> daily_signal(2026_06_09)

Apple put Google's Gemini behind Siri at Tim Cook's last keynote; the first new US coal plants since 2013 are being built to power AI

PickBits Daily Signal · Tuesday, June 9, 2026

By Mark Pickering · 9 min read · June 9, 2026

// tl;dr

At Tim Cook's final keynote as CEO, Apple rebuilt Siri on a custom Google Gemini model and shipped a standalone Siri app. The deal reportedly costs Apple about $1 billion a year. Your hardest questions now leave Apple's servers and run on Google Cloud, which changes the privacy bargain for more than a billion iPhones.
The Trump administration committed up to $850 million to build the first new US coal plants since 2013, justified under wartime emergency powers to feed AI data centers. Two new plants in Alaska and West Virginia total 2.85 gigawatts, and the West Virginia plant is wired straight to a 1-gigawatt data center.
A 339-mile line buried under Lake Champlain and the Hudson started carrying Canadian hydropower into New York City. It can supply about a fifth of the city's electricity, roughly a million homes, and is projected to cut statewide power costs by $17 billion over 25 years. It is the other answer to the same grid crunch.
ChatGPT's "dreaming" memory is rolling out to the free tier, on by default for 700 million-plus users. The system synthesizes what you tell it across every conversation, not just within one chat. You can read and edit what it stores, or opt out, but most people will never open the setting.

The AI boom showed up in two different parts of your life this week. In your pocket, the two assistants most people actually touch both changed underneath them: Apple moved Siri's brain to Google, and ChatGPT started keeping a running memory of every free user by default. On the power grid, the machines those assistants run on got two very different answers to the same question of where their electricity comes from. Washington is spending public money to build new coal plants for data centers, while a cable under the Hudson quietly started delivering clean hydropower to a million New York City homes. Two stories about the AI you talk to, two about the power behind it. The thread is that the decisions getting made this week are infrastructure decisions, and infrastructure is the part that is hard to undo.

This week Apple moved Siri onto Google's AI, the government spent $850 million to build new coal plants for data centers, a 339-mile line started carrying Canadian hydropower into New York City, and ChatGPT began remembering every free user by default.

1. Apple rebuilt Siri on Google's Gemini, and your hardest questions now leave Apple's servers

Continuing the WWDC preview thread (5.07): the keynote landed, and the headline is that the assistant on a billion iPhones now runs on a Google model.

If you own an iPhone, the assistant you have half-ignored for a decade is finally getting smart, and the brain behind it is now Google's, not Apple's. At WWDC on Monday, June 8, Apple unveiled a rebuilt Siri running on a custom 1.2-trillion-parameter Google Gemini model, a deal reported to cost Apple roughly $1 billion a year. The new Siri is agentic: it can chain multi-step commands across apps, and it ships as a standalone app with a chat interface rather than just a voice prompt. Apple paired it with iOS 27 and a preview of homeOS, the software for an upcoming HomePad smart-home hub.

The keynote was also Tim Cook's last as chief executive. Cook hands the title to hardware chief John Ternus on September 1 and moves to executive chairman. The Siri rebuild uses a three-tier routing system: simple requests stay on the device, moderately complex ones go to Apple's Private Cloud Compute servers, and the heaviest reasoning is sent to Google Cloud, running on Nvidia Blackwell B200 chips. For a company that has spent years marketing privacy as the product, routing its users' toughest queries to a Google data center is a real shift in the bargain, even with Apple's contractual guarantees layered on top.

Why this matters: If you use Siri, the AI answering your questions is changing hands, and the change is mostly invisible from the home screen. The practical trade-off is that your most complex requests now travel to Google's cloud instead of staying inside Apple's walls, which matters if you have spent years choosing Apple specifically for that boundary. It also marks the end of the idea that any one company can build the whole stack alone: Apple, which has the most reasons to want its own model, just rented Google's. Action this week: When the iOS 27 developer beta lands this week and the public beta this summer, open Settings and find the Apple Intelligence and Siri privacy panel before you turn the new Siri on. Read which request types Apple says route to Google versus stay on-device, and decide deliberately rather than tapping through the setup screen. If you manage devices at work, check whether your MDM profile already restricts third-party AI processing, and update your acceptable-use guidance before staff upgrade.

