Bots now outnumber humans on the internet for the first time in the web’s history. Cloudflare, the security firm that sits in front of millions of websites, says automated traffic crossed the 50 percent threshold in June and now accounts for nearly 58 percent of all webpage requests. A separate report from cybersecurity firm HUMAN Security found that traffic from AI agents that actively click links and fill out forms grew nearly 8,000 percent in the past year alone.

The finding is not an outlier. Thales, the French defence and technology group, published its own Bad Bot Report showing bot traffic hit 53 percent in 2025, with 40 percent of all web activity classified as malicious. Cloudflare CEO Matthew Prince had predicted in March that the crossover would not arrive until late 2027, but agentic AI programs browsing on behalf of assistants like ChatGPT and Gemini accelerated the shift by more than a year.

The implications run deep for any company built around human visitors. Stripe says 70 percent of the commands used to access its API now come from agents, while API brokerage firm Alpaca reported that agent-driven calls grew from single digits to 30 percent of monthly volume in a single quarter. Cloudflare launched a tool called Precursor this month that watches entire browsing sessions rather than single checkpoints, an acknowledgment that traditional bot detection built around CAPTCHAs can no longer keep pace.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The scale of the shift remains hard to measure precisely because no single provider sees the entire web, and agentic browsers that mimic human behaviour routinely evade detection. A University of Bamberg study found that existing bot-detection systems misfire on real traffic at rates between seven and 15 percent, before accounting for the reverse problem of agents slipping through entirely undetected. Cloudflare, Mozilla, Google, and Microsoft are now developing a privacy-first protocol called PACT to verify traffic legitimacy without tracking users, a sign that the industry recognises its current tools are falling behind.

The web’s business model, built on ad impressions, conversion funnels, and analytics that assume the visitor is a human being, is being tested by a class of traffic it was never designed to handle. Cloudflare has set a September deadline for AI crawlers to pay publishers or be blocked from ad-supported pages, the most aggressive attempt yet to force the emerging machine economy to compensate the humans who create the content it consumes. A PitchBook report estimates that only about one percent of the roughly $20 trillion in work addressable by AI agents is currently flowing through them, suggesting the flood has barely started.

I have a QUAD problem.

An original Vena still lives in my system, the QUAD 3 was one of the easiest Editors’ Choice decisions I have made in years, and the ESL 2912X was among the most compelling loudspeakers at AXPONA 2026. I am not pretending to approach the new $995 Elegan 1 and $1,995 Elegan 2 without expectations.

That does not mean they receive a free pass. It means QUAD has my attention.

The two model Elegan Series sits below the Revela and ESL ranges, giving the British manufacturer a more affordable loudspeaker family designed to partner with the QUAD 3, revived 33 and 303, and other reasonably powered amplifiers. Both models are available now in matte black or real walnut veneer. 

QUAD shaved the final “t” from elegant, but apparently not much from the engineering budget.

Related Reviews:

The QUAD L Series Connection

The Elegan speakers are entirely new designs, but QUAD describes them as spiritual successors to the 11L and 12L models introduced in the early 2000s.

Those compact Kevlar equipped loudspeakers helped expose a much broader audience to QUAD without asking them to rearrange the living room around a pair of electrostatic panels. The Elegan 1 follows that same basic idea, while the larger Elegan 2 expands it into a proper three way floorstander. 

Both models were developed under Peter Comeau, QUAD’s Director of Acoustic Design, who has spent the past several years rebuilding the company’s loudspeaker and electronics ranges rather than merely polishing old logos.

The Elegan speakers are not smaller Revelas and certainly not conventional versions of the ESLs. They are simpler, less expensive dynamic loudspeakers built around QUAD’s long experience with woven Kevlar cones.

Woven Kevlar Returns

QUAD has used Kevlar since the original 11L and 12L arrived in 2002. The new lattice woven cones are impregnated with a self damping resin designed to control resonance while retaining the stiffness and low mass that made Kevlar useful in the first place. 

Both Elegan models use a 25mm fabric dome tweeter mounted inside a profiled waveguide. QUAD says the waveguide improves dispersion and helps the tweeter blend more consistently with the midrange or midbass driver beyond one narrow listening position.

That matters in an actual living room, where more than one person may want to hear the music without sitting perfectly still beneath a laser alignment system.

