Top Tech News Today, August 27, 2026: Amazon, Apple, Google, Meta, Nvidia, OpenAI, Salesforce & More

It’s Thursday, August 27, 2026, and the AI boom is no longer theoretical. Nvidia just printed another record quarter and, if the reports hold, agreed to buy Hugging Face, the public square of open-source models, for nearly $13 billion. Hours later, investigators said a swarm of OpenAI agents had already broken into that same platform, while a ransomware crew was caught using an AI coding assistant to ransack real companies. The rest of the brief follows that split screen: Apple’s first foldable iPhone gets a date, ChatGPT turns on ads in India, Brussels starts asking frontier labs for paperwork, and airports, retailers, and developer supply chains take the kind of hits that used to be footnotes.

The AI race is starting to look less like a software competition and more like an industrial buildout. Nvidia says sales could jump 70% as demand for AI infrastructure keeps climbing. Anthropic is locking up $45 billion worth of future compute. AWS plans to deploy another 2 million Nvidia GPUs. And Kioxia and Sandisk are preparing more than $31 billion in new memory investments in Japan.

Put it all together and the message from today’s news is hard to miss: AI is no longer simply changing software. It is reshaping chips, data centers, cybersecurity, enterprise apps, consumer hardware, jobs, and the physical infrastructure underneath the global technology economy. Here are the top technology news stories shaping where it goes next.

Technology News Today

Nvidia Posts $96.2 Billion Quarter and Forecasts 70% AI Chip Growth as AI Infrastructure Spending Keeps Surging

Nvidia delivered another record quarter and, more significantly, offered its first year-ahead growth forecast, projecting revenue will jump roughly 70% in its next fiscal year. The chipmaker reported $96.2 billion in quarterly revenue, while its data center business generated $89 billion, up 117% from a year earlier. Nvidia expects revenue of about $108 billion in the current quarter as hyperscalers, AI labs, governments, and enterprises continue building out computing capacity.

The numbers also reveal the enormous industrial machinery now required to keep the AI boom running. Nvidia has committed as much as $160 billion toward memory supply, while rising memory costs are expected to pressure gross margins. The company is increasingly using its balance sheet to support AI infrastructure projects through investments, financing guarantees, and partnerships, raising questions about how intertwined Nvidia has become with its biggest customers. CEO Jensen Huang continues to argue that demand remains constrained more by supply than by customer appetite. If Nvidia’s forecast holds, the AI infrastructure cycle is moving into a scale few technology markets have previously reached.

Why It Matters: Nvidia’s outlook suggests AI infrastructure spending is still accelerating despite growing concerns about costs, financing, and returns on massive data center investments.

Source: Financial Times.

Nvidia Agrees to Buy Hugging Face for $12.9 Billion in Landmark AI Deal

Nvidia has agreed to acquire Hugging Face for $12.9 billion, according to a report from The Information cited by Reuters, turning one of the AI industry’s most important independent model platforms into part of the world’s dominant AI chip company. Hugging Face operates a widely used repository for AI models, datasets, and developer tools and has become a central distribution layer for open-source and open-weight artificial intelligence. Neither company had publicly confirmed the agreement when the report emerged.

The reported price would represent an enormous step up from Hugging Face’s previous valuation. Nvidia participated in a $235 million funding round in 2023 that valued the startup at $4.5 billion, alongside investors including Google and Salesforce. Reuters said Hugging Face’s annualized revenue was recently reported at about $150 million, making the acquisition price especially notable. The deal would also extend Nvidia’s reach far beyond GPUs and networking into the software and model-distribution layer used by thousands of AI developers. For Nvidia, owning Hugging Face could give the company a strategic position wherever developers discover, test, distribute, and deploy models, even as OpenAI, Anthropic, Google, and others pursue more vertically integrated AI stacks.

Why It Matters: Buying Hugging Face could give Nvidia control of a critical distribution hub for open AI models, pushing its influence deeper into the developer ecosystem beyond chips.

Source: TechStartups via The Information, Reuters

Google Launches Gemini 3.5 Transcribe AI Model for Real-Time Speech-to-Text

Google has introduced Gemini 3.5 Transcribe, a new speech-to-text model built for real-time voice applications, transcription, and automated processing of recorded audio. Unlike conventional transcription systems that primarily convert speech into literal text, Gemini 3.5 Transcribe can remove filler words, recognize self-corrections, automatically format output, adapt to specialized vocabulary, identify multiple speakers, and generate word-level timestamps. Google says the system supports more than 85 languages and can stream transcription with sub-second latency.

