Happy Thursday, and welcome to EnterpriseAM AI + Innovation! There are just nine days left before Halloween, enough time to stream the original 1968 Night of the Living Dead, the 1978 Invasion of the Body Snatchers (the second-creepiest film Donald Sutherland ever made), and the original 1981 The Evil Dead. Notice a theme? Now, let’s get going.
When can AI agents take over tasks on their own? And what are the benefits and the dangers when they do? That’s one of the questions bedeviling executives now as AI agents are becoming better than human employees at many tasks even as the legal and regulatory rules that would govern them flying solo are still mostly unwritten.
Taking the leap: Nils Bachtler (LinkedIn), co-founder and CEO of the Egyptian fintech Klivvr, is at one end of the spectrum. Klivvr is expecting “in the very near future” to allow AI agents to make decisions without human approval about underwriting loans and following up on late payments, Bachtler said during the Enterprise Egypt Forum earlier this month.
But what if…? Asked about the dangers of putting machines in charge, Bachtler made two separate cases.
The value to the company of making AI independent in some areas is greater than the risk of a screwup.
Putting these decisions in the hands of humans has its own drawbacks — arguably more of them. “When you hire someone, for example a risk officer or a credit officer, he comes with a past experience from a different time, from a different institution. The world in general is evolving … faster than people can cope with,” he said.
The wait-and-see approach: Khalil El Bawab (LinkedIn), CEO for local and regional markets at the investment bank Beltone Holding, isn’t ready to take humans out of the loop. He described a more cautious approach. “We trust [AI] in the collection of data at this point in time. We trust it in making suggestions, but do we trust it in deciding? Not at this moment in time, not in our line of business,” he said, speaking on a separate panel at the forum.
One big reason to be hesitant: AI agents may do the work far faster and get it right more often than humans at this point. But, when they go off the rails, there’s a good chance that they’ll do so more spectacularly than the average well-intentioned human ever would — as evidenced by the steady stream of stories about AI agents going rogue and hacking innocent victims.
The human holding the bag: “In case, God forbid, litigation [starts], who’s going to be sued? It’s the managing director or the CEO of the company, not the machine,” El Bawab said.
But, remember: We’re (probably) still in the very early days of the AI revolution. Agents’ capabilities will get far better. Their decisionmaking will grow more nuanced. Governments may even develop guardrails that give companies more confidence handing some decisionmaking over to AI. Ultimately, the balancing act we’re talking about above may come to sound quaint.
There’s a lot more from our Egypt Forum in Forum Playback.It’s online here and will be arriving in your inbox every week for the next seven weeks.
Now, let’s get to today’s Lede, which looks at the physical threats to Gulf data centers and the Western partners that haven’t been scared away — so far.
The enthusiasm for building AI infrastructure in the Gulf hasn’t noticeably dwindled more than seven months after Iranian drones damaged Amazon data centers there and amid ongoing attacks by Houthis in Saudi Arabia. That’s the good news for Saudi Arabia, the UAE and other Gulf nations as they aim to aggressively reorient their economies to take advantage of the global surge in demand for AI computing power.
But there’s a lurking danger that big Western hyperscalers might reconsider the Gulf’s long-term viability as a major computing hub — especially if Iran follows through on threats to aggressively target the infrastructure of Western firms that it considers complicit with the US and Israel.
Iranian officials haverenewed those threatsin recent weeks. Iranian Parliament Speaker Mohammad Bagher Ghalibaf warned on X this week that the nation is creating a “Persian Gulf Data Authority” to punish nations that cooperate with AI firms it considers complicit in military actions against Iran. “To governments, funds, and firms in the region: you are kindly advised to give no resources to any AI or data company aiding invasions or killing children. No finance, equity, land and energy. Assist them and face our enforcement,” Ghalibaf wrote.
The Iranian Revolutionary Guard Corps published alistof 18 companies that were potential targets back in late March, including US tech giants and the UAE’s G42.
