AI: Reset to Zero: Weekly Roundup

Nvidia Reigns On, Meta-AI IPOs File, Google I/O Consumer AI Roadmap & More. AI-RTZ #1095

Michael Parekh — Sat, 23 May 2026 05:02:08 GMT

1. Nvidia Q1 2027 Reigns On — Sans China:

Wednesday’s earnings report from Nvidia did its now-customary thing — beat the highest analyst estimates, then raised the next quarter’s guidance even higher. Revenue grew 85% to $81.6 billion in the three months ending April 26, and management projected another 95% growth in the current quarter — the fourth consecutive quarter of accelerating sales growth. As The Information put it, “the AI boom is far from peaking.” All this without (sans) China, where another $10+ billion in revenue likely sat waiting in another universe.

The story this quarter wasn’t just the headline numbers. It was the CPU callout — Jensen Huang flagged $20 billion in standalone CPU sales pace for the year (separate from the GPU-attached CPUs in Nvidia systems), with a $200 billion+ new addressable market sitting on the other side of the AI Reasoning and Agent transition. The Vera Rubin platform — Nvidia’s own ARM-based CPUs paired with Rubin GPUs — is now ramping for the second half of 2026, with meaningful volumes by Q1 2028. Add the 86% jump in free cash flow to $48.6 billion, $20 billion of Q1 buybacks and an authorization for another $80 billion more, and Nvidia’s $5.4 trillion market cap still trades at just 26.7x forward earnings.

It’s also the kind of quarter that makes the depreciation skeptics revisit their assumptions — rental prices for Nvidia chips that launched three to six years ago rose 15-20% so far this year. And it puts the Frenemies dynamic on full display — Anthropic was singled out as a relatively new major Nvidia customer despite running on Amazon Trainium and Google TPUs. The customers’ customers’ biggest customers are all still buying. More here.

2. Mega-AI IPO Filings Officially Begin — SpaceX/xAI, OpenAI, Anthropic:

The week the Mega-AI IPO Mega-Wave I’ve been writing about for over a year went from preview to public record.

SpaceX/xAI dropped its first round of S-1 filing detail, with Reuters reporting a target of June 11/12 pricing on Nasdaq — ahead of Elon’s birthday at the end of June. Implicator AI ran the sharpest read on the financials — SpaceX has Starlink revenue, the prospectus makes xAI the bill, $37+ billion in cumulative losses to date. The WSJ added the structural filing ‘secrets’. OpenAI confidentially filed for September 2026, as the WSJ pinned the timing and CNBC confirmed the confidential S-1 route. Anthropic is reportedly eyeing October on the back of what the WSJ called “mind-blowing growth” — nearly $11 billion in Q2 revenues and a $50 billion ARR by year-end (CNBC detail). The WSJ tied them all together in its umbrella framing piece.

The three are not equally positioned going in. SpaceX/xAI is first out of the chute on what is arguably the most narratively-driven IPO of the AI Tech Wave — Elon’s “financial anti-gravity machine” of XAI-into-Twitter-into-XAI-into-SpaceX-into-Starlink consolidation, plus insider-sale carve-outs ahead of the traditional 180-day lockup, plus Nasdaq positioning for early Passive Index inclusion. The ‘We are the Geese’ dynamic plays out for early SpaceX/xAI investors, helping Elon keep his ‘no loss’ promises despite the entity being the least well fundamentally positioned of the three. OpenAI has the most prep work ahead — especially on the consumer side vs Google Gemini, on developer/enterprise catchup vs Anthropic, and on the still-unresolved OpenAI Smartphone roadmap. Anthropic — the tortoise — is best positioned of the three on operational momentum, even though the loudest media coverage is going to the other two. More here.

3. Google I/O 2026 Cements Google’s Consumer AI Lead:

Google I/O 2026 was the inflection point: Google Gemini is becoming Google Search. And AI Search is becoming Google Search. Over 3 billion people use Google Search worldwide every day — vs roughly 900 million weekly users of ChatGPT and a comparable weekly figure on Google Gemini today. The daily-usage gap is enormous, and Google just wired Gemini directly into the same Search box those 3 billion already open every day. The NY Times framed it as the first Search box redesign in 25 years; The Verge called it the Search box that does everything.

