AI Agent Phones Arrive: Apps, Payments and the Gulf Adjust
AI Agents & Mobile12 min readJuly 10, 2026

AI Agent Phones Arrive: Apps, Payments and the Gulf Adjust

nubia, OpenAI and Qualcomm are racing to replace apps with AI agents on the phone, while Visa, Mastercard and Saudi Arabia race to build the rails underneath.

01

The Phone Stops Being an App Screen

The Phone Stops Being an App Screen

For a decade and a half, the smartphone was essentially a launcher for apps, a grid of icons we tapped to hail a ride, move money, or order dinner. That model visibly started cracking in the summer of 2026, and this is not a stage-keynote hypothesis anymore. China's nubia is preparing to unveil what it calls the world's first AI agent smartphone, built around OS-level autonomous operation, at the World Artificial Intelligence Conference in Shanghai running July 17 to 20. At the same time, Qualcomm chief executive Cristiano Amon has been telling investors plainly that the phone will not remain the primary device forever, while OpenAI is reportedly working with MediaTek, Qualcomm, and manufacturing partner Luxshare on hardware of its own. The premise of this piece is straightforward. What is unfolding is not a cosmetic update to another app. It is a full renegotiation of who owns the relationship with the user: the phone maker, the app developer, or the company behind the AI model running underneath everything. And the Gulf, among the most mobile-first, app-dependent markets on earth, will be one of the first regions to feel this shift in practice, whether through Riyadh's bet on domestic chip design or through HUMAIN Chat's attempt to become the Arabic layer of this new world.

02

nubia in Shanghai: Who Fires the First Shot

nubia in Shanghai: Who Fires the First Shot

nubia, owned by China's ZTE, is marketing its coming device as the successor to the M153, which ran a Snapdragon 8 Elite chipset with 16GB of RAM and 512GB of storage. This time, though, the differentiator is not silicon but software. According to a GSMArena report citing an industry tipster, the new device will carry what is described as deep system-level AI integration, plus the ability to operate autonomously across platforms without the user directing every step. nubia announced the device through its official Weibo account, calling it a pioneer of AI-native phones, with full specifications due at the World Artificial Intelligence Conference in Shanghai, scheduled for July 17 to 20. The symbolism here matters more than the marketing line. nubia is not a brand that leads Apple or Samsung headlines, yet it chose to be the one testing this shift in public rather than waiting for the giants to settle their positions first, a familiar pattern in Chinese hardware where bold launches probe the market before the larger brands follow with more caution.

03

Why OpenAI Wants Hardware of Its Own

According to TechCrunch, citing supply-chain reporting, OpenAI is working with both MediaTek and Qualcomm on chips for a phone of its own, with Chinese manufacturer Luxshare as a production partner. Final component specifications are expected to be locked in sometime between late 2026 and the first quarter of 2027, while mass production is not expected to start before 2028. The motivation is less technical than strategic. ChatGPT is now approaching nearly a billion weekly active users by widely cited estimates, and an audience that size makes it increasingly awkward for OpenAI to remain hostage to Apple's and Google's app store rules every time it wants to ship a new feature. Analyst Ming-Chi Kuo, known for closely tracking hardware supply chains, has said the device would lean on AI agents to complete tasks rather than depend on separate conventional apps, giving OpenAI access to far richer behavioral data than any app installed on a phone controlled by someone else could offer. Technically, the device is expected to blend small on-device models for instant, offline-capable tasks with heavier cloud models for anything more demanding. OpenAI has not officially commented on any of this reporting as of writing.

04

Qualcomm: Amon Prepares the Market for the End of Phone-Centrism

There is something almost paradoxical about the fact that the man running the world's largest smartphone chip supplier is the one publicly signaling the decline of phone-centrism. Qualcomm CEO Cristiano Amon told Fortune plainly: this year, 2026, we've been saying that it's going to be the year of agents, adding that the smartphone is not going anywhere, but many things will become natural to do without going through it at all. Amon expects personal AI devices to appear toward the end of 2026, with real adoption scaling between 2027 and 2028, growing from tens of millions of units to hundreds of millions within five years. More telling than the rhetoric is the target behind it: Qualcomm wants non-mobile business to reach roughly 50% of revenue by 2029, close to $22 billion in business entirely outside the phone category. Current numbers show exactly how large a gap that is. In the fourth quarter of fiscal 2025, ended September 28, the handset unit posted $7.0 billion in revenue, up 14% year over year, while automotive brought in just $1.1 billion despite 17% growth, and IoT contributed $1.8 billion on 7% growth. In other words, handsets still generate roughly twice the revenue of Qualcomm's other segments combined, which is exactly why Amon is so bullish on smart glasses as the natural heir, arguing that humans are very comfortable with glasses, a position close to the eyes, ears, and mouth that gives an AI agent constant context about its user's surroundings.