Even a minor shift in the enthusiasm of Western AI firms could have serious implications for the technology’s long-term economic prospects in the Gulf. “At the beginning of the year, I would have described the Middle East as one of the most robust data center markets, probably a couple of steps ahead of Europe in terms of growth momentum,” Tahir Gok (LinkedIn), an independent digital infrastructure consultant focused on the Middle East, tells EnterpriseAM. “Right now, considering the different geographical risks, we need to wait and see.”
Egypt, meanwhile, sees an opportunity to take market share from the Gulf, with Hassan Allam Digital Infrastructure and A15 at the leading edge after their summer announcement they had committed USD 400 mn in initial capital to go after the data center market. We went deep on Egypt’s ambitions earlier this week.
The top line: Iranian drones directly struck two Amazon Web Services data centers in the UAE and damaged another in Bahrain with a nearby strike in the early days of the war, causing structural damage, power outages and fires, according to the cloud provider’s service dashboard, dramatically jeopardizing the Gulf’s status as a safe haven for Western tech investment. A second AWS site in Bahrain was damaged in April.
Amazon has been unable to restore access to cloud services in Bahrain and parts of the UAE, the company acknowledged in a September update, adding that it is “working on replacing the affected [UAE] infrastructure and will provide an update on the restoration of our services in the coming months.” In Bahrain, AWS said it had exhausted every option for recovering data that customers hadn’t migrated before the region went offline, and that it would provide a further update in early 2027.
Staying the course: Construction on more than USD 100 bn in major AI building projects across the Gulf has generally continued since then with no broad work stoppages, Gok and other analysts say — despite intermittent bursts of fighting and despite a sharp increase in insurance costs.
There have been some delays caused by supply chain issues, Gok notes, many of them due to the large volumes of data center construction across the globe. Others are due to the closure of the Strait of Hormuz, The Register reports, citing BCS Consultancy.
Spreading out: The UAE is considering revising plans for the massive 5 GW UAE-US AI Campus that G42 is leading to make it more geographically spread out, Reuters reports.
The plan, which is not finalized, envisions building a network of data centers across the country rather than a single 26 sq km campus in Abu Dhabi, the wire service says. That broader footprint would make it more difficult for Iran or another adversary to significantly disrupt operations using drone or missile strikes. The plan is not expected to affect Stargate UAE, the campus’s USD 30 bn, 1 GW first phase that’s already under construction.
IN CONTEXT- The intermittent bursts of missile and drone strikes between the US, Israel and Iran during the past five months haven’t matched the intensity of fighting in March and April, though the lack of a peace deal or long-term ceasefire means there’s still a risk of returning to all-out conflict. The Iran-aligned Houthis, meanwhile, have dramatically increased attacks against Saudi Arabia, causing numerous casualties and, importantly for the tech world, forcing the postponement of the Kingdom’s annual Future Investment Initiative conference, known as Davos in the Desert.
Gulf AI champions want you to know it’s “BAU” — business as usual:
The UAE’s under-construction Stargate AI campus will deliver its first 200 MW of computing power in Q4 2026, Khazna, G42’s data center division, told Bloomberg.
Microsoft’s Saudi Arabia East data center region will be open for customers in November and Amazon plans to make up to 50 MW of capacity available in Saudi Arabia’s first AI zone by 2028, Bloomberg also reported.
Saudi Arabia’s Leap conference, which closed on 3 September, saw more than USD 15 bn in launches and agreements on its opening day alone, including new partnerships and expansions involving Amazon, Microsoft and other Western tech giants.
The UAE’s ambassador to Washington, Yousef Al Otaiba, pledged that work was getting done despite the challenges posed by the conflict in a Monday letter to the US-UAE Business Council. “Our commitments to the United States remain firm, our major investments are moving forward and expanded access to advanced technology is opening further opportunities,” he wrote.