What Google actually shipped is the four-piece consumer-AI stack — Gemini 3.5 Flash today, Pro soon going up against Anthropic 4.7 and OpenAI 5.5, paired with Antigravity 2.0 on the developer/coding side; Gemini Omni for any-to-any multimodal; Gemini Spark as the 24/7 agentic AI surface going head-to-head with Claude Code/Cowork and OpenAI’s Codex; plus a Gemini app redesign in the new “Neural Expressive” UI/UX and a DeepMind-shown “AI Cursor” on the agentic-pointer surface. Sundar Pichai dropped the headline number that got the analyst desks moving — Google is now processing 3.2 quadrillion tokens per month, up from 480 trillion a year ago and 9.7 trillion two years ago. That’s roughly 6-to-7x year-over-year acceleration on an already-enormous base, all leveraged by the immense TPU-powered Google Cloud AI data centers Google is investing roughly $190 billion in capex this year alone.

MP Take: Of the consumer-AI giants, Google is the one with both the daily-distribution surface (3 billion+ Search users) AND the model-tier vertical stack (Gemini 3.5, Omni, Spark, on TPUs). Nobody else has both. OpenAI has model momentum but no distribution surface near 3 billion. Anthropic has model quality and enterprise traction but a smaller consumer footprint. Apple has the device distribution but is using Google Gemini for the AI layer. It’s the opportunity Microsoft had vs Netscape in the 1990s Internet wave — and far more — without the regulatory issues that constrained Microsoft then. And with a market cap approaching Nvidia’s leading five-trillion-dollar number. More here.

4. Anthropic the Quiet Winner plugs on:

Three independent data points this week make the case that Anthropic — the third of the trillion-dollar IPO trifecta — is the quietest but most consistent operator going into the mega-AI IPO season.

First — the legal-environment shift. The jury delivered a verdict in Elon Musk vs Sam Altman / OpenAI after just two hours of deliberation. Musk lost on the contested counts; OpenAI won the narrow legal case. But the reading the rest of the industry missed: Axios wrote that “in Musk v Altman trial, the entire AI industry lost”, and while OpenAI was defending its legal foundations and Musk was running through years of internal discovery, Anthropic spent that same time shipping product, hiring talent, and signing enterprise deals. Net-net, Anthropic gets to skip the entire reputation-damage discount that both Musk and OpenAI now carry into the rest of 2026 — a useful PR win for Musk ahead of his June 12 IPO pricing, sure, but a structural advantage handed to Anthropic by way of the calendar.

Second — the subtle Mythos product-policy shift. The WSJ reported that Anthropic is opening up the Mythos cybersecurity surface to let Mythos users share cyber threat data with other users — a small policy shift on paper, but structurally meaningful as Mythos moves from sealed-preview into wider commercial availability. Less front-page drama than OpenAI is generating with GPT 5.5 ‘Spud’; more compounding revenue ramp.

Third — the talent flow. Uber-coder Andrej Karpathy joined Anthropic in an AI Research role — co-founder of OpenAI, former Tesla AI Director, creator of the most-watched AI educational content on the internet, including his “Let’s Build GPT From Scratch” lecture series. The structural read: Karpathy specifically did NOT join xAI and did NOT rejoin OpenAI. Talent flow is the leading indicator in this race, and Karpathy is one of the few AI researchers whose presence alone meaningfully moves a lab’s research velocity. He’s chosen Anthropic. That is a strong signal.

The two persistent issues with Anthropic remain — Dario Amodei’s anti-China lobbying posture, and his strident Job Loss Fear discussion. Both are about the public-facing voice rather than the operating layer. Anthropic the company keeps doing the right things on Research, People, Product, and Markets. Anthropic the public-facing voice — especially the CEO’s public-facing voice — keeps getting in the way of its own story. More here and here.

5. OpenAI’s Pre-IPO To-Do List — Rolling ‘Code Reds’ & Brockman Officially In:

Wired reported the latest OpenAI reorg — Greg Brockman officially takes control of OpenAI’s product strategy, in addition to his AI infrastructure work. The headline change: OpenAI is unifying ChatGPT and Codex into one core product experience, alongside the developer-facing API. Brockman’s memo to staff: “We’re consolidating our product efforts to execute with maximum focus toward the agentic future, to win across both consumer and enterprise.”

