The Vatican vs. Silicon Valley. Pope Leo XIV issues a landmark encyclical demanding global AI disarmament and blasting tech monopolies. Google completely deletes classic search for a conversational Gemini interface. Plus, xAI’s leaked financials show a $6.4B burn rate.
xAI & Anthropic's Billion-Dollar Co-dependency
What if I told you the world's most aggressive AI lab just burned through 6.4 billion dollars in a single fiscal year, and is relying on its biggest rival to stay afloat with a billion-dollar-a-month compute contract? I mean, it sounds like complete corporate fiction. But if you actually dig into the audited, unfiltered reality of the new AI economy, specifically the recent SpaceX S-1 SEC filing for their public offering, it's all there.
The Cost of Frontier Models (FY2025/26)
You're probably wondering how a loss that big even happens. Let's tear the lid off the economics of intelligence. The financial numbers for xAI for fiscal year 2025 are staggering. They pulled in about 3.2 billion dollars in topline revenue, which is great for a company this young, especially with Grok natively on the X platform and their premium API subscribers. But their operating loss was 6.4 billion dollars. They are spending two dollars for every dollar they make just to keep the lights on. To really understand the scale, look at their capital expenditure, the physical infrastructure, servers, real estate. In the first quarter of 2026 alone, xAI spent 7.7 billion dollars. That's in 90 days. We are talking nation-state levels of investment, deploying the GDP of a small country into server racks over a three-month period.
How do you finance a hole that deep? This is the ultimate plot twist. That massive cash burn is heavily underwritten by a compute contract with Anthropic. Yes, Anthropic. They are paying xAI 1.25 billion dollars per month through May 2029 just to lease time on their Colossus supercomputer. Let that sink in. Anthropic, the company building the Claude model and fighting tooth and nail against xAI in the enterprise market, is paying their rival over a billion a month. That is like Ford paying Chevy a billion dollars a month to use Chevy's factories to build trucks that directly compete with Chevy. It shatters the public illusion of this vendor war. Structurally, they are deeply codependent because compute is the only currency that matters.
The Compute Bank of Nvidia & Perplexity's Shift
So who is actually making the real money here? The central bank of the compute economy: Nvidia. Nvidia remains the central bank of the compute economy, reporting a staggering Q1 net income that skyrocketed 211 percent year-over-year to $58.3 billion. Half a trillion dollars of hardware is being demanded by hyperscalers and sovereign buyers to keep up with the intense need for expanded compute clusters. And they pulled this off without full access to China, even as US officials warn firms like Super Micro to tighten export compliance. These labs aren't buying chips for today's balance sheet; they believe whoever achieves artificial general intelligence first will capture global economic growth.
Click to shift Perplexity's Strategy
Legally toxic. Erodes user trust, risks copyright infringement lawsuits from major publishers.
But if you are a consumer AI company, you don't have infinite billions to burn. Look at Perplexity. They just hit a 21.21 billion dollar valuation, closed their Series E-6, crossed 200 million dollars in annual recurring revenue, and completely abandoned their ad strategy. They pivoted entirely to a subscription-only model. Why? Because ad-supported AI answers are legally toxic. Perplexity's engine relies on RAG, Retrieval-Augmented Generation. That’s a mechanism where the AI scours the web, scrapes the content, and synthesizes an answer, meaning the user never clicks through to the original website.
If you monetize publishers' work with sponsored links, you erode user trust and become a massive target for copyright infringement. Major entities like Dow Jones have already launched lawsuits over this, while others like the BBC have threatened legal action and demanded an end to unauthorized scraping. By retreating to subscriptions, Perplexity builds a legal shield. They even secured a 750 million dollar compute commitment with Microsoft Azure for a premium "Deep Research" tier.
Polsia, Microsoft & Anthropic's Autonomous Systems
Then you look at the application layer, and the economics get completely wild. Have you seen Polsia? They just closed a 30 million dollar funding round at a 250 million dollar valuation, they are nearing a 10 million dollar annual run rate, and they have zero human employees. The entire company is an autonomous AI agent system. Imagine a fully autonomous digital ad agency that monitors trends, automatically negotiates media buys, A/B tests assets, and routes invoices 24/7. When your marginal cost of human labor drops to zero, profitability skyrockets.
How Computer Use Agents (CUAs) Work
1. Vision
AI uses computer vision to "see" the screen GUI.
2. Navigation
Moves mouse & navigates legacy software.
3. Security Wall
Authenticated via Azure Key Vault / MCP Tunnel.
How is an AI physically doing this? Microsoft just announced the general availability of computer use agents, or CUAs, within Copilot Studio. A CUA doesn't just return text; it literally takes control of a virtual machine. It uses computer vision to see the screen, move the mouse, navigate legacy software, and type on a keyboard. Microsoft built a massive security wall around this, integrating with Purview log sessions so every single micro-action is immutably logged for compliance.
