YouTube reached more than 20 million people aged 18 years and over in Saudi Arabia during September 2025, while the number of people reached via Internet-connected television devices exceeded 12 million people in the Kingdom during the same month, according to data announced by the platform in conjunction with the “Made on YouTube 2026” event.
The numbers come at a time when YouTube is expanding its tools for content creators and viewers with an increasing focus on artificial intelligence in editing, channel management, and identity protection, in addition to developing live broadcasting, interacting with the audience, and opening additional paths to commercial partnerships. In the region, the platform has also reached more than 45 million adults in Egypt, and more than 20 million in Morocco.
Artificial intelligence enters production stages
One of the most prominent updates is a new interactive editing tool within “Shorts” and the “YouTube Create” application, which allows the content creator to describe what he wants to implement in a conversational way, so that the system suggests modifications such as removing periods of silence or rearranging clips to improve the rhythm, while the possibility of manual modification remains available. It is scheduled to be released globally in English early next year.
The platform also expands the role of “Ask Studio” to become more of an assistant for channel management, as it provides summaries of content performance, explanations of the reasons behind changing indicators, and suggestions related to each channel. “Ask Studio” will also arrive in the “YouTube Studio” application on phones during the fall.
Another new tool is dynamic thumbnails, where the content creator can upload up to three images for one video, while the system chooses the image that it deems most appropriate for different categories of audience. YouTube says it will begin rolling out the feature to a number of content makers later this year.
Face and voice protection
In parallel with the increased use of generative artificial intelligence tools, YouTube is expanding tools to protect the identity of content creators. The “Likeness Detection” feature will be available on the mobile application, allowing you to register, receive alerts when matches are found, and take actions from the mobile device.
The company also intends to integrate English spoken voice recognition in the first phase, with facial feature matching to improve the system's ability to detect unauthorized use of a content creator's identity. The company's data indicates that billions of channels are eligible to register in the system for automatically detecting digitally fake content.
Live streaming is expanding globally
YouTube is also pushing to expand live streaming. According to its data, more than 600 million logged in users watched live content daily on average during August 2026, including archived recordings, while more than 40 percent of YouTube Live viewing time came from international audiences.
The new tools include “Live Showdowns,” which allows content creators to integrate their broadcasts into timed live competitions, in addition to the experience of instant audio dubbing for live broadcasts. The dubbing trial is scheduled to begin in the first quarter of 2027 with a limited number of content creators in English and Spanish. The platform also expands the “Watch With” feature to include long-form videos, while adding exclusive community spaces for members and tools for moderating comments powered by artificial intelligence.
“Shorts” is close to the soap opera model
The updates also reflect a change in the way short videos are presented. Views of “Microdramas” on YouTube exceeded 6.5 billion views during the first half of 2026, with viewing time growing by more than 50 percent year-on-year, according to the company.
That is why it launched the “Shorts Series,” which allows short clips to be organized into seasons and episodes with sequential playback and custom thumbnails, making following short works closer to the series experience.
Wider business opportunities in Saudi Arabia
YouTube also expands the “Creator Partnerships” program to 20 countries, including Saudi Arabia, with the aim of helping content creators appear in front of brands. It will also use Ask Studio to help creatives prepare responses to campaign briefs, build presentations, and customize their profiles according to the brand’s needs.
In the global version of “Made on YouTube” ads, the company said that it is expanding the affiliate shopping program to 35 countries by the end of the year, adding new partners and automatically updating product offers to show options more relevant to the local market from which the user is viewing.
Among the tools for production, protection, broadcasting, and monetization, YouTube is moving towards making artificial intelligence part of the content creator’s work stages themselves, from the first version of the video to choosing how to present it, analyzing its performance, protecting its owner, and then turning the audience into commercial opportunities.
A new day, a new announcement... about a pioneering artificial intelligence model - in fact: two models, one from Anthropic, and the other from OpenAI. So the pace of new model releases seems to be accelerating, and that is true, but many of these new releases are models that repackage (and often re-price) previous flagships with different capabilities, as Mark Sullivan (*) writes.
Consecutive releases
Both releases issued on Tuesday are an example of this. OpenAI announced two systems: GPT-6 Sol and GPT-6 Luna, which are faster and less expensive models that apply the developments that distinguish its flagship model, GPT-6 Astra, in “everyday use.”