The downstream personnel changes are telling. Thibault Sottiaux — who built Codex into one of OpenAI’s fastest-growing products ever — takes over the core product and platform teams, and is one of the leaders overseeing development of OpenAI’s forthcoming “super app” that aims to combine Codex, ChatGPT, and the Atlas web browser into a unified desktop application. Nick Turley — who led ChatGPT since launch and grew it to 900 million weekly active users — moves to lead enterprise products. Ashley Alexander, a former VP of Instagram, takes over the consumer product unit. CEO of AGI Deployment Fidji Simo remains on medical leave but expects to return.

The pattern is clear. OpenAI has had several “Code Red” catalysts ahead of its IPO target — starting with the Google Gemini-driven Code Red last December, and now driven by Anthropic’s developer-and-enterprise momentum with Claude Code and Codex. The latest reorg trims the side-quests and reorients the whole company around the agentic super-app thesis — AI Reasoning, AI Agents, on the way to whatever AGI looks like down the road. And in the meantime, get to a September 2026 IPO ahead of Anthropic in October, but behind SpaceX/xAI in June. The line between consumer and enterprise — and between ChatGPT and Codex — is blurring at both OpenAI AND Anthropic at the same time. More here.

AI: Mag 5 AI Spend, Cost & Price Hikes, Elon's OpenAI trial... AI-RTZ 1074

Michael Parekh — Sat, 02 May 2026 05:02:51 GMT

Mag 4 of 7 Report Strong Quarters & AI Capex: The four big techs, Amazon, Meta, Microsoft, and Google all reported strong quarters. And renewed AI data center capex spend for 2026 and beyond. Amazon remains highest at $200 billion, with Microsoft and Google at $190 billion each, followed by a $145+ billion upper range from Meta, vs $135 billion previously. Microsoft explained that $25 billion of its $190 billion number is due to rising component prices. This provides a 10%+ inflation bump from supply chain constraints for chips and other components that is a useful proxy for all the rising capex numbers. We already know that memory chip supply constraints and price increases have been hampering both cloud and local compute providers for months now. Markets responded positively to Google Cloud and Gemini strength in particular, while selling off a bit on Meta, and others. The report points to more of the same AI Infrastructure spending thread for Q2 in July and the rest of the year. More here.

Apple Q2 sees Strong iPhone & Mac demand: Apple CEO Tim Cook cited strong demand for the iPhone and other parts of Apple’s ecosystem, in the company’s second quarterly report. Services business, the second major revenue line after iPhone, is also on a strong trend, followed by Macs and other lines of the Apple product lineup and strategy. In particular, the Mac Mini is seeing multi-month delays due to strong demand for OpenClaw inspired AI Agent deployments, and other memory and chip component shortages. Apple is raising Mac Mini prices, due to supply and pricing constraints, and in the face of AI Agent driven developer demand. Also popular with lengthy delays are orders for Apple’s new iPhone chip macbooks, the Neo line. As I outlined earlier, this represents a new ‘value’ line opportunity for Apple relative to its historic financial success with its line of computers, smartphones and wearables. Apple remains poised to roll out a Google Gemini powered Siri and other product upgrades later this year. Particularly after the June WWDC Developer conference, which should see upgrades of its core operating systems and other software. Apple is also shifting its long standing cash and debt strategy to more cash given current environment. More here.

OpenAI’s ‘Stormy Weather’: While OpenAI continues to roll out upgrades to its LLM AI products with ChatGPT 5.5 and ChatGPT Cyber, it seems to be missing some growth and financial metrics ahead of a late 2026 IPO. OpenAI also had other related product enhancements and expansions around its Codex product for developers. Especially as its being rolled out beyond Microsoft to Amazon AWS and beyond (see next item). This despite some product glitches that are being addressed. However, OpenAI remains playing catch up to Anthropic’s momentum with Claude Code and Cowork in the developer and enterprise markets. Despite the ‘stormy’ headwinds, OpenAI remains on track with its product upgrades and distribution plans beyond Microsoft Azure, to other partners like AWS. Also, OpenAI continues to make strides expanding its AI Data Centers via partners like Oracle, CoreWeave and others, with a compute infrastructure that is growing larger and faster than arch-rival Anthropic. More here.