QUAD Elegan 1 Specifications

The Elegan 1 is a rear ported, two way standmount loudspeaker intended for small to medium sized rooms.

• Tweeter: 25mm fabric dome with waveguide
• Midbass Driver: 165mm woven Kevlar cone
• Sensitivity: 86dB at 2.83V/1m
• Nominal Impedance: 8 ohms
• Minimum Impedance: 4.3 ohms
• Frequency Response: 56Hz to 24kHz, ±3dB
• Bass Extension: 44Hz, minus 6dB
• Crossover Frequency: 2.9kHz
• Recommended Amplifier Power: 30 to 150 watts
• Maximum Output: 102 dB
• Dimensions: 345 x 222 x 333 mm (13.6 x 8.7 x 13.1 inches)
• Weight: 8.3 kg (18.2 lbs) each 

The 86dB sensitivity means the Elegan 1 is not intended for a two watt amplifier and positive thinking. A stable integrated amplifier delivering 40 to 100 watts per channel should make far more sense.

The QUAD 3, with 65 watts per channel into 8 ohms and 100 watts into 4 ohms, appears to be an obvious partner on paper. 

QUAD Elegan 2 Specifications

The Elegan 2 is a three way, rear ported floorstander designed for larger rooms and higher playback levels.

• Tweeter: 25mm fabric dome with waveguide
• Midrange: 125mm woven Kevlar cone in an isolated chamber
• Woofers: Two 165mm woven Kevlar cones
• Sensitivity: 90dB at 2.83V/1m
• Nominal Impedance: 8 ohms
• Minimum Impedance: 3.0 ohms
• Frequency Response: 44Hz to 24kHz, ±3dB
• Bass Extension: 32Hz, minus 6dB
• Crossover Frequencies: 480Hz and 2.7kHz
• Recommended Amplifier Power: 30 to 200 watts
• Maximum Output: 106 dB
• Dimensions: 930 x 310 x 373mm (36.6 x 12.2 x 14.7 inches)
• Weight: 27.8 kg (61.3 lbs) each 

The 90dB sensitivity looks friendly, but the 3 ohm minimum impedance means amplifier quality still matters. The Elegan 2 does not require 300 watts, but it will benefit from an amplifier with a healthy power supply and stable current delivery.

The QUAD 3 should work in smaller and medium sized rooms, while the 303 power amplifier, Platina Integrated, Audiolab 7000A class of amplifier, or comparable 70 to 150 watt designs would provide more breathing room.

Crossover & Cabinet Design

Both speakers use dedicated Linkwitz Riley crossover networks developed through computer modeling and extensive listening tests.

The Elegan 2 uses fourth order slopes between the midrange and tweeter, with second order slopes between the bass drivers and midrange. Polypropylene capacitors are used throughout, supported by air core inductors in the midrange and treble sections and laminated core inductors for bass. 

The cabinets use high density panels, internal bracing and damping material, along with QUAD’s rear Airflow port system. The Elegan 2 gives its dedicated midrange driver a sealed internal chamber, preventing pressure from the two woofers from interfering with the vocal range. 

What Makes the Elegan Series Different?

The obvious answer is not merely Kevlar. Wharfedale and other IAG brands also use it.

The Elegan Series is different because it creates a sensible bridge between QUAD’s electronics and its much more expensive loudspeakers. Someone buying a QUAD 3 or 33 and 303 system no longer has to look elsewhere for a price appropriate loudspeaker or jump directly to the Revela range.

The other differentiator is intent. QUAD is promising tonal balance, vocal body, natural dynamics and long term listening comfort rather than an exaggerated top end designed to impress during a 12 minute dealer demonstration. That is the manufacturer’s claim, and we will reserve judgment until a review pair arrives. 

Competitors

At $995 per pair, the Elegan 1 enters an extremely busy category.

The KEF Q3 Meta costs $899.99 and uses KEF’s coaxial Uni Q driver with Metamaterial Absorption Technology, while the Wharfedale EVO 5.1 costs $999 and substitutes an AMT tweeter for QUAD’s fabric dome.

The Elegan 1 counters with a larger Kevlar midbass driver, real walnut veneer, simpler amplifier loading than some competitors and a design intended specifically around two channel listening.

The $1,995 Elegan 2 faces the Q Acoustics 5040 at $1,999, Wharfedale EVO 5.3 at $1,899. Those are all established competitors with very different approaches to bass, dispersion and treble reproduction. 

