Top Tech News Today, September 15, 2026: Apple, Gates Foundation, Google, Microsoft, Nvidia, OpenAI & More

It’s Tuesday, September 15, 2026, and the AI argument has split in two. On one stage, the White House and Nvidia tell the industry the takeover scare is a hoax. Offstage, the technology itself keeps moving: a Rome startup raises $270 million to defend robots and machines, Salesforce trains its own CRM model so enterprises can keep work off the frontier labs, Waymo maps driverless taxis onto Tokyo, and a single crafted email is enough to root a Cisco gateway.

AI, meanwhile, is moving out of the chatbot and into the physical world. Nvidia’s next-generation hardware is showing striking gains in AI output per megawatt, Samsung is backing a $230 million challenger to GPU economics, OpenAI is buying camera technology, Apple is putting a far smarter Siri into millions of iPhones, and humanoid robots are getting closer to working safely beside people. But the infrastructure underneath this boom is becoming just as consequential: Microsoft is cleaning up problems from a record Windows security update, China is tightening rules around sensitive technology and talent, and war damage in the Middle East has put parts of Amazon’s cloud infrastructure beyond reach.

Elsewhere, the Gates Foundation is committing $1 billion to push AI into healthcare, education, and agriculture, Europe is pouring fresh capital into satellites and embedded cybersecurity, and even Google is giving its engineers access to a rival AI model.

The common thread is hard to miss: AI is no longer just a software story. It is becoming a chips, energy, devices, robots, security, infrastructure, and geopolitics story—all at once. Here are the top technology news stories shaping the world today.

Gates Foundation Commits $1 Billion to Bring AI to Health, Education, and Agriculture Worldwide

The Gates Foundation is committing at least $1 billion over the next two years to expand access to artificial intelligence, backing one of philanthropy’s largest technology bets: that AI should reach far beyond wealthy companies and developed markets. The initiative, announced alongside the foundation’s 10th annual Goalkeepers Report, will focus heavily on health, education, and agriculture, particularly in communities where access to technology and skilled professionals remains limited.

The commitment is significant because much of the AI investment cycle has centered on frontier models, chips, cloud infrastructure, and enterprise software. The Gates initiative approaches the market from the opposite direction: distribution. The foundation wants AI tools placed in the hands of healthcare workers, teachers, and farmers, with human oversight remaining central to their use. That could help create demand for lower-cost models that can operate in regions where connectivity, compute capacity, and electricity are constrained. It also pressures AI developers to build systems that work across languages, economic conditions, and infrastructure environments rather than primarily serving affluent markets.

Why It Matters: The next phase of AI adoption may depend as much on getting useful technology into underserved communities as on building larger models.

Source: New York Times / Gates Foundation.

Nvidia’s Vera Rubin Shows Up to 7X Higher AI Token Throughput per Megawatt in Early Tests

Early testing of Nvidia’s Vera Rubin NVL72 platform suggests the next generation of AI hardware could deliver substantially greater inference efficiency than Blackwell. SemiAnalysis reported that pre-release Rubin systems achieved as much as seven times better token throughput per megawatt in certain large-model workloads compared with Blackwell configurations, exceeding the roughly threefold improvement Nvidia CEO Jensen Huang had previously discussed for some trillion-parameter workloads.

The results matter because electricity, rather than silicon availability, is increasingly the limiting resource for AI infrastructure. Hyperscalers can order more accelerators, but obtaining gigawatts of additional generation, transmission capacity, cooling, and data-center connections can take years. That changes the economics of chip performance. The most valuable accelerator may increasingly be the one that produces the most useful tokens from each megawatt rather than simply delivering the highest raw benchmark score. SemiAnalysis cautions that performance varies significantly by workload and serving configuration, and these are early results using pre-release software. Even so, the findings suggest Nvidia is attacking AI’s energy constraint through full-system co-design spanning CPUs, GPUs, networking, switches, and software.

Why It Matters: If Rubin’s early efficiency gains hold at scale, Nvidia could increase AI compute capacity without requiring an equivalent increase in electricity consumption.

Source: SemiAnalysis.