Microsoft/OpenAI New Deal: OpenAI and Microsoft finally recast their longstanding partnership with a revised operational and financial roadmap on working together. Key provisions include removing the AGI threshold for the end of the partnership in 2032, and redoing the revenue share agreements between the two companies. The amended deal opens the door for OpenAI to sell its AI products via other cloud companies like Amazon AWS and others. Microsoft keeps getting a share of OpenAI’s revenue until 2030 regardless of whether AGI is achieved. Microsoft also retains the rights to use OpenAI’s models and products until 2032, but the IP rights will no longer be exclusive. Microsoft also does not have to share 20% of its OpenAI related sales, while getting a piece of OpenAI revenues from other cloud providers. That toll arrangement is a key provision of the revised deal with no more AGI time-gated threshold. Also, Microsoft does not get a piece of any OpenAI/Jony Ive consumer AI products coming down the line. This includes a possible OpenAI Smartphone in the early planning stages. There are a lot more details to the deal that are worth noting, but overall it provides key ‘wins’ to both sides to move ahead with other partners and providers. It’s also timely for OpenAI, as it gets its business, financial and operational metrics focused on a late 2026 IPO. More here.

Elon’s OpenAI ‘showcase trial’: The week also saw the launch of a month long Jury trial led by Elon Musk against OpenAI, over the evolution of its journey from non-profit to for profit. The deal is a ‘showcase’ trial driven by Musk, designed to maximize positive PR for Elon’s SpaceX/xAI IPO this June, and negative repercussions for OpenAI’s Sam Altman led IPO efforts for year end 2026. A low probability verdict against OpenAI could reshape who controls OpenAI beyond Sam Altman and Greg Brockman. The case stems from a Musk 2024 lawsuit accusing both the OpenAI founders and Microsoft of betraying the company’s original nonprofit mission to “benefit humanity”. The initial days of the trail have been a bit rocky for Musk, with testimony that came across as ‘unfocused and uncharming’. Musk ended up admitting claims that xAI distilled OpenAi models for its AI model Grok’s training, as being ‘partially true’. The Judge also barred off AI existential debates in the trial, a direct nod to Elon using his 200 million plus X followers to cast himself as ‘AI’s good guy’. It’s going to be a long four weeks. More here.

AI: Apple's 2nd post Steve Jobs CEO, Elon's revealing SpaceX/xAI IPO filing, Anthropic passes OpenAI to $1+ trillion valuation, & More. Weekly Summary. AI-RTZ 1067

Michael Parekh — Sat, 25 Apr 2026 05:02:51 GMT

Apple turns to Ternus as 3rd CEO post Cook & Jobs: As expected, 50 year old Apple and its CEO Tim Cook, 65, officially appointed hardware head and 25 year Apple veteran John Ternus, 50, as the second CEO after founder/CEO Steve Jobs. John takes on the role September 1st, as Tim Cook then moves on to the Executive Chairman role, where he’ll be able to be the Apple global diplomat to heads of state here and abroad on Apple critical issues. Apple also promoted Apple Silicon maestro Johny Srouji into Ternus’s role, expanding it to Chief Hardware Officer. Together these moves cement the hardware side of Apple’s business, taking the baton from the supply-centric global behemoth that Tim Cook built post Steve Jobs in 2011. Those 15 years saw an Apple stock appreciation of close to 2000%, making it one of the best investments ever made by non-techie investor Warren Buffett for Berkshire Hathaway. Similar performance by the new CEO Ternus over the next 15 years would imply an Apple valuation of over $80 trillion vs its current $4ish trillion. An unlikely outcome even in an AI fueled age, when the global economy today is around $125+ trillion. But there’s headroom. More here.

