Apple’s Foldable “iPhone Ultra”

The Big Picture
After years of speculation, Apple has confirmed a September 9, 2026 “surprise and shine” event, and it’s now widely expected to include Apple’s first-ever foldable iPhone alongside the standard iPhone 18 lineup — though manufacturing hurdles could still push actual shipping into late 2026 or even 2027.

Design — the “Passport” Shape
Unlike tall, skinny foldables from Samsung and Google, the iPhone Ultra takes a wider-than-tall “passport” form factor — a book-style hinge (like the Galaxy Z Fold or Pixel Fold) that opens left-to-right rather than top-to-bottom.

Key Specs (per leaks)

• Outer display: 5.5 inches
• Inner display: 7.6–7.8 inches, OLED, roughly 4:3 aspect ratio (iPad mini-like)
• Thickness: ~4.5mm when unfolded — potentially Apple’s thinnest device ever
• Build: Titanium outer frame + aluminum, using a liquid-metal hinge (inspired by Oppo’s “Zero-Feel Crease” tech)
• Chip: Apple A20 Pro on TSMC’s 2nm node, 12GB RAM, up to 1TB storage
• Battery: Up to ~5,800mAh (two cells, ~1,921mAh + ~2,962mAh combined)
• Reports suggest it may drop Face ID for Touch ID and skip MagSafe — practical trade-offs for the ultra-thin foldable design

The Headline Feature: No Crease
Apple has reportedly pursued a genuinely crease-free display “regardless of cost,” developing a new material property that makes the fold essentially invisible when open — a problem that has plagued nearly every foldable phone on the market to date, including Samsung’s.

Price and Competition
Most estimates put the starting price at $2,000 or more — the most expensive iPhone ever. It’ll go head-to-head with a rumored Samsung Galaxy Z Fold 8 Wide, which reportedly shares a similar 4:3 aspect ratio, suggesting Samsung is bracing for direct competition. Analysts see this as a potential turning point for foldables moving from a niche category toward the mainstream, given Apple’s market influence.

Chrome’s Manifest V2 Extension Cutoff

What Actually Happened on August 31
This is a bit less dramatic than headlines suggest: Google permanently deleted all remaining Manifest V2 (MV2) extension listings from the Chrome Web Store. But the real functional death happened over a year earlier — MV2 extensions stopped running in Chrome entirely back in July 2025 (Chrome 138). August 31 was really a “database cleanup,” removing dormant listings, reviews, install counts, and the ability to ever reinstall them — not a new disruption to anyone’s daily browsing.

What Changes for Users

• Any MV2 extension still installed on an old Chrome version (138 or earlier) can keep running, but can’t receive updates
• If you get a new device or reinstall Chrome, you can no longer reinstall those old extensions
• Most people affected already stopped noticing a year ago when the extensions quietly stopped functioning

Why It Matters — The Ad Blocker Angle
The most consumer-relevant fallout was the transition’s effect on ad blockers: because Manifest V3 removed the old blocking webRequest API (replaced with a more limited declarativeNetRequest system), the popular uBlock Origin was removed from the Chrome Web Store — Google now only offers the feature-limited uBlock Origin Lite, which lacks dynamic filtering and real-time logging. This has fueled a longstanding “Google is weakening ad blockers to protect ad revenue” criticism, even though Google frames MV3 as a security and privacy improvement.

Why This Took 4+ Years
Google began this transition back in 2021, closing the Chrome Web Store to new MV2 submissions in January 2022. The multi-year rollout was deliberately staged to give developers time to migrate, which is why the actual “end” already happened quietly in mid-2025, with August 31 just closing the book.

Workaround
Firefox remains the main mainstream browser still supporting MV2-style blocking extensions, so users wanting the older, more powerful ad-blocking tools have been migrating there.

Why This Is a Genuinely Rare Story

This is one of the few truly bipartisan backlash issues in US politics right now. An Annenberg Public Policy Center poll found 61% of Americans oppose new data centers in their communities — including 69% of Democrats AND 54% of Republicans. Other polling backs this up: a Fox News poll found 70% oppose data centers being built in their area, and a Reuters/Ipsos poll found 59% would oppose one within 10 miles of their home.

Why People Are Angry — Three Main Threads

1. Electricity bills: Goldman Sachs projects data centers could drive a 6% national rise in electricity bills over the next year, with the sharpest increases hitting people who live nearest the facilities.
2. Water and environmental strain: Half of Gallup survey respondents opposing data centers cited environmental strain — cooling these massive facilities requires enormous water use.
3. A “techlash” narrative: Brookings frames this as tapping into deeper anger about income inequality — AI’s financial rewards concentrating among a small group of tech billionaires while ordinary communities absorb the environmental and cost burdens, even as tech firms pay relatively low tax rates.

Where It’s Playing Out Politically

Texas — the clearest reversal story
Gov. Greg Abbott, who celebrated Texas becoming an AI hub just last year, has now moved to halt roughly 1,800 data center projects, scaling back tax incentives and imposing new water/energy use limits — a stunning about-face from a Republican governor running for reelection.