TechTimes: Apple WWDC 2026: Siri Rebuilt on Gemini, homeOS Previewed in Cook Farewell Keynote (June 8, 2026)
MacRumors: WWDC 2026 Gemini-Powered Siri, iOS 27 and more

2. The US is building its first new coal plants since 2013, using wartime powers, to feed AI data centers

Returning to the data-center power thread (5.30): the fight moved from a single Michigan plant to a national coal commitment.

If you pay an electric bill, the AI boom is now reaching for the dirtiest power source on the grid to keep the data centers running. On June 4, the Trump administration announced up to $850 million to revive coal, and the Department of Energy is invoking the Cold War-era Defense Production Act to justify it, citing national-security concerns about having enough power for US AI data centers. The package funds two brand-new plants, a 1.25-gigawatt plant in Anchorage, Alaska, and a 1.6-gigawatt plant in West Virginia, which together would be the first new US coal plants to come online since 2013.

The West Virginia plant is the part to watch. DOE's own fact sheet says it would supply power behind the meter to a 1-gigawatt data center, meaning the plant exists to feed the data center directly rather than the public grid. The money also restarts a shuttered plant in Cumberland, Maryland, modernizes one in Puerto Rico, and funds a coal export terminal. Fifteen years of US energy policy pointed away from coal on cost and emissions; this reverses that, and it uses an emergency wartime statute to skip the slower approval path that normally applies.

Why this matters: If you live in Alaska, West Virginia, Maryland, or near any of the dozen plants getting this money, the AI buildout is now a thing being built in your state with your tax dollars and burned in your air, and the official reason is that data centers need the power. New coal plants run for decades, so this is not a one-year patch; it is a long bet that locks in the emissions and the costs well past the current administration. The behind-the-meter data center in West Virginia is the clearest signal yet that the grid is being reorganized around the machines rather than the households. Action this week: Look up your own utility's data-center load and planned generation: most utilities publish an Integrated Resource Plan, and your state public utility commission posts rate cases at its website. The AI Now Institute's data center map at datacenters.ainowinstitute.org/local shows what is proposed near you. If a plant or data center is proposed in your area, the public-comment docket is where ratepayers actually get heard, and the comment windows are short.

Utility Dive: Trump administration announces $850M to modernize US coal capacity, build 2 new plants (June 4, 2026)
US DOE: Defense Production Act Title III coal project selections

3. A line under the Hudson just started delivering Canadian hydropower to a million NYC homes

The other answer to the same grid crunch: not every new megawatt for the AI era is coal.

If you live in New York City, about a fifth of your electricity now comes from Canadian dams through a cable buried under the Hudson. The Champlain Hudson Power Express, a 339-mile high-voltage line built entirely underground and underwater, reached commercial operation on May 13, ahead of schedule, and its Hydro-Quebec supply contract began on June 1. It carries 1,250 megawatts of hydropower from Quebec down through Lake Champlain and the Hudson and Harlem rivers to a converter station in Astoria, Queens.

At full output, the line can meet roughly 20 percent of New York City's electricity demand, enough for about a million homes, and is forecast to cut statewide generation costs by about $17 billion over 25 years while lowering carbon emissions by an average of 3.9 million tonnes a year. It is the kind of infrastructure that takes more than a decade to permit and build, and it is the counterweight to the coal story above: as AI data centers strain the grid, the two answers landing this same month are new fossil generation in one place and a wall of clean imported power in another. Which one a region gets is now a live political fight, not a technical one.