Qualcomm QCT Revenue by Segment, Q4 FY2025

Source: Futurum Group, Qualcomm Q4 FY2025 results

05

Apple's Bind: The App Store in the Age of Agents

Apple's Bind: The App Store in the Age of Agents

Apple has landed in an uncomfortable spot: the rules it built its App Store on for a decade and a half now collide directly with the wave the company itself claims to be embracing. In March 2026, Apple blocked updates to a category of so-called vibe coding apps, tools that let users build mini-applications through natural-language prompts and AI agents, on the grounds that they violated a long-standing rule barring any app from executing code that alters its own functionality or that of other apps. The trouble is that this same rule, originally written to keep malware out of a trusted storefront, structurally conflicts with the idea of an agent capable of advanced autonomous action. According to MacRumors, Apple is now developing a new framework to host AI agent apps and AI coding tools while preserving the security and privacy standards it markets so heavily, discussing parts of this framework at its developer conference on June 8. The stated motivation is defensive rather than promotional: preventing what reports describe as rogue AI agents deleting user content or causing other damage without clear permission. The irony is that Apple, already late in shipping its AI-overhauled Siri, now finds itself rewriting the constitution governing its entire storefront on a tight timeline, while still trying to remain the industry's strictest gatekeeper.

06

Who Pays the Agent: Visa and Mastercard Build the Plumbing

If the phone is shifting from an app interface to an agent acting on the user's behalf, one plain question needs an answer: who pays, and how do you guarantee the agent does not overstep its mandate? On June 12, 2026, at the Visa Payments Forum in San Francisco, Visa announced a partnership with OpenAI embedding tokenized Visa credentials directly into ChatGPT experiences, paired with real-time authorization, live fraud monitoring, and guardrails the user controls, setting spending caps, approved merchant categories, and requiring human sign-off in certain cases. The announcement followed OpenAI's retreat from an earlier version of the idea called Instant Checkout, launched in September 2025 with Etsy, which by March 2026 had been narrowed into a per-app integration rather than one unified checkout flow. Just two days before Visa's announcement, on June 10, Mastercard unveiled a service it calls Agent Pay for Machines, built specifically for high-frequency, extremely small-value transactions between machines themselves, with more than 30 payments-industry partners including Stripe, Coinbase, Adyen, Checkout.com, and Cloudflare. Mastercard Chief Product Officer Jorn Lambert said machine payments can make it possible for services to be bought and sold among agents at fundamentally different scales, while Alchemy co-founder Joe Lau went further, arguing that machines paying each other constantly, for things too small to bother a human with, unlocks business models nobody can build today. This race between the two payment giants is not a technical footnote. It is the underlying layer that will determine whether an agent economy actually functions or remains a demo.

07

The Gulf Is Ready for Agents Before Agents Arrive

If you had to pick one region of the world to test an agent economy first, it would be hard to ignore the Gulf Cooperation Council states. Internet penetration exceeds 97% across all six countries, according to 23HubLab's 2026 analysis of digital growth in the GCC, among the highest rates anywhere. More telling, over 82% of the region's e-commerce transactions happen specifically on mobile, not desktop, which means any shift in behavior at the phone level lands harder here than in markets where the desktop still holds real ground. The same estimates put total GCC e-commerce transaction value at roughly $57 billion in 2025, with projections putting it past $72 billion by the end of 2026, driven mainly by mobile-centric shopping behavior among consumers under 40. Add to that the fact that Gulf phone users are already used to handling everything, moving a salary, booking a doctor's appointment, paying a utility bill, through one app or maybe two, and it becomes obvious why the region does not need to be convinced en masse to hand a task to an AI agent. It just needs someone to build that agent with enough trust that a user will hand over access to their bank account.

GCC E-commerce Transaction Value

Source: 23HubLab, State of Digital Growth in the GCC 2026

08

Who Owns Attention: Gulf Ad Spend and Platform Maps

Any AI agent that wants to operate in the Gulf needs to understand the region's attention map first, and it looks different from most Western markets. By 23HubLab's estimates, digital ad spend across the GCC is projected to reach roughly $12.4 billion in 2026, growing about 19% year over year, more than double the 10 to 12% global average. Saudi Arabia alone accounts for roughly 48% of that spend, followed by the UAE at about 35%, with Qatar, Kuwait, Bahrain, and Oman splitting the rest. On platforms, Snapchat remains the dominant force specifically in Saudi Arabia, with more than 23 million monthly active users and a daily-to-monthly usage ratio above 74%, a rate that outpaces most competing platforms globally. TikTok pulls in about 28 million regional users with an average daily session exceeding 52 minutes, while LinkedIn stands out as a particular case in the UAE, where its 4.5 million users come out of a population of roughly 10 million, an exceptional rate of professional penetration rarely matched elsewhere. Any company building an agent aimed at Gulf consumers will have to treat Snapchat and TikTok as core distribution channels rather than afterthoughts, which partly explains why AI advertising firms have already started designing their agent interfaces around behavior native to these platforms instead of copying models built for American or European users.