Microsoft Rushes Out Emergency Windows Fix After Record Patch Tuesday Causes New Problems

Microsoft has issued an emergency out-of-band Windows update after its enormous September Patch Tuesday release introduced problems affecting Remote Desktop, Hyper-V Linux virtual machines, USB audio devices, and other systems. The original security release addressed roughly 1,000 vulnerabilities, including actively exploited zero-days, making it the largest Patch Tuesday in Microsoft’s history by several measures.

The episode illustrates a growing challenge for software vendors: fixing vulnerabilities at unprecedented scale without introducing operational failures. Microsoft’s September release touched hundreds of Windows components and enterprise systems, meaning even a small percentage of regressions can affect large numbers of organizations. Emergency out-of-band patches were once unusual but have become more common as software stacks grow and security teams face pressure to patch immediately. Enterprises now confront competing risks: delaying security updates leaves known vulnerabilities exposed, while deploying large updates immediately can disrupt critical systems. The situation reinforces the importance of staged deployment, testing, rollback procedures, and endpoint visibility for organizations managing large Windows environments.

Why It Matters: Microsoft’s emergency patch shows how the rising volume of software vulnerabilities is making secure software maintenance itself an increasingly difficult engineering problem.

Source: The Verge.

Amazon AWS Says War Damage Has Left Bahrain Cloud Region and UAE Data Zone Inaccessible

Amazon Web Services says it remains unable to restore access to its cloud infrastructure in Bahrain and one availability zone in the United Arab Emirates following damage from missile and drone strikes during the regional conflict earlier this year. AWS has moved many customer workloads to alternative regions, but some infrastructure and data remain inaccessible after physical damage exceeded the resilience assumptions built into the affected facilities.

The incident provides a rare real-world test of cloud computing under wartime conditions. Hyperscale cloud architecture is built around redundancy, with applications distributed across availability zones and regions so failures do not take entire systems offline. But physical attacks can challenge assumptions designed primarily around equipment failure, weather events, fires, and conventional utility outages. The disruption has reportedly prompted discussion in the Gulf about geographically dispersing data centers and designing facilities with greater physical protection. As countries pour billions into sovereign AI and cloud infrastructure, geopolitical resilience is becoming another factor in where data centers are located and how workloads are distributed.

Why It Matters: AI and cloud infrastructure are now strategic physical assets, and the Middle East outages show that geographic redundancy must account for geopolitical and military risk, not just technical failure.

Source: Reuters.

Trump Phones Nvidia’s Jensen Huang Onstage and Calls AI Takeover Fears a “Hoax”

President Donald Trump called Nvidia chief executive Jensen Huang mid-interview at the All-In Summit in Los Angeles on Monday and was put on speakerphone in front of a live audience. Trump said “the robots will not be taking over” and described extinction-style warnings as a “hoax,” arguing that opposition to data-center construction plays into the hands of political opponents and China. Huang replied that Nvidia would make sure “everybody wins in the AI race in America.” The exchange landed a day after Anthropic’s Dario Amodei published a call to “pace the frontier,” a position later echoed by other lab leaders.
theverge.com

Huang separately struck a more measured tone than the president, backing independent evaluators while rejecting a broad halt. Trump also praised data centers as wealth engines for states and compared AI infrastructure to “the oil of the next 20, 25 years.” Nvidia sits at the center of both the safety debate and the capex boom: chip stocks sold off Monday as investors priced a possible slowdown in frontier training, even as hyperscalers continue to lock in multi-year GPU supply. The public alignment between the White House and the industry’s most important supplier hardens a policy line that treats deceleration as a strategic gift to Beijing.
axios.com

Why It Matters: The most powerful chip vendor and the White House are now openly rejecting a frontier slowdown, which will shape capital spending, export policy, and how startups plan model roadmaps.

Source: The Verge.

Apple Rolls Out iOS 27 With Siri AI, Bringing Its Biggest AI Upgrade to Millions of iPhones

Apple has released iOS 27 and iPadOS 27, beginning the broad rollout of Siri AI and a new wave of Apple Intelligence features across supported devices. The updated Siri can use personal context, understand on-screen information, perform actions across apps, and access broader knowledge sources. The software is available on iPhones dating back to the iPhone 11 and second-generation iPhone SE, although the headline Siri AI and Apple Intelligence capabilities require an iPhone 15 Pro or newer. Siri AI is initially launching in beta, with availability and capabilities varying by device, language, and region.

