Meta Has a Custom Chip and a Plan to Double Its Compute.

You’ve saved $1.5 million. Now comes the real test.

Can it produce $10,000 a month, or will that pace drain your portfolio?

Most retirees do not get a clear answer until it is too late.

The issue is not just how much you have. It is whether your portfolio was built to pay you, not just grow.

That difference can determine whether your money lasts decades or starts breaking down early.

Sequence of returns, taxes on withdrawals, healthcare costs, and whether the 4% rule still applies all play a role.

Fiduciary advisors created a breakdown showing what drives sustainable income and why the same $1.5M can produce very different outcomes.

If you have $1M or more invested, do not guess.

See How Long My Money Lasts

Learn More

3 things that happened while you were busy

1.  Meta plans to double its computing power by 2027, and Wall Street cheered.

An internal memo revealed Meta intends to double total compute within two years, backed by long-term supply deals including one with Samsung, and the stock jumped more than 7% on the news. The plan includes a $10B, 1-gigawatt data center in Alberta, Meta's 33rd worldwide and its first in Canada, a facility that draws roughly as much power as a small city.

2.  Gemini 3.5 Pro finally has a date. Allegedly.

Leaked launch plans point to July 17 for Google's long-delayed flagship, with a 2-million-token context window, a Deep Think reasoning mode reserved for the $250-a-month Ultra tier, and API pricing near $1.25 per million input tokens. Worth stressing: Google has not confirmed the date, and this model has slipped before.

3.  Claude Code's desktop app grew a browser.

The coding agent now ships with a built-in sandboxed browser, letting Claude pull up documentation, designs, or any website and interact with them the same way it works with your local dev servers. Small feature, big workflow change: less copy-pasting context, more pointing the agent at the source.

The models get the applause. The hardware gets paid.

The buildout.  Meta's compute-doubling memo is one data point in a much bigger pattern. The company is also putting its custom AI chip, code-named Iris, into production in September, designed with Broadcom and manufactured by TSMC. Officially it complements the Nvidia and AMD chips Meta keeps buying. Practically, every hyperscaler that builds its own silicon also gains a bargaining chip against Nvidia's pricing.

The club.  Meta is not alone in going custom. Anthropic has opened preliminary talks with Samsung about manufacturing its own AI accelerator on a 2-nanometer process, hiring silicon

engineers and sketching chip specs. When the model labs start designing hardware, the message is clear: whoever controls the stack controls their margins.

The bill.  All of this burns power at a startling rate. Google's latest environmental report showed electricity use jumping 37% in a year, its largest increase ever, with data centers alone consuming more than 42 million megawatt-hours, in the range of what entire countries like Denmark use annually. The company admits the AI buildout is outpacing grid decarbonization.

The takeaway.  This is the same lesson SK Hynix's $1T debut taught last week, now playing out across the industry: the reliable money in AI flows to compute, memory, chips, and power. For users, the buildout means the price war we covered yesterday has fuel to continue. The real constraint to watch over the next two years is not model quality. It is electricity.

What people are actually watching and sharing

Meta's Muse.  Meta AI's new image model works from your personal photos: restoration, Renaissance-portrait and claymation styles, room restyling, and product shots, a direct shot at Adobe Firefly and Google Imagen.

Accra calling.  Google launched the Africa Applied AI Lab, based in Accra, giving African researchers and entrepreneurs early access to its AI tech plus direct guidance from Google engineers to build solutions for the continent's own challenges.

Your car is watching you.  Since July 7, every newly registered car in the EU must include an AI driver-distraction detection system that reads gaze and head movement to catch inattention, without recording or transmitting footage.

The IPO scoreboard.  OpenAI is reportedly prepping its listing with Goldman Sachs and Morgan Stanley, possibly as soon as September at around a $730B valuation, while Fortune reporting puts rival Anthropic ahead on revenue, roughly $47B annualized, with Claude Code alone growing from $1B to over $2.5B in about two months.

Interrogate any document pile like an analyst

With Gemini 3.5 Pro reportedly arriving this week with a 2-million-token window, and every major model already swallowing hundreds of pages, the skill that matters is asking documents the right questions. This prompt turns a contract stack, research pile, or report dump into cited answers instead of vague summaries.

Works in Claude, ChatGPT, or Gemini with attachments. Try it on a lease agreement, a set of competitor reports, or your own meeting notes from the quarter. Advanced tip: the "what do these documents NOT answer" list is usually the most valuable part, so if you only read one section, read that one.