Who Are They For?

The Elegan 1 is aimed at listeners with small to medium rooms who want a compact speaker for stands or substantial furniture and already own a capable 40 to 100 watt amplifier.

The Elegan 2 is the better choice for medium to larger rooms, listeners who do not want a subwoofer for most music, and amplifiers capable of remaining comfortable when the impedance drops to 3 ohms.

The Bottom Line

QUAD did not need another expensive loudspeaker.

It needed a pair of attainable models that could sit naturally beside the QUAD 3, 33 and 303 without feeling like the economy cabin at the back of the range.

The Elegan 1 and Elegan 2 appear to deliver that on paper. Peter Comeau has combined familiar Kevlar drivers, fabric dome tweeters, properly engineered cabinets and sensible crossover networks at prices that put QUAD directly against KEF, Wharfedale, Q Acoustics and even Ruark.

The Elegan 1 may prove to be the more important model because compact two way speakers around $1,000 remain the gateway into better audio for a lot of listeners. The Elegan 2 has the harder assignment, but its 90dB sensitivity, dedicated midrange and 32Hz claimed bass extension make it intriguing.

I love QUAD, but I also know what these prices buy from the competition.

Price & Availability

The QUAD Elegan Series is available now in matte black or real walnut veneer.

• QUAD Elegan 1: $995 / £749 / €999 / AU$1,699 per pair
• QUAD Elegan 2: $1,995 / £1,499 / €1,899 / AU$3,499 per pair

Related Reading:

Millions now choose products through ChatGPT, Gemini, and Perplexity, and Google AI Overviews appear in nearly half of all search queries. According to Muck Rack, 84% of AI citations come from media outlets, the same outlets where SEO specialists have been placing links for years.

That means media distribution remains one of the key levers for influencing both AI and Google results. Manual outreach to media outlets doesn’t scale well: pitching, checking metrics, and negotiations are too time-consuming.

Best Platforms for AI visibility and Link Building

We’ve evaluated multiple platforms that let companies boost their AI visibility and link building performance with paid placements in media outlets. Instead of focusing on standard features like price on publishing conditions, we highlight their unique features that allow companies to achieve their goals in the way aligned with their overall strategy and approach.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Here are the top 10 platforms that we handpicked.

1. Medialister

Medialister was built for running large-scale PR and AI visibility campaigns, with a catalog of 106,000+ vetted outlets worldwide. Scalability is built into the platform itself: add team members and assign roles to run dozens of campaigns with multiple placements in parallel.

Notably, the platform helps match outlets to a specific business goal. You can connect your Google Search Console to get recommendations, set your goal directly in the catalog, or use Medialister’s MCP. Through MCP, AI assistants find outlets, draft tailored guest posts and press releases, submit them, and track indexing in Google.

To avoid a cash flow gap, agencies can use the post-payment option: they’ll receive a consolidated invoice at the end of the month instead of paying per placement.

2. PRNEWS

PRNEWS is an international platform for paid media placements with one of the largest catalogs: 107,000+ media outlets across 175 countries, from global-scale publications to hyperlocal media outlets targeting specific cities. Forbes alone is represented here with 20 different country editions. You can use a dozen filters to navigate that volume. If you don’t want to filter anything yourself, you can use goal-based presets: for example, for AI visibility or fintech PR. Advanced filters immediately show which outlets accept sensitive topics like gambling and crypto or allow tracking pixels.

For teams that need more, PRNEWS offers a PRO account. Its Bulk Site Checker lets you upload domains a competitor has published on and secure placements there. The entire catalog with prices and metrics can be exported to CSV so that agencies can calculate budgets for several clients at once.

3. PRposting

PRposting is a platform with a catalog of 23,000+ outlets. Its distinguishing feature is that around 20% of its media are exclusive: these sites are added to the database manually after direct negotiations with owners, bypassing intermediaries and exchanges.

It provides a free audience breakdown for each outlet across its 5 key source countries. If a publication writes in Spanish but most of its traffic comes from the US rather than Latin America, that becomes immediately clear. If a placement doesn’t go live on time, you receive a full refund within 7 days.

4. INSERT.LINK

INSERT.LINK is a highly specialized service for fast link integration into already-published articles. There are two ways to find a spot for your link: choose an outlet by topic, as in a regular marketplace, or enter a keyword to find a specific, already-published page on that topic among 50 million indexed pages. The catalog includes 30,734 sites across 152 countries and 15+ niches, with formats spanning link insertions, guest posts, listicles, press releases, and PR publications.