The release is a critical test of Apple’s ability to close the generative-AI gap with Google, OpenAI, Anthropic, and other rivals while preserving the privacy-focused architecture that differentiates its ecosystem. Siri AI can search across messages, emails, photos, and other personal information, understand on-screen content, carry out systemwide app actions, and provide broader answers to questions. Apple is also integrating AI deeper into everyday applications, including natural-language Shortcuts creation, intelligent photo editing, contextual suggestions in Mail, and smarter Calendar actions.

The broader strategic question is whether Apple can turn its enormous installed device base into an advantage in consumer AI. Unlike standalone chatbots, Siri is embedded directly into the operating system and can potentially combine personal context, apps, sensors, cameras, and on-device processing. If the experience proves reliable, Apple could make AI assistance a routine part of hundreds of millions of daily device interactions rather than a destination users deliberately open. But hardware restrictions mean many iOS 27 users will receive the operating-system improvements without its most advanced AI capabilities.

Why It Matters: iOS 27 marks Apple’s most consequential attempt yet to turn Siri into a modern AI assistant and use its massive hardware ecosystem to compete for control of the consumer AI interface.

Source: The Verge / Apple.

Musk’s X and SpaceXAI Drop Apple From Their AI Antitrust Lawsuit

X Corp and SpaceXAI asked a Texas federal judge on Monday to dismiss their antitrust claims against Apple with prejudice, saying those claims have been “resolved.” Apple did not oppose the motion. The companies are keeping the case alive against OpenAI Foundation, OpenAI L.L.C., and OpenAI OpCo. The 2025 suit alleged that Apple’s exclusive integration of ChatGPT into Apple Intelligence and Siri locked rivals out of smartphone and chatbot distribution. A judge had allowed the case to proceed last November. Terms of the Apple resolution were not disclosed.
9to5mac.com

Dismissal with prejudice means Apple cannot refile the claims in the same form. The OpenAI track remains on track for a January 2027 trial setting, with a mediation deadline in December. The split outcome is notable because Apple is simultaneously shipping a more capable on-device Siri and, according to earlier code analysis, building hooks that could route requests to third-party models. Removing Apple from the docket lowers one legal overhang on the iPhone maker while leaving OpenAI exposed to a competitor-led monopoly fight over distribution and default placement.
reuters.com

Why It Matters: Default placement on the iPhone is still the highest-value distribution prize in consumer AI, and OpenAI now faces that fight without Apple sitting beside it as a co-defendant.

Source: 9to5Mac.

MediaTek Unveils First 2nm Dimensity 9600 Pro Chip as On-Device AI Race Accelerates

Taiwan’s MediaTek has unveiled the Dimensity 9600 Pro, its first flagship smartphone processor manufactured using TSMC’s advanced 2-nanometer process. The company is targeting the premium smartphone segment with a chip built to handle increasingly sophisticated AI workloads directly on devices. MediaTek says its neural processing unit delivers a 51% improvement in prompt-processing performance over the previous generation, while the architecture also improves efficiency for always-on AI workloads.

The launch represents more than another annual smartphone-chip refresh. As generative AI moves from cloud servers onto phones, PCs, vehicles, glasses, and other edge devices, semiconductor companies are racing to deliver enough local computing capacity to run useful models without constantly sending requests to data centers. MediaTek is also expanding beyond smartphones: its first AI accelerator for a major U.S. cloud provider is expected to enter mass production later this year. That puts a company traditionally associated with mobile chips deeper into the global AI infrastructure contest alongside Nvidia, Qualcomm, Broadcom, and hyperscalers developing custom silicon.

Why It Matters: The Dimensity 9600 Pro shows how quickly advanced AI computing is moving from massive data centers into everyday consumer devices.

Source: Reuters.

Google Gives Engineers Access to Anthropic’s Claude Despite Building Rival Gemini AI

Google is giving engineers across the company access to Anthropic’s Claude through its internal Antigravity development environment, according to Business Insider. Claude had previously been available to selected teams and high-priority projects, but the broader rollout means Google developers can now use one of Gemini’s strongest outside competitors for coding tasks. Google says Gemini remains its primary and foundational model for internal development.

