AI Is Now Priced by Proof, From Memory Chips to Cloud Earnings

By Crypt0 · Oct 28, 2026

Artificial intelligence moved past promises toward proof this week, and the proof arrived in three different forms at once, earnings, governance, and mathematics. A memory maker posted a record quarter built on chips that sit beside AI accelerators, two cloud giants reported the same AI demand with very different bottom lines, a leading lab told California how courts should handle powerful models, and two rival labs claimed a breakthrough in theoretical computer science. The common thread is accountability. Buyers, regulators, and researchers are all asking the same question in different language, show the work, show the numbers, and show the safeguards, and the companies with clear answers are setting the pace for the whole field.

SK Hynix supplied the most striking earnings proof on Wednesday, reporting third quarter revenue of 104.62 trillion won, up 327.9 percent from a year earlier, and operating profit of 83.33 trillion won, up 632.1 percent from a year earlier, for an operating margin of 79.7 percent. The South Korean memory maker said high bandwidth memory demand from AI accelerators drove the result, and it expects its newer HBM4 memory to pass the older HBM3E in shipment volume and revenue during the fourth quarter. Long term supply agreements with 14 major customers already cover more than half of planned production capacity for 2027 and 2028, and the company expects demand to run ahead of supply through 2030. Shareholder returns of about 100 trillion won for 2026, including a 14 trillion won third quarter dividend, complete a picture of a supplier turning AI demand into cash, contracts, and capacity plans at the same time.

Microsoft and Alphabet then showed how the same cloud demand can land differently in two earnings reports on the same evening. Microsoft reported fiscal first quarter revenue of 92.3 billion dollars, up 19 percent from a year earlier, with Azure revenue up 48 percent to 33.2 billion dollars and total Microsoft Cloud revenue up 30 percent to 63.9 billion dollars. Alphabet reported third quarter revenue of 127.6 billion dollars, up 25 percent from a year earlier, with Google Cloud revenue more than doubling to 30.7 billion dollars, while earnings per share landed at 2.12 dollars against a consensus estimate of about 3.02 dollars, shaped by an 8.7 billion dollar loss on equity securities. The split is instructive for anyone building with AI tools. Cloud growth is strong across the board, and the market now weighs that growth alongside margins, cash flow, and capital spending, with each company earning its valuation one quarter at a time.

Governance took a concrete step forward in California, where Anthropic told Governor Gavin Newsom that government officials should be able to seek court orders pausing the release of AI models that pose a serious public safety concern, according to an October 15 letter confirmed as authentic by an Anthropic spokesperson. The letter followed a September panel of artificial intelligence experts convened by Newsom to develop safety proposals, and OpenAI sent its own recommendations on October 20. Anthropic also argued for targeted interventions over whole model shutdown switches, favoring steps such as disabling a specific AI agent or ending a specific customer's access to a model, on the grounds that precise tools work better in practice. The proposal matters because it moves safety from abstract debate toward a court tested process, with evidence, review, and a defined remedy, and that kind of structure gives builders a clearer path to ship powerful systems with public trust intact.

The most eye catching claim of the week arrived from theoretical computer science, and it deserves careful handling as an unconfirmed result awaiting independent review. OpenAI and Anthropic each announced algorithmic work challenging the Exponential Time Hypothesis, with machine checked Lean formalizations backing their claims, and OpenAI reported confidence that current cryptographic systems remain intact under the new algorithms. Both labs trace their breakthroughs to October 16, and both present deterministic subexponential algorithms for counting satisfying assignments of complex formulas, a result that would reshape how researchers classify hard problems if it holds up. Independent theorists still need to inspect the definitions, reductions, and formal proofs, so the honest description today is a striking unconfirmed claim from two competing labs, published openly for checking, rather than a settled theorem. That openness is itself a healthy sign, since science advances fastest when bold claims meet public scrutiny.

For readers, the week offers a practical way to judge AI news going forward. Look for revenue tied to real products, contracts tied to real capacity, governance tied to real process, and research tied to real proof that others can check. Memory earnings show the infrastructure build is generating cash, cloud earnings show the market rewarding discipline alongside growth, California shows safety policy moving toward courts and targeted remedies, and the mathematics claim shows frontier labs willing to publish work for the whole field to test. The builders who thrive in this phase will be the ones who treat proof as a habit, shipping numbers, safeguards, and open results, and inviting the world to verify each one.

Quick answers

What is this story about?

Why does this story matter?

Sources

New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.