Capex, Code, and Bitcoin: what Tesla cutting its AI spend, AI finding holes in 70 percent of the web, and Bitcoin dropping back to 81K mean for people starting from zero

Bitcoin touched 85K late Wednesday. By Thursday afternoon it had given almost all of it back. Same story across every screen: tech sold off, oil spiked, crypto bled, and the question on every trading desk changed from ”is AI demand real” to ”who pays for the buildout.”

And in the middle of it, AI just found two critical holes in the code that runs 70 percent of the web. Then researchers showed the same class of tools is already being used to attack at machine speed.

Three stories. One pattern. The money is no longer funding the dream. It is auditing the bill. Let me explain what that means for you.

The market stopped asking if AI works. It started asking who pays.

Meta reported Wednesday and the numbers were good on the top line: revenue 64.60 billion dollars, up 26 percent. But investors read the rest of the release and sold anyway. Meta raised its 2026 capital spending to a range of 135 to 145 billion dollars. Free cash flow went negative 2.12 billion dollars. Operating margin was cut in half, from 40 percent to 20 percent. The 10-Q shows 454.52 billion dollars of non cancelable commitments, with 121.41 billion of that falling due in 2027. The company also plans a new publicly listed, non voting Class C stock as a one time dividend effective January 19, 2027, and it is carrying a 25 billion dollar New Mexico privacy penalty. The stock fell about 2 percent after hours.

Tesla went the other direction and got punished too. It trimmed its capital spending plan to about 20 billion dollars, cut its 2030 sales goal to 5 million vehicles, and reported net income down on flat revenue. Investors read that as doubt about the pace of AI spending. The stock fell more than 5 percent premarket.

So both directions lose. Spend too much, you get sold. Spend too little, you get sold. Chipmakers including Nvidia and Intel fell with it. In Asia, the Kospi closed down 3.5 percent, the Hang Seng fell 3 percent, the Nikkei slipped 2.4 percent. Nasdaq 100 futures pointed about 1.8 percent lower, the S&P 500 about 1 percent lower.

Meanwhile Brent crude sat near 112 dollars, close to a five month high. Gold actually fell 1.2 percent to about 4,090 dollars. Treasuries rallied, with the 10 year yield near 5.08 percent. France’s 10 year yield hit 5.126 percent, its highest since 2002.

What does this tell you, starting from zero? Yesterday’s post was about SpaceX, where AI is now 52 percent of revenue. Monday’s was about the 7 trillion dollar AI spending warning. Today investors looked past the demand, which everyone agrees is real, and stared at the financing. The constraint was never the technology. It is always the money. And here is your edge: you do not need billions. The buildout is expensive. Building with AI is cheap. That gap is where you enter.

Bitcoin: a 3,000 dollar round trip in under a day

The attempt to break 85K got rejected near 84,876 late Wednesday, and Bitcoin slid all the way to about 81K by Thursday. Live numbers as I write: Bitcoin 81,271 dollars, down 3.5 percent over the last 24 hours. Ethereum 2,379.36 dollars, down 4.29 percent. Total crypto market cap 2.74 trillion dollars, down 6.02 percent over 24 hours. Bitcoin dominance 59.49 percent. The Fear and Greed index reads 51, neutral.

Leveraged traders got burned on both sides. About 412 million dollars of longs were liquidated market wide, including roughly 135 million in Bitcoin longs. The single biggest hit was an 11.8 million dollar Bitcoin long on Binance closed near 83,500 dollars. DeFi tokens led the decline, with one sector index down 5.6 percent.

But look at what did not break. US spot Bitcoin ETFs took in more than 1.08 billion dollars across the first three days of the week, even as the price fell. Analysts point to support at 80,000 to 81,500 dollars, which lines up with the cost basis of the spot ETF buyers. Real money keeps buying the dips.

The calendar from here: US jobless claims land today, flash PMIs on Friday, and the Fed meets October 27 and 28, with a rate hike widely expected in December.

The money does not ask whether AI works. It asks who pays for it.