The decision says something important about the emerging AI software market: even companies building frontier models may not rely exclusively on their own technology. Coding models have become essential productivity tools inside technology companies, and engineers increasingly want access to whichever model performs best for a specific task. Google also has a substantial financial relationship with Anthropic, making the competitive dynamic unusually intertwined. For enterprise customers, the development reinforces a broader shift toward multi-model environments where organizations route different workloads to different AI systems based on cost, speed, reliability, security, or capability instead of standardizing on one provider.

Why It Matters: Google allowing widespread internal Claude use suggests the enterprise AI market may evolve toward multi-model stacks rather than winner-take-all platforms.

Source: Business Insider.

OpenAI Buys AI Camera Startup Glass Imaging in Deal Valued Above $300 Million

OpenAI has acquired Glass Imaging, a startup founded by former Apple engineers that uses artificial intelligence to improve smartphone photography, in a transaction valuing the company at more than $300 million, according to The Wall Street Journal. Glass Imaging has developed computational photography technology intended to produce higher-quality images from small smartphone camera sensors and had previously raised funding from investors including GV and Insight Partners.

The acquisition adds another piece to OpenAI’s growing consumer-hardware strategy. The company has already moved deeper into physical products through its relationship with former Apple design chief Jony Ive and its acquisition of io Products. Camera technology could become especially important for AI devices because cameras give machines continuous visual context about the physical environment. That could let future assistants understand objects, locations, documents, gestures, and events rather than relying primarily on typed or spoken prompts. Glass Imaging also brings engineers with direct experience in smartphone computational photography, an area where hardware, optics, and AI increasingly converge.

Why It Matters: OpenAI’s purchase of Glass Imaging points toward AI devices that see and interpret the physical environment rather than functioning primarily as voice or text interfaces.

Source: Wall Street Journal.

Samsung Backs Euclyd in €200 Million AI Chip Round to Challenge GPU Economics

Dutch AI chip startup Euclyd has raised more than €200 million, roughly $230 million, in a Series A round co-led by Samsung, Somerset Capital Partners, the EQT-managed Scaleup Europe Fund, and Innovation Industries. The Eindhoven-based company is developing an inference architecture that departs from conventional GPUs, combining custom compute, memory design, and data-center systems. Former ASML CEO Peter Wennink is joining Euclyd as chairman.

Euclyd is entering one of technology’s most difficult markets. Nvidia’s dominance rests on far more than GPUs: its software ecosystem, networking technology, developer base, and supply relationships create formidable barriers for challengers. Euclyd is betting that inference economics will create room for architectures built specifically around running AI models rather than general GPU workloads. Samsung’s participation is particularly notable because memory, manufacturing relationships, packaging, and supply-chain access can be as important to an AI chip startup as capital. Euclyd plans to begin commercial deployment of its physical systems in 2028, so it still needs to prove its claims at scale.

Why It Matters: Investors are continuing to fund alternatives to Nvidia as inference cost, energy consumption, and memory bandwidth become central constraints on AI deployment.

Source: TechStartups via Euclyd, CNBC.

Exein Raises $270 Million at $1.7 Billion Valuation to Secure the Physical AI Era

Rome-based cybersecurity startup Exein has raised $270 million in a funding round led by Headline, valuing the company at $1.7 billion. Exein builds security directly into the firmware of connected devices, including industrial equipment, vehicles, routers, robots, and consumer electronics. The company says its technology is already present across more than two billion devices and is developing an AI security model intended to identify and respond to attacks against physical systems in real time.

The timing is important. AI is moving out of browsers and cloud applications into machines that interact directly with the physical environment. Robots, autonomous vehicles, drones, industrial controllers, medical equipment, and smart infrastructure all create a different security problem from conventional software. Compromising one of those systems can have physical consequences. Exein’s approach treats security as part of the device itself, not something added later through network monitoring or endpoint software. The funding also highlights Europe’s growing position in embedded and industrial cybersecurity, an area where automotive, defense, manufacturing, and critical infrastructure increasingly overlap.

Why It Matters: As AI gains control of physical machines, cybersecurity is shifting from protecting data to protecting devices that can act in the real world.

Source: TechStartups via Exein.

Cornelis Raises $205 Million to Attack the AI Data Center Networking Bottleneck

Cornelis Networks has raised $205 million while unveiling Active Compute Fabric, a new networking architecture intended to make the network itself participate in AI computation rather than simply move data between accelerators. The Intel spinout says the architecture spans both scale-up and scale-out networking and incorporates programmable compute and in-fabric acceleration. Cornelis is also working with Qualcomm around networking for its Dragonfly AI accelerators.

