Eight hundred and eleven billion dollars in AI infrastructure — and it just met a governor with a stop sign. If you're just joining us, New York's gone from local skirmishes over AI campuses to a statewide pause. Any new or expanded data center that would pull 50 megawatts or more is temporarily blocked while state agencies study the environmental, utility, water, noise, and community impact. The fight also includes Hochul's push to repeal the sales-tax exemptions that massive existing data centers have enjoyed. This is the AI Daily Briefing. Today, Alphabet's future spending commitments jumped by half a trillion dollars in a single quarter. New York hits pause on big data centers, and Etched lands a ten-billion-dollar valuation. Cassidy, let's start with that Alphabet tab. The Next Web writes:
Alphabet had $811 billion in contracted future spending commitments at the end of June, up nearly $500 billion from just three months earlier. The figure, disclosed separately from capital expenditure in a quarterly filing, represents purchases Google has already committed to under supply agreements and open purchase orders. Roughly $200.7 billion of the total is short-term.
Alphabet's contracted future spending hit $811 billion at the end of June. Everyone's chasing the capex line — $195 to $205 billion for 2026 — but the commitments are the number that really moved: up $500 billion in one quarter. Yeah, and that's the part that's reframed my whole week. I'd been treating these compute deals as press-release theater — this is a filing, disclosed separately from capex, and it covers chips, power, data centers, content licenses. The demand's contracted, not projected. That settles it for me. The infrastructure behind the model launches has been the main story all along. But $500 billion in ninety days? I want the counterparty structure. Because “contracted” spans everything from take-or-pay deals you can't walk away from to letters of intent with escape hatches all over them — and the filing sure isn't telling you which. Exactly. At $811 billion, whoever's on the other side of those supply agreements has enormous leverage — and we still don't know who most of them are. At the current rate, that's four years of spending locked in. If the demand curve so much as flinches, it's far too big to unwind quietly. NBC News, with Allan Smith:
New York Gov. Kathy Hochul said she felt the need to sign the nation’s first statewide moratorium on data center development because tech giants “are flooding the zone” with construction efforts faster than states and localities can keep up.
So, right after Alphabet locks in $811 billion, Hochul signs the first statewide freeze on new or expanded data centers pulling 50 megawatts or more. That's where an $811 billion balance sheet runs into the physical world. And what jumps out in her NBC interview is this: New York doesn't need the data centers to support the AI boom. She's the first governor to draw a hard line between the buildout and what her state's grid actually needs. She's also calling this a response to tech giants flooding the zone. Strip away that demand framing, and you're looking at a permitting fight. And 50 megawatts lands squarely in hyperscale territory — the exact tier of sites Alphabet's supply agreements assume will get built. If three more states copy this, the $195 billion to $205 billion in 2026 capex guidance hits a wall it can't buy its way around. That's why I stopped treating the model launches as the headline. Now the regulatory clock is running against the investment clock, and this is the week you can watch them collide. The Economic Times writes:
Etched secured three hundred million dollars in a Series C funding round. This investment values the artificial intelligence chip company at ten point three billion dollars. Sequoia led the funding round with several other prominent investors participating. The company plans to use these funds to expand production and customer deployments.
Etched just raised $300 million in a Series C led by Sequoia, at a $10.3 billion valuation. And here's what they're actually building: a chip that runs only Transformers. They're baking that one architecture straight into silicon. So you're staking ten billion dollars on the bet that the Transformer remains the architecture to build around for the next several years. If the architecture shifts, that chip becomes a very expensive paperweight. Sequoia says it's the highest valuation ever for a Series C they've led. Great press-release line. But at that valuation, I still don't see a benchmark against Nvidia's inference stack that anyone outside the cap table has actually checked. The investor list has a16z and Jane Street, with SK Hynix coming in on the memory side. Now put that next to the Alphabet story we just covered. That $811 billion in supply commitments presumably still leans on Nvidia and AMD, so a pure-Transformer ASIC is a very narrow lane for a ten-billion-dollar valuation. But I'll give them this: they're going after inference, not training. That's where the cost curve actually bites. A chip that's cheaper per token wins the enterprise deal, full stop. Can they ship it before the design assumption expires? This one's from The Jerusalem Post:
Israeli cybersecurity startup Glow emerged from stealth on Wednesday with $180 million in funding at a $1.2 billion valuation, launching an AI-powered platform designed to prevent security threats from reaching corporate devices. Glow focuses on endpoint security, which protects the laptops, desktop computers, and other devices through which employees connect to corporate systems.
Glow comes out of stealth with $180 million in funding at a $1.2 billion valuation. The pitch is that AI agents will continuously map every laptop in your company and rip out risky software automatically. Automatically. On endpoints. That's the layer where one false positive bricks the CFO's machine mid-quarter-close. The founding team has real credentials, though. The CEO came out of Meta, the CTO ran cybersecurity strategy at Snowflake, and the R&D lead came from Claroty. That team has seen enterprise security at scale. Pedigree, I'll grant them. But the product depends on agents making unsupervised enforcement decisions: which software stays and which gets removed. I want the error rate on those removal calls before I let it touch a single production fleet. And that's a launch-day valuation. The company was founded in 2025, so we're at $1.2 billion before we've seen it survive a real breach attempt. Right, and this lands the same morning we're staring at Alphabet's $811 billion. Endpoint security is a real market, but agents haven't proved they can handle it well. That $180 million is a bet on the founders to solve the failure modes. If AI Daily Briefing helps you stay informed, subscribe and leave a review wherever you listen. It helps more people find the show, and we really appreciate it.
You'll find links to every story in today's show notes if you want to dig deeper. That's it for AI Daily Briefing this Friday. This is a Lantern Podcast.