Why this matters: If your electric bill keeps climbing because of AI demand, this is a working example of the alternative to building a new fossil plant: import clean power on a dedicated line and use it to hold prices down. The $17 billion projected savings is the number that matters to your household, because transmission costs get spread across every ratepayer in the state. The catch is timing: this line was contracted years ago and took over a decade, so the regions deciding their AI-era power mix today are choosing what arrives in the 2030s. Action this week: Find out whether your region has a major transmission line in the queue: search your regional grid operator (PJM, MISO, ERCOT, NYISO, ISO-NE) plus "transmission projects" to see what clean-power delivery is planned near you. If you sit on facilities or sustainability at work, pull your utility's marginal-emissions data, because a cleaner grid changes the carbon math on every kilowatt-hour your data center or office burns.

THE CITY: New Yorkers, Electricity Will Now Come to You From Dams in Canada (June 3, 2026)
S&P Global: 1,250-MW CHPE line into NYC reaches commercial operation early

PickBits Daily Signal is free. If this lands in your inbox every day and it is worth something to you, the best way to support it is to share it with someone who would read it. Subscribe at pickbitsai.substack.com.

4. ChatGPT's free tier just started remembering everything you tell it, by default

The capability shipped quietly: the change is on by default and the off switch is buried in a menu most people never open.

If you use the free ChatGPT, it is about to start remembering what you tell it across every conversation, not just within a single chat. On June 4, OpenAI began rolling its "dreaming" memory system (internally Dreaming V3) out to free-tier users for the first time, after a roughly 5x cut in the compute it takes to run. The system runs in the background and synthesizes a running summary of what it knows about you, which it then uses to personalize future answers. OpenAI says the feature reaches 700 million-plus weekly users, with some regions waiting until late July.

The design includes real controls: you can open a readable memory summary, edit or delete what it has stored, give it standing instructions, or switch back to the old "saved memories only" behavior. Plus and Pro accounts get twice the memory. The honest part is that the feature is on by default, and most people never open the settings menu, so for the typical free user the practical reality is a chatbot that quietly accumulates a profile of their work, health, relationships, and habits. The EU AI Act's chatbot transparency rules take effect August 2, which is the first external pressure on how clearly this gets disclosed.

Why this matters: If ChatGPT is where you think out loud about your job, your money, or your family, it is now building a persistent model of you that carries from chat to chat, whether or not you asked it to. That is genuinely useful, it stops you repeating context, and it is also the thing to be deliberate about, because a default-on profile you never reviewed is a profile you do not control. The pattern is the same one running through this whole edition: the capability ships first and switched on, and the consent is a setting you have to go find. Action this week: Open ChatGPT, go to Settings, then Personalization, then Memory, and read the memory summary it has already built about you. Delete anything you would not want surfaced, and decide whether you want it on at all; you can switch back to saved-memories-only or turn it off. If your team uses ChatGPT for work, remind people that personal-account memory is not covered by your enterprise data agreement, so sensitive work context does not belong in a free personal account.

TechTimes: ChatGPT Memory Dreaming Update: OpenAI Rewrites Personalization Engine (June 5, 2026)
Implicator: OpenAI expands ChatGPT memory to free users after 5x compute cut

» What to watch this week

The iOS 27 public beta and independent testing of where the new Siri actually routes requests, plus any regulator questions about the Apple-Google data arrangement.
Whether the Anchorage and West Virginia coal plants draw state-level permitting challenges, and which other behind-the-meter data-center power deals surface under the Defense Production Act.
New York's Hochul decision on the one-year data-center moratorium passed by the legislature, which would set the terms for where the state's next wave of AI power demand can land.
How clearly OpenAI discloses default-on memory as the EU AI Act's August 2 chatbot transparency rules approach, and whether US state privacy regulators follow.
The Philadelphia police document showing surveillance of anti-data-center speech, and whether other cities are running the same monitoring as the buildout fights spread.

Tomorrow's signal lands here.