AI: Mega IPO Preps, Google AI Re-Org, ‘RAMageddon’-DC, & More. AI-RTZ #1179

1. Anthropic & OpenAI prep for their mega-AI IPOs: Anthropic and OpenAI are ramping up their preparations ahead of their planned mega-AI IPOs later this year and next respectively. Both have filed confidentially with the SEC. Anthropic, last privately valued around $965 billion, is rehearsing its roadshow story with institutions ahead of an anticipated October float, with early investors floating trial-balloon valuations of $2 trillion or more: the math being 30x revenues, the ‘incredibly low end’ against Palantir’s 55x, on $100 to $120 billion in annualized revenues by year-end, implying a $3 trillion upper boundary. It is even shopping on the way to the roadshow, in talks to buy Decart for about $6 billion. OpenAI, at a $40 billion-plus run rate that has roughly doubled since year-end, is rebuilding its revenue engine for the public markets: a second new chief revenue officer in under a year in Wiz’s Dali Rajic, senior departures from Brad Lightcap to Fidji Simo, about half its business now enterprise, and price cuts plus Codex momentum to close the AI Coding gap with Anthropic. More here and here.

   

2. Google puts AI Products ahead of AI Labs: Google is moving its global AI management from DeepMind London to California, post recent senior management changes, with Koray Kavukcuoglu, ‘KK’, running models, the Gemini app and developer teams as one product group reporting to CEO Sundar Pichai. The consensus scores Google on the frontier model-of-the-month game; my view remains that Google is playing a different one: efficient models scaling to billions with Gemini, and Google Cloud growing 82% as the company’s big focus, selling to the AI who’s who. A product, not a lab, is the right job description for this race. More here.

   

3. ‘RAMageddon’ lobbying reaches Washington DC: The global shortage of memory, aka ‘RAMageddon‘, is moving its lobbying forces to Washington DC in force, with everything from Defense Production Act allocations to onshoring subsidies on the table, a supply picture I mapped in ‘US hamstrung on Memory Chip Supply via China’. The receipts kept arriving this week: Google added $100 to every Pixel 11 while trimming RAM, Lenovo managed through by leaning into AI servers, and China’s state memory maker CXMT passed Tencent as the country’s most valuable company, through the roof on the memory boom. New memory supply is 2027 at the earliest, so near term the lobbying decides who waits, while consumers pay retail. More here.

   

4. AI Compute Demand Bull Case debates: The industry is debating the long-term supply and demand for AI compute, for both training and inference, and the latest estimates outline the bullish end of that spectrum. The ground-level data this week backed the near-term bulls: H100 rental prices up 50% in six months on a 2022 chip, with young AI ‘neolabs’ pushed into 3-to-5-year contracts just to train at all, while the neoclouds cash in on record pricing: Nebius’s first capacity auction cleared 15% above its highest-ever price, and CoreWeave, the ‘Nvidia rocket fuel’ story I profiled early among the ‘Neoclouds’, raised prices 25% in July with capacity effectively sold out. My view: near-term pressure is UP, tempered a few years out as hundreds of billions of buildout land as live capacity. More here and here.

   

5. Nvidia’s $500 billion Financing Move: Nvidia, led by founder/CEO Jensen Huang, has assembled a ‘Who’s Who’ financial consortium, the top of the institutional list with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR among the members, with plans to raise half a trillion plus dollars for AI infrastructure globally over the coming years. The ‘Kingmaker’ I have long described is now something closer to an ‘Uber AI Investor & Banker’, converting its trillion-plus backlog into financed, recognized demand across the ecosystem, with Stratechery’s risky-business framing a useful read on the trade-offs. It compounds the gigawatt-scale supply buildouts already underway across the industry, and this week’s neocloud pricing power shows why the financiers keep saying yes. Circular financing concerns deserve their airing; so does the sheer operator craft. More here.

   

Additional Items.

1. Mark Zuckerberg moves Meta back to Open Weight models: Zuck laid out a renewed open-weight AI vision for Meta in a 6,500-word essay, swinging the company back toward the open models flow it helped pioneer with Llama, and rejoining the US open source AI champions race where, as I have noted, open source ‘Rockets are Hard’. More here.

2. SpaceXAI joins the AI Coding/Frontier race with Grok 4.6: The new release benchmarks right up against the frontier leaders, making the case for Elon’s $60 billion Cursor purchase and the boundless AI ambitions laid out in the SpaceX/xAI IPO filing, all running on his gigawatt-scale compute buildout. More here.

(Additional Note: AI Ramblings is now a weekday Daily podcast called AI Ramblings Daily (ARD). Different content than AI-Reset to Zero (AI-RTZ), which remains a daily morning substack with now over 1170+ ‘MY TAKES’ on key AI events and issues turbulently flowing by. AI Ramblings Daily is typically a 20 minute afternoon podcast on my take on additional AI developments of the day. Both daily substack and podcasts typically discuss different AI issues and items. And are free to subscribe.

Try this week’s series below:

ARD 137, ‘Slow & Steady’, the New US AI Speed. OpenAI, Meta & Microsoft: OpenAI paces its IPO into early next year, Zuck swings Meta back to open weights, and Microsoft compounds quietly, tortoise-style.

ARD 138, ‘Gaming the System’ for AI. Nvidia, OpenAI & Anthropic: Nvidia’s $500 billion AAA Wall Street consortium, OpenAI’s $7 billion tender prunes its cap table, and Anthropic books Riot’s $9.1 billion of capacity.

ARD 139, ‘Going with the Flow’ in AI. SpaceXAI, AT&T & Apple: Grok 4.6 joins the frontier flow, AT&T routes 45 billion daily tokens toward open models, and Apple re-staffs its Washington DC operation.

ARD 140, ‘Through the Roof’ in AI Valuations. Anthropic, Cerebras & CXMT: Anthropic’s $2 to $3 trillion trial balloons, Cerebras beats and raises yet drops 16%, and China’s CXMT passes Tencent’s peak valuation.

ARD 141, ‘For Better or for Worse’ in AI commitments. OpenAI & Anthropic: OpenAI’s new revenue chief at a $40 billion-plus run rate, ChatGPT remembers your Mac, and Anthropic’s ‘First Lady’ profiled):

Tomorrow, Sunday, we’ll have The Bigger Picture.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)