Some Western partners are also projecting stability: Microsoft led the pack by announcing plans in September to invest more than USD 10 bn in Gulf technology between now and 2030, a figure that includes USD 2 bn in new commitments. The outlay is focused on cloud and AI infrastructure, among other priorities. Amazon Web Services CEO Matt Garman told Forbes in April that his company “continue[s] to be extremely bullish about our partnerships in the Middle East and about the long term potential of the region.” The Trump administration also gave a boost to the region in July when it announced new export rules granting G42, its cloud arm Core42, and US companies operating in the country access to advanced AI chips without export licenses.
Also important to note: The relationship between US tech giants and Gulf AI champions is unusually tight — including numerous USD multi-bn deals. That makes the relationship extremely difficult — though not impossible — to unwind during tough times. OpenAI and Oracle, for example, are set to operate Stargate UAE, which G42 is building with partners including Nvidia, Cisco and SoftBank in a roughly USD 30 bn transaction.
But companies have also generally been tight-lipped about their investment plans in Gulf infrastructure beyond existing commitments. US firms including Amazon, Google, Microsoft, Cisco, and Nvidia either declined to comment for this story or didn’t respond to a request for comment.
“I think a lot of companies are waiting to see what happens,” Sam Winter-Levy (LinkedIn), a senior fellow focused on AI and geopolitics at the Carnegie Endowment for International Peace, a US think tank, tells EnterpriseAM. “Data centers are soft targets. They’re big, sprawling industrial structures. They’re on Google Maps. It’s hard to protect them against drones.”
A balancing act: Whether foreign companies make additional investments in Gulf AI infrastructure will depend on balancing the risk of renewed fighting with the Gulf’s clear advantages as a computing provider, Gok says.
“I’ve heard from a number of international data center developers that their boards have become more cautious about committing to additional investment in the Middle East,” he notes. But investing in our part of the world is “still attractive, because these countries are offering modern facilities, competitive power economics, and access to advanced GPUs.”
Part of a bigger picture: In addition to worrying about physical attacks from Iran, foreign firms are waiting to see if Gulf AI data centers can attract a broad enough customer base to justify additional building, Gok says. He warned against viewing the Iran war as the only — or even the main — factor in the success or failure of the region’s AI ambitions.
There are a variety of security measures Gulf players can take to mitigate the risks of physical attacks, security analysts tell EnterpriseAM. At the cheaper end, these include installing jamming technology that disrupts drones’ flight paths and nets that drones would get tangled up in. The costlier end includes Iron Dome-like systems to shoot attacking drones out of the sky or even building data centers underground.
Humain and G42 have both announced plans to harden their infrastructure, though they have not provided specific details of new security protections.
THE CHALLENGE: The more money Gulf players spend protecting their data centers from physical attack, the less profitable those centers become. “There are many things you can do to harden a data center, but it’s expensive,” Winter-Levy says. “The economics of building [in the Gulf] look much more challenging, or at least more uncertain.” (Winter-Levy has been a critic of US companies building AI data centers in the Middle East since before the war, arguing that doing so makes the US too dependent on other nations for a critical component of its economic vitality and national security.)
Another strategy is to keep spare parts on hand, so data centers can quickly recover from minor damage. “I’d start with the principle that resilience matters more than trying to make any single facility invulnerable,” Joe Scheidler (LinkedIn), a former White House and US State Department official focused on infrastructure and investment, tells EnterpriseAM. “Companies should look at a layered physical security approach including site design, hardening, backup power and cooling, multiple connectivity routes and making sure they have rapid recovery capability,” says Scheidler, who is founder of the tech firm Helios.
One advantage cloud systems and networked data centers have over other types of facilities is that they can zip data from one facility to another, so damage to one portion of their infrastructure doesn’t necessarily mean long interruptions for customers. Mohammed Soliman (LinkedIn), a senior fellow at the Middle East Institute, argued in a Foreign Policy essay that the Gulf should take lessons from Ukraine, which transferred its sensitive data out of the country after Russia’s 2022 invasion, removing it as a Russian target. However, that would require dramatic policy changes for Gulf leaders who — like a growing number of politicians globally — have insisted on housing sensitive government and other national data inside their countries for sovereignty reasons, Soliman notes.