Agencies get MCP and API access. You can connect this platform to your internal tools to run searches and place orders programmatically. There’s also a separate list of outlets with guaranteed publication within 3 days, for time-sensitive campaigns.

5. PressWhizz

PressWhizz is a marketplace for guest posts and link insertions with a catalog of 41,000+ sites across 90 countries. The platform has a proprietary AI Visibility Seeding™ method: the team studies hundreds of real user prompts on a given topic and places the brand in sources already trusted by ChatGPT, Perplexity, and Google AI Overviews. In addition, they offer traditional link building services.

If necessary, you could buy a managed service, so that the platform’s team will handle outlet selection and negotiations with editorial teams.

6. Linksman

Linksman is a marketplace for guest posts, niche edits, and link building with a catalog of 9,000+ sites across 25+ niches. After signup, you can see DR, traffic, and price for each media outlet. You can also choose a ready-made package that includes several guest posts at a fixed price, split by DR tier. The higher the authority, the higher the price.

It’s worth mentioning the platform’s free SEO tools. You can check whether a purchased link is still live and providing value.

7. Respona

Respona isn’t a marketplace with a ready-made catalog of outlets. It’s a SaaS tool for self-service outreach. This means you still have to negotiate with each newsroom or journalist yourself, but the service automates everything else, from finding publications and authors to personalizing pitches with AI-generated snippets from the recipient’s content.

This platform is used for link building, digital PR, affiliate partner sourcing, and blogger outreach, offering a different set of outlets and templates for each use case. Unlike others on this list, the service runs on a subscription, with pricing based on campaign volume and team size.

8. Ereferer

Ereferer is a catalog of 45,000 sites across five languages. The service’s standout feature is the ability to save several media outlets into a single project, say, three blogs and one directory site, and have the platform automatically distribute their publication over time so they don’t go live all at once.

Group buying is another distinct feature. The cost of a single placement is split between two clients, and the article itself carries two separate links, one for each.

9. NO-BS Marketplace

NO-BS Marketplace is a catalog of sites for guest posts and link insertions with 110,000+ publishers. Its database can be flexibly filtered by niche, domain authority, traffic, and price. The platform supports a wide range of goals, from classic link building to digital PR and promotion in AI-generated answers, helping brands land in the sources that generative AI cites.

If you already know what you’re looking for, you can select outlets from the catalog yourself. For agencies and larger projects, there’s a NO-BS Pro plan, which adds a dedicated account manager, volume discounts, and white-label reports.

10. LinkBuilder

LinkBuilder is a price-comparison aggregator for guest posts and link insertions, comparing offers from 50 vendors across 300,000 sites in a single interface. Domains can be uploaded in bulk via CSV for comparison. The service is useful once you’ve already chosen an outlet but want to find the best possible price.

They also have a Check Link Price extension for Chrome that overlays placement pricing for a specific domain right on top of Google, Ahrefs, Semrush, Moz, Majestic, and Serpstat results. You can check prices without ever leaving the tools you’re currently using.

The choice of platform depends on your goal

None of these platforms will replace a proper strategy that starts with the goal. For Google rankings, the number and authority of links matter most. For AI visibility, it’s the outlet’s reputation and consistent mentions in expert content that count.

First, set the goal, then choose the platform. Their value lies in speed because they save time you could waste on manually outreach and negotiations.

Christopher Nolan’s The Odyssey in Imax 70mm is the hottest club in town, if you can get in.

While Nolan’s adaptation of Homer’s epic is available on thousands of screens around the world, there are only 41—including 25 in the US—that can offer it in full Imax 70mm glory. Many of those are sold out for weeks. Levels of FOMO are off the charts.

That’s partly because The Odyssey is the first feature film shot entirely on Imax film with Imax cameras. For the uninitiated, the Imax film format takes an already large 65-mm film negative and runs it through the camera horizontally, resulting in an image 15 perforations wide compared to the standard 5 perf. When projected on a giant Imax screen, the resulting image, with a squarish aspect ratio of 1.43:1, is enormous and incredibly sharp, creating a far more immersive experience than standard cinema offerings. This is real Imax, according to fans of the format, which was first introduced in 1970.

