Why AI may have just broken a 25-year-old computing rule

TOGETHER WITH*

A free newsletter with the marketing ideas you need

The best marketing ideas come from marketers who live it.

That’s what this newsletter delivers.

The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.

Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.

Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.

Sign Up Free

*Sponsor

TECH BRIEF

Welcome back to Front Research. Tuesday's news came down to who pays for AI, who builds it into products, and what it can now prove:

• Nvidia and AMD are putting a combined $4B into Meta's stock sale, which Meta lifted to $97.5B

• Apple launched M6 versions of the 14-inch MacBook Pro and iMac and added AI assistants built partly on OpenAI models to its creative apps

• OpenAI says its models disproved the Exponential Time Hypothesis, a decades-old pillar of computer science

Let's get into it.

Nvidia and AMD Buy $4B of Meta as the AI Stock Sale Grows to $97.5B

• Nvidia will invest $3 billion in Meta's equity offering and expand work on Meta's Muse models; AMD is committing $1 billion

• Meta sold $25 billion of common stock at $625 a share and $12.5 billion of 6.5% mandatory convertible preferred, with up to $60 billion more in nonvoting Class C shares to follow

• The total raise now stands at $97.5 billion, up from $95 billion announced Monday

• Meta will deploy more Nvidia Vera Rubin systems and Vera CPUs, and has begun testing AMD's Helios racks

• Why it matters: Meta's two biggest chip suppliers are now also its shareholders, which deepens the circular financing behind the AI buildout

Nvidia said it will invest $3 billion in Meta's equity offering, subject to closing, as the two companies co-optimize training and inference for the Muse model family. Meta plans to deploy Vera Rubin systems and Vera CPUs more widely for its next generation of AI agents, along with Spectrum-X Ethernet and BlueField-4 networking. Jensen Huang called agents "a new computing paradigm," and Mark Zuckerberg described Muse as "how we bring personal superintelligence to everyone." TechSpot notes that Meta lifted the overall fundraising target to $97.5 billion on Tuesday, and that the $60 billion follow-on program will run after the new Class C shares begin trading in early 2027.

AMD followed with its own check. According to Yahoo Finance, AMD will commit $1 billion to the offering, Meta has begun installing and testing AMD's Helios rack-scale systems, and Meta is an early partner on the MI450 architecture, including a custom accelerator. AMD shares rose about 5.4% on Tuesday while Meta slipped about 1.6%.

The terms show how much the raise is costing. A separate Yahoo Finance analysis reports that the $25 billion common tranche priced at $625, below Friday's close, and that the preferred converts in 2029. Meta's free cash flow turned negative last quarter for the first time since 2012, with capital spending of about $34 billion against operating cash flow of about $33 billion, and its five-year credit default swap spread hit a record 127 basis points on Friday.

Why it matters: Nvidia and AMD are effectively recycling part of Meta's chip spending back into Meta's balance sheet, which supports their own order books for 2027. For Nvidia and AMD holders, the question is how much future revenue rests on customers they also finance. For Meta holders, the company has moved from $45 billion a year of buybacks in 2021 to nearly $100 billion of dilution, and record CDS spreads show credit investors are pricing more risk too. Capex guidance on the next earnings call will set the tone for the whole AI supply chain.

Apple Ships M6 Macs and Brings OpenAI Models Into Its Creative Apps

• The 14-inch MacBook Pro with M6 starts at $1,999 and the 24-inch iMac at $1,499, both with no price increase; preorders open October 28 and shipping begins November 6

• Apple claims up to 30% faster graphics and 25% faster AI workloads than M5, plus up to 24 hours of battery life on the MacBook Pro

• Keynote, Pages, Numbers and Logic Pro gain beta AI assistants built on a mix of Apple and OpenAI models

• Keynote, Pages and Numbers also add Model Context Protocol support so outside AI agents can read and edit documents

• Why it matters: Apple is opening its productivity apps to third-party models and agents, a bigger strategic shift than the chip refresh

Apple said the M6 has a 12-core CPU and 12-core GPU with Neural Accelerators in each GPU core, and a dual 16-core Neural Engine that it says is up to 1.9x faster than M5's. Memory starts at 16GB and goes to 32GB, with bandwidth up to 170GB/s, and both machines get Wi-Fi 7 and Bluetooth 6. TechCrunch reports that the iMac comes in seven colors and claims up to 5x faster on-device AI than the M4 model, and that the launch came a day ahead of Apple's "Pro and Tell" event on Wednesday. Yahoo Finance notes that the Mac was a $33.7 billion business in 2025, out of $416.1 billion in total revenue, and that Apple is expected to show its first touchscreen MacBook on Wednesday.

The software news may matter more. Apple is adding beta assistants to Keynote, Pages and Numbers that work with document context, and one in Logic Pro that can set up tracks and refine projects, all built on Apple's own models plus OpenAI's. The same apps add MCP support, letting external agents read content, make edits and run actions. 9to5Mac reports that the features ship in November as a free update for Creator Studio subscribers, who pay $12.99 a month or $129 a year in the US.

Why it matters: the M6 refresh holds prices flat into the holiday quarter, which protects Mac margins and gives Apple a clean setup heading into its earnings report later this week. The more important signal is that Apple now relies on OpenAI models inside first-party apps and is supporting MCP, the agent standard that Anthropic created and OpenAI, Google and Microsoft have adopted. That makes iWork a target for any company's AI agent, which helps Apple's ecosystem stay relevant in agentic workflows but also concedes that its own models are not enough on their own.

OpenAI Says Its Models Disproved the Exponential Time Hypothesis

• OpenAI published three model-generated results: a disproof of the Exponential Time Hypothesis (ETH), a proof of Sogge's local smoothing conjecture in all dimensions, and a proof concerning four-dimensional quantum field theories

• Outside expert groups, which OpenAI calls "shepherds," will verify and write up the results; OpenAI says it will not seek academic credit

• MIT's Ryan Williams said a confirmed ETH refutation would be a major development and that experts are checking the proof

• An earlier OpenAI Navier-Stokes claim turned out weaker in its formally verified form than as announced

• Why it matters: if the proofs hold, AI labs move from selling productivity to producing original research, which supports the valuations now being set for OpenAI and Anthropic

OpenAI announced the results along with a new process for releasing AI-generated science through academic groups. According to Engadget, hundreds of papers in databases, computational biology and geometry rest on ETH, which assumes some problems cannot be solved much faster than brute force; if it is false, better algorithms may be possible across those fields. The release follows earlier October manuscripts covering progress on 372 math problems. For computer science, an external group that includes the scientists who proposed ETH will write up the result, while the physics group is holding back its materials to write its own interpretation.

Experts are interested but careful. Al Jazeera quotes MIT's Ryan Williams saying "we are working to verify and examine their proof and determine if it holds up," and adding that it would not produce near-term real-world applications. The University of Rochester's Alex Iosevich said OpenAI appeared to have "cracked" the local smoothing conjecture, calling it a "historic development." Gizmodo points out that human readers found the formally verified version of OpenAI's earlier Navier-Stokes result to be weaker than the published claim, and that OpenAI itself says its choice of shepherds "creates asymmetry of information."

Why it matters: the timing is not accidental. With Anthropic's IPO filing setting a $1.5 trillion to $2 trillion benchmark, OpenAI is making the case that frontier models are now generating new knowledge, not just automating tasks. Verified results would strengthen the argument for paying for scarce frontier compute and support demand for Nvidia, Broadcom and AMD hardware. If the ETH proof fails review, it adds to a pattern of overstated claims that investors and enterprise buyers will discount. The verification timeline, likely weeks, is worth watching.