There’s always a danger that costly anti-drone systems won’t protect against some future attack method devised by Iran or another adversary. “It’s not entirely clear how to provide comprehensive protection at the scale that we’re talking about,” Vincent Carchidi (LinkedIn), a defense industry analyst at the US firm Forecast International, tells EnterpriseAM. “It’s not as simple as building a wall around your data center and being done. The types of aerial or other threats that might be posed to these data centers are going to change over time.”
Insurance costs are one clear window into the challenges facing Gulf AI infrastructure since the war kicked off.
Back in the day: Before the war, the regional limit per policy for political violence insurance — which covers missile and drone strikes — was USD 1 bn and customers were paying between 0.1% and 0.3% of their policy limit in annual premiums, Omar Gemei (LinkedIn), a top executive in Marsh’s Dubai office in charge of numerous types of coverage, including political violence, tells EnterpriseAM.
Lower policy limits, higher premiums: Now, market capacity for political violence has dropped to between USD 100 mn and 300 mn and customers are paying from 2% up to 10% of their policy limit in annual premiums, Gemei says.
Details: Premiums were at the top of that scale early in the war but have dropped closer to the bottom six months later, Gemei says. Rates for critical infrastructure, including big data center projects, are on the higher end of the scale, he says.
Despite that sharp price increase, the market is booming: “This product used to be seen as an easy, ‘nice to have,’” Gemei says. “I think it’s now become a product that’s essential to the business.”
Premium costs will likely drop if and when there’s a lasting ceasefire, Gemei says, but it will take a while: “Say there’s peace today — my own estimation is it would probably take another two years to get back to something close to where we were before the conflict… And, arguably, it may never fully get back to where it was.”
Insurance companies may also be hesitant to lower premiums because they took such a beating early in the conflict. Insurance companies received about USD 2 bn in political violence claims after early rounds of missile and drone strikes compared with about USD 1.5 bn globally in annual premium payments for that type of coverage, Gemei says.
MGX wants a bigger bite of OpenAI, Reuters reports.
The Abu Dhabi state AI investor is among the existing OpenAI backers in talks to join a new fundraising round of about USD 30 bn. That round could value the ChatGPT maker at more than USD 1.4 tn, Reuters says, citing two people it says are familiar with the matter. That’s c. 64% above the USD 852 bn OpenAI was valued at in March. SoftBank is also in talks to take part.
OpenAI wants to close the round before rival Anthropic’s expected public listing in mid-November. Some pundits have suggested that slowing month-on-month growth could put downward pressure on Anthropic’s valuation, giving OpenAI good reason to want to raise now at what could be a high-water mark.
Why it matters for Abu Dhabi: If it closes, this would be the second OpenAI round backed by MGX this year. It co-led a USD 122 bn round that closed in March.
Nearly 500 scientists from 34 countries signed anopen letterlast week calling for the creation of an international body to oversee testing of advanced AI models and for more transparency about when AI models go rogue. The letter cited the torrent of recent testing failures in which AI agents disobeyed instructions and hacked external victims.
We caught up with Christina Pöpper (LinkedIn), a computer science professor and associate dean for faculty affairs at New York University’s Abu Dhabi campus, who was one of the signatories. The interview has been edited for length and clarity.
EnterpriseAM: Why did you sign the open letter?
Christina Pöpper (CP): We are not where we should be in terms of risk containment. Hacking has always happened, but, before, it took humans to find a vulnerability that they could exploit. Now, we don’t need humans to do that anymore. The AI agents are getting so good that they can do it at scale. They can find subtle vulnerabilities that would take humans way, way longer to find. These are problems that cannot be solved individually by companies or universities or government authorities. Standard frameworks are needed. A regulatory framework is needed.
E: Are you more concerned about bad actors using AI for hacking and in other dangerous ways or about the AI itself going rogue?
CP: My current major worry is the first one. That’s very hard to prevent. As for AI itself becoming the malicious actor, that’s what’s behind the open letter. We need regulatory frameworks because you cannot solve this by technology alone.