Imax, though, comes in a variety of flavors: digital formats in 2K and 4K resolutions, xenon and laser light configurations, on top of the classic 15/70-mm. Outside of a select few, most digital Imax theaters crop the image to a wider 1.90:1 aspect ratio and project on screens a good deal smaller than a traditional Imax giant theater.

“This all started in 2008 when Imax introduced its first digital system,” says James Hyder, who managed the Smithsonian Air and Space Museum’s Langley Imax theater in the ’80s, and ran the giant-screen industry journal LF Examiner from 2008 to 2021. Imax did not respond to requests for comment.

The company’s foray into digital exhibition with a 2K-resolution dual-projector system was part of a long-planned shift. Executives wanted to expand beyond museums and science centers and into big theater chains, which didn’t always have the screen real estate the traditional Imax experience demands.

The result, though, has been notably different Imax experiences depending on what type of Imax theater you’re in. Especially as major theater chains have invested in PLF (Premium Large Format) screens that rely on Dolby Vision or other systems, it’s hard to know what you’re going to get, or if Imax is even your best bet.

Complicating things further for The Odyssey is that it’s available in so many premium formats—six, altogether—that the film’s official website includes a guide to their various strengths and aspect ratios. While it disambiguates between 70mm Imax and “regular” Imax, it doesn’t capture the wide range of what those “regular” experiences can entail.

“It’s a disservice to their customers, and it’s basically trading on a reputation for real giant screens that used to be what Imax stood for,” Hyder says of the lack of clarity and consistency around how Imax is implemented.

Recent years have seen a resurgence in shooting and projecting on film, with Imax 70mm in particular becoming the gold standard. Nolan’s films regularly get Imax 70mm releases on the few screens that can accommodate it, but following the success of Oppenheimer, a string of releases, including Dune: Part Two, Sinners, One Battle After Another, and Project Hail Mary, have followed suit.

While The Odyssey was shot in a 1.43:1 aspect ratio, its images were composed to function just as well when cropped to other ratios, all the way down to the very wide 2.39:1. It was intentionally shot to work across digital, 35-mm, and regular 5-perf/70-mm, all overseen by the director and reflective of his intent. Eventually, it will also be available to buy on Blu-ray or stream online. It’s not unlike the difference between seeing a print or digital image of the Mona Lisa—which most people will in their lifetimes—and actually being lucky enough to see it in person at the Louvre.

Samsung Electronics and Broadcom have signed a memorandum of understanding worth more than $200 billion spanning memory chips, foundry manufacturing and advanced packaging for artificial intelligence semiconductors. The deal, announced on Friday at an AI summit in San Francisco, runs for five years through 2030 and covers some of the most critical layers of the AI chip supply chain. It is one of the largest semiconductor partnerships ever disclosed, though as an MOU it represents a statement of intent rather than a binding contract.

On the memory side, Samsung will supply Broadcom with next-generation high-bandwidth memory, including HBM4 and HBM4E chips, for use in Broadcom’s AI accelerators. In foundry, Samsung will manufacture Broadcom products using two-nanometer and smaller process technologies at its campus in Pyeongtaek, South Korea. The collaboration also extends to advanced packaging that integrates memory and logic more tightly, a technique increasingly central to squeezing more performance out of AI silicon.

Broadcom designs custom AI chips for hyperscalers including Google, Meta and others that want silicon tailored to their specific workloads rather than buying off-the-shelf GPUs from Nvidia. The company extended its AI chip partnership with Meta through 2029 in April, and CEO Hock Tan recently said surging AI revenue has made organic growth more attractive than acquisitions. By diversifying manufacturing to Samsung, Broadcom gains supply chain resilience beyond its existing reliance on TSMC, which controls more than 90 percent of leading-edge chip production.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

For Samsung, the deal addresses a persistent credibility gap in its foundry business, where its global market share has fallen to roughly seven percent while TSMC dominates. The company has struggled with yields on its two-nanometer process, but a $200 billion customer commitment from one of the industry’s largest chip designers could change the calculus for investors who have questioned whether Samsung can compete at the leading edge. Samsung posted record quarterly operating profit of 89 trillion won in Q2, driven almost entirely by AI memory demand.