The company is targeting an increasingly expensive weakness inside AI clusters: idle accelerators. Thousands of GPUs can deliver enormous theoretical computing capacity, but performance falls when processors spend time waiting for data or synchronization across the network. That means companies can spend billions on accelerators and still fail to use them efficiently. Nvidia has built networking into its own full-stack strategy through technologies such as NVLink and InfiniBand, while competing ecosystems are rallying around standards including Ultra Ethernet and UALink. Cornelis is betting customers will want high-performance networking without being locked into a single accelerator vendor.

Why It Matters: As AI clusters grow, networking is becoming nearly as important as GPUs in determining how much useful compute data centers actually get from their hardware.

Source: Network World.

Agility Robotics Unveils Digit 5 Humanoid Built to Work Safely Alongside Humans

Agility Robotics has unveiled Digit 5, a new generation of its humanoid robot built to operate more closely alongside human workers in warehouses and factories. The robot uses additional sensors and AI-based collision-avoidance technology to detect people approaching and automatically slow, stop, or shut down when necessary. Digit 5 can lift up to 50 pounds, uses interchangeable hands for different tasks, and can recharge in about nine minutes following roughly 90 minutes of operation.

Workplace safety is one of the biggest barriers to deploying humanoid robots at scale. Traditional industrial robots are often separated from employees by cages or restricted zones because their speed and force can cause serious injuries. Humanoids are supposed to work in environments originally built for people, making safe human-machine interaction essential. Agility says customer feedback from companies including Amazon, Toyota, and GXO influenced the new design. The company reports more than $300 million in multiyear orders, although commercialization remains capital-intensive and competition from Figure AI, Tesla, Unitree, and other robotics companies is intensifying.

Why It Matters: Humanoid robots will have limited economic value if they cannot safely share workplaces with people, making human-aware operation a critical step toward commercial adoption.

Source: Bloomberg / Business Insider.

China Tightens Exit Rules to Protect Advanced Technology and National Security

New Chinese exit-and-entry regulations took effect Tuesday, giving authorities broader tools to prevent citizens from leaving the country when officials believe national technology or industrial security is at risk. The rules include restrictions tied to violations of technology import-export controls and can impose exit bans ranging from six months to three years in certain cases involving conduct abroad that authorities determine harmed national security or interests.

The changes arrive as technology expertise itself becomes a strategic asset. Governments are increasingly treating semiconductor engineers, AI researchers, proprietary data, manufacturing knowledge, and intellectual property as components of national security. China has already tightened controls around advanced technology exports and sensitive information while the United States has restricted China’s access to advanced chips and semiconductor manufacturing equipment. Talent mobility is now becoming another front in that contest. Beijing says the regulations target unlawful behavior rather than ordinary travelers, but multinational technology companies will closely watch how broadly the rules are applied to engineers, researchers, executives, and others with access to sensitive technology.

Why It Matters: The U.S.-China technology conflict is moving beyond chips and export controls to the movement of people with strategically important technical knowledge.

Source: Associated Press.

Open Cosmos Raises €300 Million as Europe Builds a More Independent Space Tech Industry

British satellite startup Open Cosmos has raised €300 million to expand satellite manufacturing and its space-based connectivity and intelligence services. The round includes backing from Lightrock, ETF Partners, Institut Català de Finances, Entrepreneurs First, the UK’s National Security Strategic Investment Fund, and other investors. The company operates facilities across Britain, Spain, Portugal, and Greece, serving governments, space agencies, and commercial customers.

The financing arrives amid a broader European push to strengthen sovereign space capabilities. SpaceX has demonstrated the strategic and economic advantage of vertically integrated launch and satellite infrastructure, while European governments increasingly view independent communications, Earth observation, and intelligence capabilities as national-security priorities. Open Cosmos takes a different approach from consumer broadband constellations such as Starlink, focusing instead on smaller satellite networks and data services for governments and enterprises. The company says it has achieved five consecutive years of profitable growth, a notable claim in a sector where many startups remain dependent on external financing.

Why It Matters: Europe is putting serious private and public capital behind homegrown satellite infrastructure as space becomes critical to communications, defense, climate monitoring, and economic security.

Source: The Next Web.