E: Are there things that excite you about AI?
CP: Oh, absolutely. The capabilities now are mind-blowing. And it keeps developing faster and faster. You could not imagine the things that are possible now, even a year ago.
E: Do you see particular benefits and dangers from AI in the GCC?
CP: The GCC, and in particular the UAE, is fully embracing AI. We have this commitment by the government to involve AI in every step of what people are doing. I definitely see risks involved with that, but also a lot of potential. What’s important is, if it becomes a human safety issue, can we still pull a plug somewhere? Increasingly, this is not realistic because we’re becoming so dependent on AI that we cannot pull the plug anymore. If you pull the plug, then other things collapse, basically. That’s a real risk.
E: What advice do you give to undergraduates worried about AI making them less employable?
CP: The more we have AI, the more, I think, it’s important that we have humans who understand the basics. Critical thinking, really understanding things, questioning things, is something we need more than ever now. This is what, for me, university education is about. Because nobody is able to predict now what the workforce looks like in five years.
E: How do you incorporate AI into the classroom?
CP: There are courses that are minimally impacted by AI and there are courses that are entirely changed by AI. I try to be clear with every task. In many cases, students are instructed to use AI in a certain way, but then they need to critically reflect on that and report on it. For some tasks, they should only be using AI for a specific part, because otherwise, they would lose the learning that comes with doing the work. There are other tasks where I ask them not to use AI at all.
E: Do you recommend certain models for these assignments?
CP: For me, it’s model-neutral. It doesn’t make sense to make it model-dependent, because the models themselves are evolving so quickly. I think the differences between different versions of ChatGPT, Gemini and Claude are often bigger than the differences between the models themselves.
DeepSeek gets the boot: The US is banning Chinese AI model DeepSeek, and any tools built on it, from civilian federal agencies’ computers, CNBC reports. The order came in July, but was made public this week.
The move highlights how AI is increasingly becoming a battleground between the US and China — a conflict that could create problems for players in the Gulf and elsewhere who are trying to stay on good terms with the two big tech rivals. The US previously banned Huawei tools from the computers of government agencies and their contractors.
China could use DeepSeek tech for spying or to introduce malware into US government computers, the Trump administration warned.
China’s Foreign Ministry spokesperson called that far-fetched. “China consistently opposes the over-generalization of national security as well as the politicization of science, technology and trade issues,” Ministry Spokesperson Mao Ning said.
MEANWHILE- Macron’s Gulf AI tour: French President Emmanuel Macron toured the Gulf this week, and AI was high on the agenda. Kuwait and France signed defense and AI agreements, and Macron brought along the head of Campus AI — a tech hub near Paris backed by USD bns in UAE funding — to Abu Dhabi.
AND- Emirati startup goes to Silicon Valley: CNTXT AI, a UAE startup that builds Arabic-first AI and data tools for training robots, is opening its first US office, the Wall Street Journal reports. It says it has USD 300 mn in “committed revenues,” mostly from US tech companies.
The business model: Make humans obsolete. CNTXT records human workers in the Middle East doing everyday tasks so the robots can learn to imitate them and do those jobs.
The plot thickens? A Saudi military official paid for the wife and baby of Abdulrahman Alfozan, an OpenAI employee and Saudi national, to fly out of the US a day after he was arrested for allegedly bugging a company office, prosecutors allege in a court filing. The Saudi consulate also offered to pay his bail, prosecutors claim.
Staying put: A judge refused Alfozan bail on Monday, saying he might flee, the BBC reports. Prosecutors say more serious charges could follow. Here’s the background on Alfozan from our coverage last week.
MEANWHILE- The AI turned off its own alarm: Two rogue AI systems at OpenAI weakened another AI system that was supposed to flag their misbehavior this month, Reuters reports.
How? An agent overinterpreted an order not to send Slack messages to include turning off safety alerts as well. Another deleted a safety check while copying code, then said it hadn’t.