The agents authenticate using Azure Key Vault, think of it like a high-security bodyguard. The AI walks up, asks for access, and if approved, the bodyguard unlocks the door. The AI never holds the keys. And the cost? Four cents a step. That obliterates outsource data entry. Anthropic is building similar walls. At the Code with Claude event in London, they introduced the MCP tunnel, the Model Context Protocol. Imagine your database is a bank vault. The MCP tunnel is a secure, vacuum-sealed tube built from the public cloud directly into your vault. The AI reaches its hand through, processes data inside your firewall, and pulls its hand out. Your data never travels to Anthropic's public servers.
ASUS, Cursor & DeepMind's Escaping Capabilities
But running these agents 24/7 costs money. ASUS just unveiled a hybrid architecture strategy across their commercial ExpertBook and NUC mini PCs to cut cloud inference costs by up to 70 percent. They use adaptive memory extension to trick standard systems into using fast solid-state storage as expanded VRAM. It allows local machines to run massive 26-billion and 35-billion parameter models. It's the ultimate pressure release valve, routing complex tasks to the cloud and mid-tier tasks locally.
And as processing shifts, the cost of building software is collapsing. Cursor is quietly becoming one of the fastest-growing software companies ever, hitting a 3 billion dollar annualized revenue run rate. They shipped Composer 2.5, dropping the price for frontier coding to almost zero, 50 cents per million input tokens and 2.50 dollars for output. They did this through reinforcement learning post-training. Instead of just pre-training a base model, like teaching a student to read by reading the whole internet, they locked their Moonshot Kimi K2.5 model in a room with a compiler. Trial and error. They punish it when code breaks, reward it when it works.
It's so effective that SpaceX secured the right to acquire Cursor for 60 billion dollars outright, or pay a 10 billion dollar walk-away fee. Managers are freaking out. Spotify reported 99 percent of their engineers use AI coding tools weekly, resulting in a 94 percent productivity lift and a 76 percent increase in pull request frequency. This is why the talent war is escalating. Andrej Karpathy just officially joined Anthropic as a research scientist to work on Claude's pre-training. And Google DeepMind just proved AI is moving past predictive text with their AlphaProof Nexus breakthrough. They paired a creative large language model with a rigorous formal proof assistant called Lean. The LLM guesses creatively, Lean verifies the formal logic, and they iterate. Together, they solved nine open Erdős mathematics problems, two of which were unsolved by humans for 56 years, and 44 open OEIS conjectures.
The Native Gemini Overhaul of Google
If AI is solving 56-year-old math problems and driving our computers, what happens to the consumer web? Google completely overhauled their core search engine at I/O 2026. The concept of keyword crawling is dead. They transitioned the entire search experience to be natively powered by Gemini 3.5 Flash. No more AI tabs; it's just the default. This coincided with a massive 14-day core algorithmic update that threw SEO into chaos. The web is being restructured for AI agents to retrieve data.
I've built your custom PC under $1,500. I sourced components from three retailers, verified motherboard compatibility, and added them to your universal cart.
Over a billion monthly active users are now in this AI-first interface. They also rolled out Gemini Omni, which natively edits multi-modal video, and the universal cart. You can ask your Google agent to build a custom PC under 1,500 dollars, and the AI automatically sources components from three different retailers, checks compatibility, and checks you out across all three businesses simultaneously. They are pushing this onto your face, launching smart glasses with Samsung, Warby Parker, and Gentle Monster. Plus, Anti-gravity 2.0 lets anyone build complex apps through natural language. It's like replacing the Dewey Decimal system with a super librarian who reads the book, writes a sequel, and buys the merch for you.
The Global Map: Alibaba, DeepSeek & Vernacular Models
Meanwhile, the global map of AI power is being redrawn. Alibaba launched their Qwen 3.7 Max model, scoring a 69.7 on the Terminal Bench 2.0, beating Western models. DeepSeek made their 75 percent price cut completely permanent on their V4 Pro model. And India is executing a massive blueprint for Vernacular AI. At the Bihar AI summit, policymakers prioritized hyper-localized models natively fluent in regional dialects like Hindi, Bhojpuri, and Maithili to bypass English-centric literacy barriers for rural populations.
China
Qwen 3.7 Max beats benchmarks; DeepSeek permanently cuts Pro API costs by 75%.
India
Vernacular AI push for rural inclusivity. Models natively fluent in Hindi, Bhojpuri, Maithili.
Canada
$16.5M injection into Toronto to scale sovereign AI & retain IP borders.
Canada just injected 16.5 million dollars into the Greater Toronto Area to scale homegrown AI, funding sovereign companies like Cosm Medical and Edgecom Energy to keep intellectual property within their borders.