Anthropic also announced the launch of the Claude Opus 5.5, which it says matches the performance of its leading model, the Claude Fable 5.1, in most missions, and its operating cost is about 40 percent lower than its predecessor, the Claude Opus 5. While SpaceXAI, on Monday, launched the Grok 4.7 model, which it says excels in programming and cognitive work, and can run longer missions and verify its outputs. But the Grok 4.7 is an all-new flagship model with new performance and capabilities, not an improved version of a previous model.
Pioneering new models
The pace of issuance of Anthropic's leading models has nearly doubled in 2026, from every 46 days in the first half to every 26 days so far in the second half. While the pace of “OpenAI” issuance increased from a model of every 46 days in the first half of the year to a model of every 51 days so far in the second half. But the time period between the launches of all-new flagship models hasn't changed much through 2026 for either company, as the release dates show.
Thus, this year, Anthropic released eight pioneering models, while OpenAI released six.
Cheaper and more specialized versions
A faster pace of releases does not necessarily mean faster improvement in the underlying technology. Often, the completion of one main model may lead to the emergence of a range of cheaper or more specialized versions aimed at different customer segments; This makes the pace of product launches a less reliable indicator of the speed of progress.
“Iterative self-improvement” generates new models
AI Labs reports that it is increasingly using AI itself to design and build new models, a practice known as “iterative self-improvement.” Anthropic reports that as of last August, its Cloud models were driving 26 percent of AI R&D, and collaborating on more than 90 percent of it.
Researchers have begun using artificial intelligence models to design new computing infrastructure that provides greater computing power, efficiency, and speed. They also use artificial intelligence to generate synthetic training data, and to improve the entire software framework that manages model training. They also use AI coding models to write and improve the code that defines the model, implements it.
OpenAI notes that before June 2026, the total uptime of its AI agents was less than human hours, but by mid-August, AI agents were working 3.1 days for every human working day (based on an eight-hour workday).
Innovation...or reassurance of financial markets?
Arnal Dayaratna, vice president of software development at research firm IDC, wonders whether labs have already begun to see the effects of true self-improving systems? He argues that any acceleration in model releases is driven by the AI industry as a whole better understanding what the market wants, as well as competitive pressure from open source model developers.
Arun Chandrasekaran, an AI analyst at Gartner, points out that model release dates may be affected by many different factors, not just the speed of innovation. “AI labs schedule releases to maintain market share, secure institutional clients, or shape investor expectations,” he says in an email to Fast Company. Frequent releases are an indicator of flexible development and provide evidence of an improved model price/performance ratio.”
Convince investors
Both OpenAI and Anthropic are seeking to go public within the next year, and this important event may impact model release schedules and marketing campaigns. Both companies have gone to great lengths to convince investors and the public that they have a reliable formula for systematically improving AI models until they are able to perform most human-level tasks, and then outperform them at almost everything.
Before listing, both companies will have to prove that the billions of dollars spent on computing and research produce a repeatable engine, not just a single breakthrough, Harrison Rolfes, a senior analyst at PitchBook, said in an email to Glass Company. “The tougher question for IPO investors is whether faster versions generate sustainable revenues and margins, or simply shorten the life of each model while keeping computing, security and infrastructure costs higher.”
* Fast Company magazine
An earphone can run an artificial intelligence application and connect to a cloud service directly via Wi-Fi without always relying on the phone. The glasses can process what the user sees locally, while the computer can run more demanding models, and the cloud remains ready when the task requires computing power that exceeds what is available on the device.
These are some of the examples that Qualcomm presented during its annual global conference, “Snapdragon Summit 2026,” in the American city of Hawaii, revealing a broader shift than simply adding more artificial intelligence capabilities to devices. The vision you are trying to build is based on distributing intelligence among a group of devices and data centers, so that the information and task are transferred to the most appropriate place for implementation instead of the experience being linked to a single device.
In one model presented, the glasses know what the user is looking at, the phone carries location information and its digital context, the watch collects signals related to movement and the body, while the speaker provides a continuous audio interface. The computer has more memory and computing power, while the cloud provides the largest models and more complex reasoning capabilities. The challenge becomes collecting these parts so that it appears to the user as a single experience.
Which device knows what?
In one of the sessions attended by Asharq Al-Awsat, Ziad Asghar, Senior Vice President and General Manager of Extended Reality, Wearables, and Personal Artificial Intelligence at Qualcomm, explains that each device around the user has a different “point of view.” In response to a question about what happens when several devices are active at the same time, and which of them has or creates the personal context, he explained that the company is working to collect the context coming from different devices at a low software level. The glasses see what the user sees, the phone has good location information, while the watch provides other types of data. He added that if the phone is present, it can play a role in gathering this context, and if it is not available, a watch, glasses, or headset can do this. Thus, Qualcomm is not necessarily looking for a new device to replace the phone, but rather for a system that can use the capabilities of several devices at the same time. Asghar said that today the phone usually represents one device for each person, while the personal artificial intelligence market could be different, as an individual owns several devices specialized in different uses, such as glasses for sports, for example, a watch, headphones, and perhaps another device for work.