Why it matters: Researchers have warned that AI agents could get harder to watch and control as they become more independent — a problem that will only get worse as agent “swarms” are more commonly employed. OpenAI says it doesn’t “yet have a satisfying explanation” for what happened.
SpaceX + Tesla?: SpaceX and Tesla have each set up special committees to study a possible merger, Elon Musk said on Tuesday, the BBC reports. If concluded, the deal would create a company worth more than USD 3 tn, the biggest corporate tie-up ever, though Musk says winning over Tesla shareholders will be a “high hurdle.”
The Gulf angle: Saudi Arabia’s PIF owns just under 1% of SpaceX, as we reported in June. Saudi’s Humain, Abu Dhabi’s MGX, and Qatar’s sovereign wealth fund also hold SpaceX shares, which they got by backing Musk’s xAI before it merged into SpaceX in February, Semafor reports.
MEANWHILE- Anthropic’s big loss: Anthropic lost USD 80.7 bn in the first half of the year, according to a draft IPO prospectus seen by Reuters. Most of that — USD 72 bn — came from accounting charges rather than day-to-day business, the Wall Street Journal reports.
Spending big on computing: The company spent close to USD 9 bn on computing power in 1H, up from USD 2.3 bn a year earlier, the Journal says.
Can AI help keep students in college? A US startup called Straia is pulling together data scattered across university departments to help advisers spot students who are struggling to pay tuition or attend classes regularly, Bloomberg reports. Straia just raised USD 3.1 mn in a seed round led by Speedrun, Andreessen Horowitz’s startup accelerator.
Qatar gets an AI HQ: Qatar is setting up a National Center for Artificial Intelligence to steer the country’s AI strategy and write its AI rules, after Emir Sheikh Tamim bin Hamad Al Thani signed off on the plan, The Peninsula reports. The center will be housed at the Communications Ministry and report directly to minister Mohammed bin Ali Al Mannai.
Worried that angry locals will block your data center project? There’s now an insurance policy for that. New York-based insurer Arbol has launched coverage that pays out if a government halts or suspends construction on a data center project, Bloomberg reports. Local opposition blocked or delayed at least 45 data center projects worth USD 68 bn in the second quarter alone, according to tracker Data Center Watch.
26-27 October (Monday-Tuesday): The Dubai AI Festival returns as part of Dubai AI Week.
28 October (Wednesday): Alphabet reports 3Q earnings after the US market closes, with AI capex and Gemini / cloud growth as the headline watch items.
28 October (Wednesday): Microsoft reports fiscal 1Q 2027 earnings after the US market closes, including Azure and OpenAI-partnership capex disclosures.
29 October (Thursday): Amazon reports 3Q earnings, including AWS and Bedrock AI growth.
30 October (Friday): UAE businesses with annual revenue of AED 50 mn or more must have an accredited e-invoicing provider in place by this date.
November 2026
7-10 November (Saturday-Tuesday): Cairo ICT runs in New Cairo alongside AIDC Expo, its dedicated AI, data center, and cloud event.
17-19 November (Tuesday-Thursday): Bengaluru Tech Summit brings businesses, startups, researchers, and policymakers to Bengaluru, India, under the theme “AI & Beyond.”
17-20 November (Tuesday-Friday): Microsoft Ignite runs in San Francisco and online, covering Microsoft’s AI, cloud, security, and data products.
30 November-4 December (Monday-Friday): AWS re:Invent brings Amazon’s cloud and AI conference to Las Vegas, with product announcements and sessions on deployment, costs, and security.
December 2026
1-3 December (Tuesday-Thursday): Black Hat MEA brings cybersecurity leaders, researchers, and technology companies to Malham, Riyadh.
6-12 December (Sunday-Saturday): NeurIPS holds its main AI research conference in Sydney, offering a look at advances that could shape the next generation of commercial AI.
7-11 December (Monday-Friday): GITEX Global opens with a leadership summit at Dubai World Trade Centre on 7 December, followed by its exhibition at Expo City Dubai on 8-11 December.