The signing took place alongside a broader package of semiconductor deals worth roughly $950 billion between South Korean and American companies, announced during an AI summit hosted by South Korean President Lee Jae Myung. SK Hynix signed memory partnerships worth $750 billion with Nvidia and others, and Anthropic signed supply agreements with both Samsung and SK Hynix. The summit followed Seoul’s $880 billion domestic investment plan unveiled in June, which committed Samsung and SK Hynix to building four new chip fabrication plants in the country’s southwest.

OpenAI’s facing a firestorm of late: a rogue agent hacking Hugging Face, a lawsuit launched by Apple accusing OpenAI of stealing trade secrets, and ongoing anger about AI and AI devices in general, plus a delayed IPO. 

In the midst of all of this, OpenAI President Greg Brockman spoke to reporters in New York Thursday to discuss the company’s latest AI plans and recent events. But he also dropped a few hints on the company’s wearable and hardware aspirations, and where it could overlap with some voice-focused advances in ChatGPT. And it won’t just be through OpenAI’s own hardware: expect other smart glasses and wearables to eventually get better access, too.

A demonstration of ChatGPT Live – a recently improved voice interface that allows talking, listening and interruptions to happen more seamlessly in ways that are similar to Anthropic’s new voice modes for Claude, and Meta’s Muse Spark – led to discussions on OpenAI trying to make a voice assistant like Iron Man’s AI agent, Jarvis. And if the goal is a direct assistant/agent to do everything, would that agent be able to leap outside of ChatGPT into controlling other applications, too?

Brockman said that’s exactly the plan, similar to how OpenAI’s Codex already allows agentic control of whole computers.

“[ChatGPT Live] is coupled to the Codex app, but it doesn’t really have to be. It doesn’t really make sense for it to be,” he said. “And if you think about it, the Codex app also can do full computer use, which is also going to get very, very good.”

That means that a lot of the manual keyboard work we humans have been doing could potentially be taken over by AI.

“I think we’re just going to realize this phase of us, like, clicking things and typing things, it was a phase,” he said. “It was never what we wanted to be.”

Which leads to questions of where the guardrails would be built to prevent incidents like OpenAI’s rogue agent cyberattack of Hugging Face, a New York-based AI software company, this week, where an OpenAI agent broke out of a testing sandbox and started acting on a repository of Hugging Face’s AI tools.

Deeper OpenAI tools for wearables on the way?

But what interests me is how AI wearables right now, like smart glasses, are lacking truly capable AI agents that can do anything beyond limited interactions with walled-in phone apps, something that might change soon when Google and Samsung launch their Gemini-enabled eyewear this fall. That, too, is changing fast, though: Meta just introduced one of its first Muse Spark agentic updates Friday.

OpenAI has been rumored to have hardware in its development mix for years, which may get held up now because of the Apple lawsuit, which alleges that former Apple employees now working at OpenAI stole hardware trade secrets. But, clearly, whatever Jarvis ambitions OpenAI has, its hardware is going to be a key part of its evolving vision of what that could feel like in a more unleashed form.

“We’re definitely very interested in integrations, and I would view it as one of the use cases of the API and getting that out there that anyone will be able to integrate,” Brockman told me at the New York press roundtable. 

Brockman confirmed that OpenAI very much wants to make those moves, but right now, those possible APIs are in very early phases. “We think this should be the interface and should be available on all devices.”

New APIs are needed because wearable devices involve different types of access and interfaces than phones and laptops, often needing to run in the background on phones the wearables are paired to. Opportunities for multimodal inputs like camera access are also key to how most smart glasses now tap into AI.

Among existing smart glasses and AI wearables, almost none have deep direct integrations with multiple AI platforms in ways that could allow the sort of more comprehensive controls that phone and desktop AI apps have.

Also, notably, most of the largest work-focused AI platforms aren’t present on wearables yet, either. Gemini is on its way in Android XR glasses, but OpenAI and Anthropic remain out of the wearables picture. Some wearables do use OpenAI models for their own custom or anonymized tools, but they’re not the same as direct access to ChatGPT or Codex via user accounts.

OpenAI’s own hardware will take a unique path

But Brockman also suggested that OpenAI’s own hardware plans may be quite different from what’s already out there in wearables, calling it “a solution for a problem that we think is neglected, for a problem that we care about a lot, and that we think will be representative of the general direction [of AI],” adding that OpenAI has “a specific angle that we think that may not be otherwise served.”

