Good morning. Somewhere in a data center last week, an OpenAI model sat for a cybersecurity exam, decided the honest route was too slow, broke out of its sandbox, reached the open internet, and went rummaging through the servers of AI startup Hugging Face looking for the answer key. OpenAI says the model took roughly 17,000 actions over a weekend before anyone clocked what was happening, and both companies now call it "day one" of AI cybersecurity. HackerOne CEO Kara Sprague told Bloomberg Technology the disclosure was "an example of responsible behavior," which is one way to describe a model that cheated on a test by hacking a stranger.

Every horror movie eventually reaches the line where the operator realizes the call is coming from inside the house. Wall Street just got its first look at the caller ID.

Big Tech

Google spent two decades printing money; this quarter it learned to set some of it on fire on purpose.

Alphabet's second quarter, reported after Wednesday's bell, was the kind of number that should have sent the stock up: revenue rose 24% year over year to $119.8 billion, and Google Cloud grew a blistering 82% to $24.8 billion, comfortably ahead of Wall Street's estimate. Then investors read the spending line, and the applause curdled. Shares fell nearly 5% after hours.

The problem isn't the business. It's the invoice attached to it.

Alphabet is laying the foundation for an AI empire, and the cement trucks never stop arriving. Management lifted full-year capital-expenditure guidance to $195–$205 billion, up from the $180–$190 billion floated in April, and the CFO warned that 2027 spending will run higher still. The bill is already showing up in the cash flow statement.

By the numbers: Q2 capex hit $44.92 billion, up 100% from a year ago; free cash flow swung to negative $5.9 billion, Alphabet's first cash burn since it went public more than two decades ago; and the Google Cloud backlog — contracted work not yet booked as revenue — swelled to $514 billion, up from roughly $460 billion in Q1.

Blueprint or Money Pit: The backlog is the bull case in one number — half a trillion dollars of demand waiting for data centers that don't exist yet, which is why the concrete matters. The bear case is the same number read backward: every dollar of that backlog needs steel, silicon, and power before it becomes revenue, and Alphabet is now the first hyperscaler burning cash to get there. With Microsoft, Meta, and Amazon all reporting next week, Google just set the price of admission to the AI build-out, and it's being paid in cash flow.

Autos & EVs

Tesla had its best second quarter ever and got graded on a curve it drew itself.

The company delivered 480,126 vehicles in Q2, up 25% and a record, and revenue climbed 26% to $28.24 billion — both ahead of estimates. But the money that reached the bottom line disappointed: adjusted earnings landed at $0.33 a share against the $0.53 analysts wanted, and free cash flow went negative for the first time in more than two years as AI and robotics spending ramped.

Investors have quietly re-badged Tesla as an AI-and-robotics story. The engine underneath is still, stubbornly, a car company — and the car company is downshifting.

Here's the tell buried under the record deliveries:

That last data point is the whole margin story in one driveway. Tesla stopped being a luxury marque and became a mass-market one, and mass-market metal carries mass-market margins.

The Street's complaint is almost funny: it wants Elon Musk to spend more. Tesla has guided to roughly $25 billion of capex this year for robotaxis, Optimus robots, and AI compute, and after-hours the message from analysts was to put the money where the promises are.

Autopilot Engaged, Profits Not: Tesla is asking investors to fund an AI future with the cash from a shrinking car margin, and this quarter the two lanes visibly diverged. Until a Cybercab is actually collecting fares or an Optimus is actually working a line, the mainframe of this business is still the assembly line, and it just posted a 1.4% operating margin. — AllThingsWallSt, our take.

Semiconductors

If you can't out-chip the champion, buy a seat at the table where the chips get bought.

AMD announced a strategic partnership with Anthropic under which the Claude maker will deploy up to two gigawatts of AMD's next-generation Instinct MI450 GPUs, with the first gigawatt going live in the first half of 2027. AMD is backing the tie-up with an equity investment of up to $5 billion in Anthropic, and, per Yahoo Finance, is separately in talks to guarantee some of Anthropic's future data-center lease obligations. AMD opened lower on the news, then rebounded with the rest of the chip sector.

The move is a page torn straight from the Nvidia–OpenAI playbook, and it lands the same question on the table: why does a chipmaker need to buy equity in its own customer just to sell it chips? Anthropic has already struck compute deals with Google and Broadcom; adding AMD as both a supplier and a shareholder is less about one order and more about locking in an ecosystem before Nvidia locks it up first. Franklin Templeton's Sara Araghi, an early Anthropic investor, said on Bloomberg Technology that "there's a need for these investments" and that every player is now doing the same — while conceding the arrangement will "drum up the circular financing topic." That topic is the whole game: the AI hardware market has started to look like a table where the house stakes the players so they keep betting.

All In or Folding: For AMD, timed a day ahead of its Advancing AI event, the bet is that a marquee frontier-lab customer running MI450 racks at gigawatt scale is worth more than the $5 billion of chips it's effectively pre-funding. For the market, the tell to watch is whether these stake-for-silicon deals keep multiplying, because when the chipmaker, the lab, and the landlord are all financing each other, the pot gets very large and very circular before anyone shows their hand.

Content Is King, Rent Is Due: Reddit (RDDT) fell about 8% after the WSJ reported it has discussed cutting off Google's access to its content for AI training, reassessing a 2024 licensing deal reportedly worth around $60 million a year.

Lease on Life: Klarna popped 11% after Bloomberg reported Apple will name it the partner for a new "Apple Upgrade" device-leasing program, filling the financing hole left when the Apple–Goldman Sachs relationship fell apart.

Big Iron, Small Comfort: IBM (IBM) slipped again as full results confirmed the 42% collapse in Z mainframe sales that had already triggered a historic 25% one-day crash on July 14, as clients diverted spending toward AI-adjacent hardware.

Three of the biggest names in tech reported within an hour of each other, and the market's verdict rhymed: spending is fine, so long as somebody, somewhere, someday turns it into cash. Google poured a record foundation, Tesla sold a record number of cars for record-low margins, and AMD staked a competitor to keep the table full. The AI build-out is no longer a promise — it's a capex line, and the bill is finally arriving in the mail.

That's the tape. We'll see you at the open. — AllThingsWallSt