From Google book to smart speaker
In computers, Qualcomm and Google announced “Googlebook,” a new class that runs on the Snapdragon X Elite platform and is designed around Gemini Intelligence. These devices support native Android applications and transfer between the phone and the computer, including accessing photos, files and downloads on the phone from the computer.
Google also offers functions such as “Magic Pointer,” which allows Gemini’s capabilities to be called within the context of the work itself, instead of moving to a separate application, while the system can benefit from the mail, calendar, and information that the user is working on.
On the smaller end of the spectrum is the Snapdragon Sound Elite Gen 2, which Qualcomm says is its first audio platform specifically designed to combine premium audio, local AI, and the connectivity required for proxy experiences. The platform integrates low-consumption Wi-Fi 6E connectivity that allows voice devices to directly access services and agents in the cloud. The company states that it provides up to twice the AI capabilities on the device compared to the previous generation, with up to a 40 percent reduction in power consumption in the testing scenarios identified by the company. The platform targets more than traditional headphones and includes headphones equipped with cameras, audio glasses, open-ear devices and other multimedia formats.
Glasses of different capacities
Glasses occupy an important place in the event because they are the device located directly in the user's field of view, Asghar explains. He adds that Qualcomm sees nearly three-fold growth annually in this category, but stresses that glasses are not the only expected form of personal intelligence; There are opportunities in headphones, pendants, rings and other forms of devices.
To accelerate the emergence of these products, Qualcomm is using the Snapdragon START program, which combines hardware modules with a software package and a network of manufacturing partners to help companies turn personal device ideas into products. The program began with smart glasses, and Inspecs was the first eyewear company to cooperate with it. In the main session of the summit, Asghar pointed out that work on the “Inspecs” model went from the beginning to a working device in about three to four months, providing an example of reducing the time needed to move from the idea to a ready model.
At a higher level of visual computing, Qualcomm also has the Snapdragon Reality Elite platform aimed at extended reality. The company says that it provides 48 trillion operations per second for artificial intelligence capabilities, and can run large models of vision and language locally, with its arrival on devices running the “Android XR” system, including “XREAL Project Aura.” This reflects the different roles even within the same category of glasses: light devices focused on sensing and context, and others with more computing power for extended reality experiences.
Who decides where the task runs?
But the presence of this number of devices raises a fundamental question: If glasses, headphones, phone, computer, and the cloud can all participate in the task, who chooses where it is carried out?
During the Computing Speech session, Qualcomm presented a vision of what it described as a “distributed intelligent fabric.” In this architecture, the cloud provides the largest models and deepest inference capabilities, while the edge and on-premises devices provide lower latency and closer context to the user, and the orchestration layer chooses the best place to execute each operation. The result may be that part of the task starts on the headset, another part goes to the phone or computer, and then the most demanding part is sent to the cloud. According to what the company presented, the goal is not to transfer the entire artificial intelligence from the data center to the device, but rather to make the system able to determine the correct location for each process at the right moment.
“Surface” expands local processing
The computer appears to be a candidate to play a larger role in this system due to its memory space, power, and ability to run continuous computer loads. A practical example appeared during the summit with the 12-inch Microsoft Surface Pro and the 13-inch Surface Laptop, both of which operate on the Snapdragon X2 Plus platform.
According to Qualcomm, the new devices provide 80 percent faster performance in local artificial intelligence inferences compared to the previous generation, in addition to more than 15 hours of battery life and an improvement of up to 55 percent in graphics performance, according to the company’s figures. These numbers are important because the value is not just increased performance, but rather keeping more work on the device when speed, continuity, or privacy is important. A Microsoft spokeswoman during the conference confirmed the same idea, saying that the issue is not a choice between the cloud and the device. She explained that some loads benefit from being local, while others need the size of the cloud, and the task may move between the two even within the same session.
Partnership with “Human”
This model gains a Saudi dimension through Qualcomm’s cooperation with HUMAIN, as the HUMAIN Horizon Ultra device runs on the Snapdragon X2 Elite platform. Qualcomm says its software extends across Linux and Windows while implementing intelligence