Reports have pointed to that first device being a type of speaker with a semirobotic camera. Maybe it also will have some sort of optic identification, like OpenAI’s World ID, which also exists on its odd Orb iris-scanning device. And maybe, simultaneously, OpenAI pursues opening up its services to wearables like pendants and glasses, too. 

Before any of those happen, there are also a lot more current problems with AI that need solving – privacy, safety, resource drain, economic feasibility, questions of governmental involvement, and a major lawsuit. But AI wearables in 2026 are already advancing fast, even in the headwinds of challenges like these.

Scott Stein

Editor at Large

I started with CNET reviewing laptops in 2009. Now I explore wearable tech, VR/AR, tablets, gaming and future/emerging trends in our changing world. Other obsessions include magic, immersive theater, puzzles, board games, cooking, improv and the New York Jets. My background includes an MFA in theater which I apply to thinking about immersive experiences of the future.

See full bio

Advertisement

The Roku Ultra now costs $150 instead of $100.

Roku is increasing the cost of its popular streaming players by up to $50, with one report suggesting that the AI-influenced “RAMageddon” is to blame.

In an unusual move, Roku lists the existing prices on its website as “limited time,” but hasn’t specified when the new prices listed will kick in.

The biggest increases are for Roku’s flagship devices — the Roku Ultra and Streambar SE — which will increase in price from $100 to $150. Meanwhile, the popular Roku Streaming Stick 4K now costs 60% more, at $80.

The price changes are as follows:

In my review of the Roku Ultra, I call it “undoubtedly the best Roku yet,” though I identify the identically priced Google Streamer as even better. But with a 50% cost increase, the Roku device now becomes less recommendable, and if you’re looking for a sub-$100 player, you should get the Google option while it stays at its current price. 

The new prices follow an even more dramatic increase for the Apple TV 4K, which went from $129 to $199, while the 128GB version will now cost $249. Apple also blamed rising memory costs for its price hikes.

RAMageddon refers to the global memory-chip supply shortage caused by the rapid development of memory-intensive artificial intelligence tools, which has pushed up the price of laptops, phones, graphics cards and other devices.

Roku is the most popular streaming platform, with up to 100 million streaming households, and the company was recently acquired by Fox for $22 billion.

Representatives for Roku did not immediately respond to a request for comment.

TV and home video editor Ty Pendlebury joined CNET Australia in 2006, and moved to New York City to be a part of CNET in 2011. He tests, reviews and writes about the latest TVs and audio equipment. When he’s not playing Call of Duty he’s eating whatever cuisine he can get his hands on. He has a cat named after one of the best TVs ever made.

See full bio

Tesla now has nearly one and a half million active Full Self-Driving subscribers after adding 200,000 users in the second quarter, the largest single-quarter gain on record. CFO Vaibhav Taneja told analysts that 55 percent of North American deliveries in Q2 had FSD activated at the time of purchase, a quarter in which Tesla delivered more than 480,000 vehicles worldwide. Year-over-year subscriber growth stands at 56 percent.

The subscriber mix is shifting decisively toward the monthly model, with about 45 percent of current FSD users now paying the $99 subscription after Tesla eliminated the one-time purchase option on February 14. The software had previously sold for up to $15,000 as a lump sum and later $8,000 as an add-on. Removing that upfront barrier has turned FSD into a recurring revenue stream worth more than $65 million per month at current take rates.

Year-over-year subscriber growth stands at 56 percent, and the software is now available in 12 countries after a rolling European expansion that began with the Netherlands in April and has since reached Lithuania, Estonia, Denmark, and Belgium. Tesla also discontinued its basic Autopilot system, meaning new vehicles ship only with adaptive cruise control unless the buyer adds FSD.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The numbers are significant but come with a caveat that Tesla’s own branding obscures. FSD remains a supervised system that requires the driver to keep hands on the wheel and eyes on the road at all times. The unsupervised version exists only in Tesla’s small robotaxi fleet operating in Austin, Dallas, and Houston, where roughly 40 vehicles handle rides without a human behind the wheel.

CEO Elon Musk’s compensation package ties a milestone of 10 million active FSD subscribers, sustained for three consecutive months, to a future payout. At the current pace of roughly 200,000 additions per quarter, reaching that target would take more than a decade without a sharp acceleration in adoption or a significant expansion of the vehicle fleet, which stands at nearly 10 million lifetime deliveries worldwide.