The AI news: AI found the holes in 70 percent of the web

On Wednesday, Wiz Research published the technical disclosure of two unauthenticated remote code execution flaws in PHP: CVE-2026-109618 and CVE-2026-109619, with severity Wiz rates at up to 9.2 out of 10. The part that stops you: nearly every PHP version is affected, all the way back to PHP 5.4. PHP runs roughly 70 percent of the web. The fix is PHP 8.5.12, and the NSA advised pulling unpatchable sites offline.

Here is how they found them. Wiz says plainly: “We used Wiz Atlas, Gemini 4 Argon and GPT-5.6 Sol throughout this research.” A security company called Calif found one of the flaws with GPT-5.6 Sol under OpenAI’s Patch the Planet initiative.

Let me explain what that means in plain talk. An AI agent read code the way a tireless reviewer would, traced how memory gets handled, and found bugs that humans missed for 20 years. Then Wiz issued the warning that matters more: this ”significantly lowers the barrier to scaling attacks.” Adapting an exploit across versions and configurations used to be tedious human work. Models erase the tedium. The same capability that finds the bug can farm it at scale.

And attackers are already there. Microsoft attributed the ConnectWise supply chain compromise to a China aligned group running ARTEX, an open source penetration testing framework that works on any model and runs an autonomous penetration test from reconnaissance through data exfiltration with little human involvement. Microsoft logged sign in attempts against more than 2,250 tenants peaking at 2,900 a minute, and in several cases the attackers went from first targeting to privileged cloud access in under three hours.

The AI supply chain took its turn as the target, too. Version 1.107.0 of LiteLLM shipped a credential stealer and a backdoor through PyPI, built to take SSH keys, cloud secrets, and AI provider API keys. That is the project’s second compromise.

One sober measurement came from Trail of Bits: 21 confirmed regressions across more than 130 agent written patches, and none of them broke a security protection. Translation: AI written fixes work when a human reviews them. OpenAI also published a five stage framework for disclosing AI discovered vulnerabilities, and said it expects similar findings to increase sharply in the coming months.

So what does this mean for someone starting from zero?

This is the clearest signal yet that AI security skills are about to be one of the most valuable things you can learn. Here is one concrete thing you can do today: pick any open source project you use, or your own website code, and ask an AI agent to review a file for security issues. You are practicing the exact skill Wiz used. Reading code with an AI reviewer is free, it takes an hour, and it is a real job category now.

And if you run a website on PHP or WordPress, update to PHP 8.5.12 today. That is the builder’s version of this news.

The agent economy is getting real, and it has a price tag

Watch the other big pattern forming. Meta said its Muse agent now has more than 10 million users in the US and Canada, about 25 percent of them daily, and called it highly unprofitable. Anthropic put Claude Duet into research preview: a proactive agent that sends a morning brief and can buy things through Stripe with a spending authorization request. Walmart opened an Agent Gateway, an MCP server that lets approved agents search inventory and buy on a customer’s behalf, with per transaction approval or a spending cap. Amazon, meanwhile, is blocking purchase agents. Banks are pushing back: one CEO said agent only bank transactions “will never happen.”

Agents that can spend money are the next battlefield. The protocol underneath Walmart’s move is called MCP, the Model Context Protocol. It is open, it is documented, and it is what everything is standardizing on. Learn MCP this weekend. It is free to learn, and it puts you inside the conversation where the real building is happening.

The uncomfortable truth

The AI boom will not be paid for by everyone. Today investors drew the line in public: revenue without a credible path to covering the buildout gets sold, and even cutting spending gets sold when it reads as doubt. Bitcoin will not rescue you from that. It trades with the same fear as everything else right now.

The only asset nobody can dilute is a skill you can demonstrate. The models that found holes in 70 percent of the web just told you which skill is scarce: people who can read what the machine writes and check it. Be that person.

Start tonight. Pick a piece of code. Ask an agent to audit it. Learn what it found. That is one hour, zero dollars, and you will know more about how AI security works than 99 percent of the people talking about it online.