GCC Digital Ad Spend by Market, 2026

Source: 23HubLab, State of Digital Growth in the GCC 2026

09

Saudi Arabia Bets on Silicon: Qualcomm, Aramco, and HUMAIN

Saudi Arabia Bets on Silicon: Qualcomm, Aramco, and HUMAIN

While American and Chinese firms fight over who builds the first agent phone, Saudi Arabia is moving on a quieter but arguably more consequential front: who designs the chips and architecture running this generation of devices in the first place. The Design in Saudi Arabia with AI 2026 program, running from April through November 2026, brings together four parties not normally seen working this closely: Qualcomm through its international arm, Saudi Aramco through its accelerated innovation lab known as aramcoSAIL, the Kingdom's Research, Development and Innovation Authority, and HUMAIN, positioned as the country's national AI champion. The program received more than 124 applications and selected only ten startups, among them Deqa AI, Digital Petroleum, Finix Systems, Nommas.ai, and Tawkeed, which now get technical mentorship and patent guidance alongside direct access to Qualcomm's AIC200 AI accelerators and HUMAIN cloud computing credits. Aramco Executive Vice President Ahmad Al-Khowaiter said the program has not only supported unique and technically advanced startups, pointing to an ambition wider than simply funding companies, one aimed at building a local design base for chips and embedded systems. The implicit message is clear. If the future of phones and devices gets built around agents running locally on-device, whoever can design the silicon that runs those agents efficiently, not just buy it, keeps far more leverage once the industry matures, and that is exactly the bet Riyadh is making by tying Qualcomm into its domestic industrial base this early.

DISAI 2026: From Applications to Selection

Source: Zawya, Qualcomm/Aramco/RDIA/HUMAIN DISAI 2026 announcement

10

Arabic Voice and the Agent: From Voice Search to HUMAIN Chat

Arabic Voice and the Agent: From Voice Search to HUMAIN Chat

Any AI agent operating on a phone will mostly talk with users by voice rather than text, and that is exactly what makes the Arabic-speaking market interesting. 23HubLab data shows voice search now makes up about 31% of mobile queries in the region, with Arabic-language voice queries growing 67% year over year, while Arabic text search in Saudi Arabia is growing at 2.3 times the rate of English search. These are not side statistics. Any AI agent built primarily around English-language understanding will simply miss half the real demand in this market, and that is precisely the gap HUMAIN Chat tried to fill when it launched on August 26, 2025, billed as the first fully Arabic conversational AI app, built on the 34-billion-parameter ALLaM model, independently verified by Canada's Cohere. The app, owned by the Public Investment Fund through HUMAIN, was built by a team of 120 experts, half of them women, and openly targets 400 million Arabic speakers and 2 billion Muslims worldwide, an ambition that clearly extends well past Saudi Arabia's own borders. HUMAIN CEO Tareq Amin said the app's launch is a point of pride for Saudi Arabia and marks a historic milestone, adding that this is not the end state, but the beginning of a journey with limitless potential. Whether or not those larger ambitions land, the underlying logic holds up: whoever builds the more accurate layer of Arabic voice and text understanding will be in a far stronger position once these apps evolve from simple chatbots into agents capable of booking an appointment or completing a purchase on a user's behalf.

11

The Risks and What Actually Deserves Watching Next

None of this means the transition will be smooth or risk-free. Once an AI agent has simultaneous access to a bank account, a call log, and a live location, the privacy question becomes far deeper than a routine data-use policy, especially since regulatory frameworks in most markets, the Gulf included, were not written with an agent that acts without direct human sign-off on every transaction in mind. There is also a genuine concentration-of-power risk. If only three or four companies, spanning chipmakers, model builders, and payment giants, end up shaping what an agent actually is, any user or small merchant will find themselves with no real options outside that closed loop. A regional bank or a mid-sized Gulf retailer, for instance, would have almost no leverage to negotiate terms with whichever agent platform ends up carrying most of the region's mobile traffic. Then there is the simpler, more immediate question: does the average user actually want to give up manual control over every small purchase decision? Recent NIQ data shows only 9% of consumers are buying new phones today specifically for AI-driven improvements, still a limited share despite all the noise, even as the firm forecasts 8% global smartphone-sector growth in 2026, partly driven by this same upgrade cycle. NIQ's global VP for mobile networks, Anastasia Bourneli, summed it up precisely when she said smartphones in 2026 are moving beyond connectivity and specifications toward intelligence and orchestration. That gap between announced ambition and measured consumer appetite is worth sitting with, because it suggests the hardware and payment rails may arrive well before most people are ready to hand an agent their card. The coming months should settle three things worth watching closely: whether Apple actually ships its new App Store agent framework, whether OpenAI's phone specs lock in on schedule by early 2027, and whether Riyadh turns DISAI from a small pilot cohort into a locally built, exportable chip-design ecosystem. None of these three is guaranteed, and each would independently reshape how fast an agent-first phone becomes ordinary rather than experimental, in the Gulf and everywhere else.

12

References

// Want to apply this?

Let's discuss how this applies to your business.

A senior engineer reviews every inquiry and responds within one business day.

Start a Conversation