Anthropic & OpenAI grind-on pre IPO with Opus 4.7 & OpenAI ‘Spud’ GPT 5.5: The week saw a flurry of new LLM AI models from Anthropic with Mythos and its cybersecurity challenges, and its Opus 4.7, an upgrade from Opus 4.6, which has been powering forth its world beating Claude Code and Claude Cowork coding and productivity AI software applications. OpenAI also released it’s long awaited ‘Spud’ model, now officially known as GPT 5.5. Both Anthropic’s Opus 4.7 and OpenAI’s GPT 5.5 are initially shown as highly competitive with each other on various key benchmarks. The pricing for both sets of models is higher in AI tokens against preceding models, underlining the ability of these vendors to adjust the a la carte pricing away from ‘all you can eat’ subscription pricing for the latest models. Especially given the current supply constraints of fast ramping AI data centers. Enterprise customers are rapidly coming to terms with these pricing shocks. Uber’s CTO is a recent example, citing how fast its developers are crashing through the company’s AI budgets, with the current ‘tokenmaxxing’ culture amongst developers. Expect this ‘velvet rope’ pricing dynamic to continue to play out this year, as both Anthropic and OpenAI plan mega-AI IPOs later this year. Anthropic is already crossing a trillion valuation in private secondary markets given its momentum with Claude Code in particular. More here.

Google flexes with new TPUs vs Nvidia et al: Google continues to improve its core vertical tech stack AI TPU (tensor processing unit) chips, which provide one of the main competitive offerings against Nvidia’s global market leading AI GPUs chips and attendant infrastructure architecture. Google, which has used TPU based AI data centers for its own global infrastructure, is making available its TPU infrastructure to third parties like OpenAI, Meta and others. It’s #3 ranking Google Cloud unit headed by CEO Thomas Kurian, is vocal on how Google plans on building TPU powered AI infrastructure vs not just Nvidia, but other competing chips architectures from #1 cloud provider Amazon AWS and others. Microsoft, the #2 cloud providers with Azure also have its own chip architecture in the works. Google’s TPU infrastructure gives it an edge in its global distribution deals for Google Gemini, with partners like Apple and others around the world. More here.

SpaceX/xAI IPO filing reveals xAI wagging the dog: The SpaceX/xAI confidential S-1 IPO filing highlights the boundless AI ambitions of founder/CEO Elon Musk. Across the range of businesses beyond SpaceX/xAI with its Starlink unit, and into the AI priorities of sibling public entity Tesla. Revelations include the open-ended AI capex investments needed for AI Data Centers for SpaceX, far beyond the needs of just its core rocket launching and Starlink internet communications businesses. Elon executed a creative $60+ billion merger with AI Coding company Cursor in 12 months, with a $10 billion walkaway fee if no merger is concluded by then. This is to accelerate xAI’s multi-gigawatt data centers with circular ‘paying customer’ Cursor, as it ramps up its global AI coding software business against Anthropic’s Claude Cowork, OpenAI’s Codex, Microsoft’s Github Copilot, and other entrants. All this makes it challenging for investors to assess the SpaceX/xAI IPO against its intended $1.75+ trillion valuation expectations. More here.

DeepSeek drops next open source LLM AIs: As expected, DeepSeek released its new flagship models V4 Pro and V4 Flash, nearly matching state of the art closed US models by about 3-6 months. With globally competitive AI token pricing. At 1.6 trillion parameters, it’s one of largest open source models, with both versions having 1 million token context windows.M This after DeepSeek seared into global attention early last year. Chinese AI companies private and public race to take a global lead in open source LLM AIs. Wresting away that crown from Meta with Llama back then. Meta under founder/CEO Mark Zuckerberg of course choosing to go its own direction with mega AI Talent hires, and ramped up AI Data Center spend to over $135+ billion a year. More here.

AI: Anthropic & OpenAI inch forward to IPOS, & more. Weekly Summary. RTZ #1060

Michael Parekh — Sat, 18 Apr 2026 05:02:41 GMT

Anthropic’s Opus 4.7 Pitstop & Pricing changes before Mythos: Anthropic is following up its cybersecurity concerns around its ‘peek-a-boo’ reveal of its next gen Mythos model, with Opus 4.7. It’s a small upgrade to its Claude Code and Cowork driving Opus 4.6. While it beats the earlier model on some benchmarks, there are anecdotal complaints that the model Claude feels ‘different’ vs a few days ago. Speculation ranged from a ‘nerfing’ of Claude ahead of a full Mythos release, to possible AI Compute infrastructure shortage that both Anthropic and sibling rival OpenAI are facing in the near term. Mythos, now with its Glasswing cyber security tour and testing with a small cohort of corporate and government partners, could likely take a few weeks or months. In the meantime, Anthropic continues to navigate to its mega-AI IPO by year-end. In that context, the company changed pricing to bill firms based on ‘a la carte’ AI use, as demand for its AI Agent products jumped. Anthropic working hard to become Coke in this picture. More here.