Ohio — a Senate race flashpoint
The National Republican Senatorial Committee has warned that data center backlash could hurt GOP Sen. Jon Husted in his special election against Democratic challenger Sherrod Brown — prompting Husted to recalibrate his messaging. The NRSC has dropped multiple new ads addressing the issue in battleground states.

Florida — a primary election issue
Rep. Byron Donalds won Florida’s Republican gubernatorial primary while backed by crypto PACs but simultaneously proposing data-center restrictions — showing how even pro-tech Republicans are hedging.

Pennsylvania — executive action
Democratic Gov. Josh Shapiro signed an executive order imposing strict new standards on data center development in his state.

The scale: Newsweek reports the issue is playing a role in all six US Senate races currently rated as toss-ups by the Cook Political Report.

The Culture-Jamming Angle

Beverage company Liquid Death released a satirical ad mocking data centers’ water usage timed to a major primary election day — a sign the backlash has moved from policy circles into pop-culture mockery.

Trump’s Response

Trump has pushed back hard, posting on Truth Social that communities opposing data centers risk becoming “backwards and poor,” while insisting successful, “rich” communities should “let Data Reign.” He’s separately floated a “ratepayer-protection pledge“ requiring tech companies — not ordinary utility customers — to cover the power generation and grid upgrade costs their projects require.

The Political Fallout Angle

Axios reports this backlash caught the political establishment “flat-footed,” with Republican operatives privately frustrated that groups like the NRSC spent months fundraising from tech companies while under-resourcing candidates now facing attacks on the issue.

Historical Momentum

This didn’t come from nowhere — in 2025, Democrats flipped two Georgia Public Service Commission seats by 25+ points campaigning on rising utility costs, and Virginia Gov. Abigail Spanberger won partly on an affordability message tied to energy bills. In 2024, Warrenton, Virginia voters ousted their entire town council after it approved an Amazon data center.

What Was Announced

At the Hot Chips conference on August 25, 2026, OpenAI published its first real benchmark data for Jalapeño, its custom AI inference chip developed with Broadcom (chip/networking design) and Celestica (systems integration) — part of an October 2025 deal to co-develop 10 gigawatts of custom AI accelerators. The chip was first unveiled in June 2026, and OpenAI went from initial team hiring to manufacturing tape-out in roughly 16 months — an unusually fast development cycle for a custom chip.

The Benchmark Numbers

Tested against Nvidia’s Blackwell-generation systems (GB200 and GB300) on InferenceX, an independent public benchmark from semiconductor research firm SemiAnalysis, which physically visited OpenAI’s labs to verify the results:

• 1.5–1.9x more AI work per watt at peak throughput across three open-weight models (GPT-OSS 120B, DeepSeek R1 670B, Kimi K2.5 1T)
• 1.7–3.6x lower end-to-end latency
• On interactive workloads specifically, gains reached 2.1–4.1x, with interactivity hitting roughly 700–1,400 tokens per second per user
• Against an Nvidia GB200 running GPT-OSS 120B: Jalapeño hit ~85,448 tokens/kW vs. Nvidia’s 44,960, with 1.03-second latency vs. 1.80 seconds
• Jalapeño runs at 700 watts per package (drawing as low as 550W in testing) — roughly half the 1,200–1,400W of the Blackwell systems it beat

OpenAI’s head of hardware, Richard Ho, called it “a very, very significant performance advance over state of the art.”

Why It’s Purpose-Built, Not a GPU

Jalapeño isn’t a general-purpose chip — it’s an “intelligence processor” designed specifically around large language model inference (running trained models to serve responses), not training. It handles both major inference phases — prefill (processing the prompt) and decode (generating tokens) — in one architecture, and OpenAI used its own AI models to help with the chip design itself.

The Important Caveats

• This is inference-only, not a training chip, and won’t replace Nvidia — OpenAI explicitly said it will “continue to widely deploy accelerators from NVIDIA and other partners.”
• Deployment starts in “very small volumes” by end of 2026, with meaningful scale-up only in 2027.
• The comparison was against Blackwell — Nvidia’s next-gen “Vera Rubin” platform hasn’t been tested yet, so the lead could shrink by the time Jalapeño actually ships at scale.
• Some of OpenAI’s strongest claims (an even wider advantage on its own frontier models) are self-reported and haven’t been independently verified.
• A Gen 2 chip is already in development.

Why It Matters for the Bigger Picture

This lands squarely in the “Nvidia’s monopoly under threat” narrative — OpenAI is one of Nvidia’s largest customers, and it’s now publishing benchmarks beating Nvidia’s own hardware, all timed just one day before Nvidia’s own Q2 earnings report. It also puts OpenAI in the same company as Google, Amazon, and Meta, all of whom are building their own AI chips to reduce reliance on Nvidia. Analysts told CNBC this could specifically pressure Nvidia’s margins in the fast-growing inference market (as opposed to training, where Nvidia remains dominant) — and inference is significant because it’s the ongoing, recurring cost of actually running AI products at scale, unlike the one-time cost of training a model.

Investor Angle

This is also being read as a pre-IPO signal — since OpenAI is expected to go public eventually, demonstrating in-house infrastructure cost efficiency (cheaper compute per dollar) is a meaningful story to tell public-market investors ahead of any listing.