing as certain conditions are met.

Amazon and OpenAI also expanded an earlier deal to include a commitment to consume 2 gigawatts of computing capacity powered by Amazon's in-house Trainium chips, the companies said.