OpenAI tests Ad Pricing: OpenAI is trying out various pricing schemes for ads on its global ChatGPT franchise. This is in addition to its API enterprise pricing for its various models. Ads are a revenue line that the company is modeling to grow from $2.4 billion in ad revenue in 2026, to $11 billion by 2027, growing to over $100 billion by 2030. That would make it a strong, global ad business after Google, Meta, Amazon and others. The current ChatGPT ads pilot is extended, as OpenAI is testing early advertisers’ budgets slowly. Across less than 5% of its global ChatGPT inventory. Early marketer feedback indicated difficulty in figuring out if the ads are working. And stress the need for stronger tracking and measurement tools. As well as ad system management tools. Also, there is a demand for more inventory, for more robust trials and testing. Also in flux is what the ads will look like, with so far adding short headlines and a small image adjacent to AI results. The current experiments build on traditional online advertising, with presumably more AI specific innovation to come. Separately, OpenAI also launched GPT-Rosalind an AI model for life sciences research, including drug discovery. As research preview for customers like Moderna and Amgen. This is an area where OpenAI has discussed value capture pricing. More here.

Amazon’s $12 Billion Globalstar Buy w/ Apple Deal: Amazon’s $12 billion acquisition of satellite company Globalstar marks more ante in the low earth orbit broadband satellite business, going up against Elon Musk’s Starlink. That unit of course is the crown jewel of his SpaceX/xAI, getting ready for its mega-AI IPO this June, in time for Elon’s birthday. The Amazon Leo unit’s deal also includes a partnership with Apple, who is a 20% equity owner of Globalstar. That was a $1.7 billion investment two years ago, with Globalstar satellites providing emergency SOS messaging services to billions of iphone and Apple Watch owners on the ground. For Amazon, it means a couple of hundred legacy satellites, as well as an important radio spectrum portfolio, that potentially allows it to compete more aggressively with Starlink. These are long-term AI infrastructure investments in space in low earth orbit. Which will take billions more in investments, and a patient timeline to execute networks of thousands of Amazon Leo satellites encircling the earth. Starlink already has over a 60% share with over 10,000 satellites and counting. The holy grail for all the large players here is the chess game towards direct global broadband connectivity with billions of smartphones on the ground. Initially supplementing, and eventually bypassing terrestrial cellular wireless networks. More here.

Apple’s Big Splash coming into AI Smart Glasses: Bloomberg reports Apple is leaning into its long anticipated move into ‘simpler’ AI Smart Glasses. It marks a shift away from its higher end AR/VR Vision Pro experiment. Moving more towards the smart glasses that Meta has spent tens of billions ramping up with a range of third party brand and distribution partners. Apple seems to be going full vertical stack with its portfolio of four types of glasses in various styles. Presumably distributed via its hundreds of global, iconic retail stores. The key here is Apple’s intention to leverage its core asset, over two billion iPhones that could be more tightly coupled with the AI Smart Glasses. That then provides an off loaded processing and storage module, keeping the AI Smart Glasses light, and potentially relatively less expensive. It’s a capability not easily matched by competitors like Meta, Google, Amazon, Snap and others. More here.

Google Gemini AI Skills for Chrome Browsers: Google as long expected, is making its next move with AI features in its Chrome browsers. They call it ‘Skills, and they’re ‘Ai Powered’, using Google Gemini. The premade Skills, available via a Gemini sidebar in Google’s ubiquitous Chrome browsers, include ways to keep at hand a host of light-weight AI Agent tasks for its billions of global users. The strategy is an extension of the AI Browser trend I’ve been talking about for months. The Skills are also poised to assist in AI Shopping, an incremental revenue opportunity that Google and others are very keen on developing with AI Agents. Also to be expected would be seamless first party integrations with Google’s multiple billion user apps on the desktop and mobile, as AI Skills roll out over time. Separately, Google is also rolling out a local and global AI Search app on Windows. A smart move to gain distribution across hundreds of millions of windows machines worldwide. Additionally, Google is also updating AI Mode in Chrome, letting users open links side by side with AI mode on desktop. It’s all strong leveraging of Google’s global distribution opportunities with Gemini. More here.