State Laws, NSA Black Ops & Pope Leo XIV
In the US, regulation is experiencing massive whiplash. States are advancing aggressive bills. California is pushing bans on unconsented digital replicas. Illinois passed the AI Safety Measures Act for frontier developers. New York is mandating third-party bias audits and pushing to ban interactive chatbot toys. But at the federal level, the administration reversed a planned executive order for safety oversight. Instead, the White House approved a classified 9 billion dollar funding request to acquire high-end AI processors for deep espionage.
The Anthropic NSA Contract
Deploying frontier models directly into secure, air-gapped government networks. Driven by Project Glasswing & Mythos model finding 10,000+ critical vulnerabilities in a month.
Anthropic is finalizing a highly classified contract with the NSA to deploy their frontier models directly into secure, air-gapped government networks, networks physically disconnected from the internet. This urgency is driven by Anthropic's Project Glasswing and the unreleased Mythos model, which found over 10,000 high or critical severity vulnerabilities in a single month. It can independently write zero-day exploit code to completely take over systems.
Right in the middle of this military escalation, Pope Leo XIV issued a landmark papal encyclical titled Magnifica Humanitas. Standing alongside Anthropic's Chris Olah, the Pope explicitly called for the strict disarmament of AI, warning against autonomous weapons systems operating beyond human moral reach and the opaque concentration of digital infrastructure in Big Tech's hands.
The Double Distortion & Meta's Restructuring
This velocity is taking a massive physical and psychological toll. A new study from GCheck exposed the double distortion of the workforce. 63 percent of workers surveyed, and 80 percent of Gen Z, admit they are actively exaggerating their AI skills to management. They are terrified of losing their jobs to automation, but pressured to look like AI power users. They pretend to use Copilot in meetings, but do the work manually in secret. Over 80 percent admitted they discourage or limit using AI at work.
Management Persona
"I use AI for everything! My workflows are 10x faster thanks to Copilot."
Companies using AI as a scapegoat for layoffs are pouring gasoline on this resentment. Meta just executed 8,000 layoffs, reassigned 7,000 employees to their AI department, and doubled their 2026 AI budget to between 125 and 145 billion dollars. Leaked audio even revealed they are using tracking devices to feed employee workflows directly into their LLMs.
The physical world is feeling the strain, too. In Utah, and across states like Virginia, Maine, Texas, and New Jersey, public pushback is rippling across the US. Local residents are furious and organizing protests over the land, water, and power that these massive AI data center facilities will drain.
The Physical Chaos of Waymo & Final Takeaways
And for all these trillions spent, AI still struggles with physical chaos. There are times to take a robotaxi, and times you need a human driver. I found this out the hard way this week. I was staying in San Francisco, heading to Mountain View for the Google I/O keynote. I took a Waymo. It cost me 102 dollars and 72 cents. We hit massive traffic in Mountain View, and instead of taking a creative alternate route like a human driver, the Waymo got hopelessly confused. It pulled into a Google campus parking lot and essentially froze. I had to get remote support on the speakerphone just to get dropped off safely. Waymo gave me a 10 dollar credit for my trouble, but it was a perfect illustration: human intuition, context, and rule-breaking still vastly outpace machine logic in the chaotic physical world.
Subscription-First Web
Ad-supported internet breaks as AI search natively synthesizes answers.
Local Hardware Returns
Hybrid compute solutions are vital to bypass crushing cloud inference bills.
Beware the Double Distortion
Mandated AI use without safety leads to faked productivity and resentment.
To summarize: We are in the middle of the fastest technological infrastructure rollout in human history. The immediate actionable insights for you are clear. First, prepare for the subscription-first web. The traditional ad-supported internet is breaking as AI search engines synthesize answers natively. Second, local hardware is back. Look into hybrid computing solutions for your enterprise to bypass crushing cloud inference costs. And third, be aware of the "double distortion" in your own teams, mandated AI use without psychological safety just leads to faked productivity.
And that's your daily dose of AI Know-How from ainucu.com, AI News You Can Use. The biggest takeaway today is this: As AI writes our code, edits our video, searches the web, and negotiates our contracts, we are rapidly approaching an internet that is just a closed loop of machines talking entirely to other machines, with humans doing nothing but paying the monthly subscription fees to watch it happen. Keep questioning the systems around you. You are now the most informed person in the room. Thanks for taking the time with us.
Key Terminology
Click to flip and learn
Model Context Protocol (MCP)
A secure, vacuum-sealed tunnel that allows AI to process data directly inside a private firewall without sending it to public servers.
Knowledge Check
1 / 4Which company is paying xAI $1.25 billion per month for compute power?
Module Complete!
You scored out of 4.