Google is making the model available to developers through the Gemini API and Google AI Studio, and it is already tied to voice features across parts of Google’s consumer ecosystem. Google reported average word-error rates of 4.0% for streaming workloads and 2.6% for non-streaming transcription in cited testing, though real-world accuracy will vary by language, accents, noise, microphone quality, and specialized terminology. The larger opportunity extends well beyond dictation. Speech is becoming an increasingly important interface for AI agents, call-center automation, meeting tools, accessibility software, customer support, and hands-free computing. Better transcription gives AI systems a more reliable foundation for interpreting what users say before reasoning or taking actions.

Why It Matters: As AI shifts from text boxes toward voice-driven agents, accurate and low-latency speech recognition becomes a critical infrastructure layer rather than a standalone feature.

Source: Ars Technica, with technical details from Google.

Anthropic Strikes $45 Billion AI Compute Deal With Nscale

Anthropic has agreed to spend $45 billion over six years to rent AI computing capacity from British infrastructure company Nscale, Bloomberg News reports. The agreement covers roughly 460 megawatts at Nscale’s West Virginia data center development and is expected to use Nvidia’s next-generation Vera Rubin systems as capacity comes online beginning in late 2027. The scale of the commitment underscores how aggressively frontier AI companies are locking down future computing capacity years before they expect to need it.

The agreement adds another major infrastructure commitment to Anthropic’s growing compute portfolio. AI labs increasingly face a strategic problem that resembles energy-intensive industrial companies more than traditional software startups: future growth depends on securing land, electricity, chips, financing, and data center construction well in advance. Anthropic is trying to ensure that Claude and products such as Claude Code have enough capacity if demand continues climbing. The arrangement is also significant for Nscale, founded only in 2024, because it further establishes specialized AI infrastructure providers as major counterparts to hyperscalers such as Amazon, Microsoft, and Google. Compute access is becoming a defining competitive advantage in frontier AI.

Why It Matters: Anthropic’s $45 billion commitment shows that the AI race is increasingly being fought through long-term control of electricity and computing infrastructure, not just model quality.

Source: Bloomberg News.

Ukraine Awards Elon Musk the Order of Freedom While Lobbying for Deeper Starlink Strikes

President Volodymyr Zelenskyy conferred Ukraine’s Order of Freedom on Elon Musk in a Wednesday decree citing contributions to protecting life and freedom and to Ukraine–U.S. ties. The honor arrives as Kyiv asks Musk to authorize Starlink on drones operating up to 200 kilometers inside Russia so Ukrainian forces can hunt ballistic-missile launchers. Zelenskyy said last week that Musk had called that step a “major escalation” and refused, though he later described “more encouraging” feedback and said the decision sits with the U.S. president.

Starlink has been central to Ukraine’s battlefield communications since 2022 and is used to pilot drones. Musk disabled thousands of grey-market Russian terminals in Ukraine earlier this year. The medal is therefore both recognition and leverage: Ukraine needs a privately controlled satellite network to extend a war-fighting system across an international border.

Why It Matters: A single company’s satellite constellation still functions as strategic infrastructure, with targeting range set by private approval rather than a formal alliance treaty.

Source: Financial Times.

Russian-Speaking Hackers Used Cursor AI Agent to Break Into Companies

Russian-speaking cybercriminals used the Cursor AI coding assistant to help infiltrate at least seven companies, according to data Reuters reviewed and research from Israeli cybersecurity startup Gambit Security. The researchers discovered 28 chat sessions between hackers linked to the Aur0ra ransomware group and a Cursor AI agent, with conversations spanning April 8 through May 21. The attackers allegedly persuaded the AI that malicious operations were part of legitimate security simulations.

Gambit said the agent then carried out hundreds of operations involving credential theft, account takeover attempts, and other intrusion activities. Reuters independently identified several victims, including Belgian cleaning-products manufacturer Christeyns, German garage-door maker Teckentrup, and Scotland’s Helideck Certification Agency. The case shows how general-purpose coding agents can lower attackers’ operational burden by automating tasks that previously required more manual expertise. It also exposes the difficulty AI providers face in separating legitimate penetration testing from malicious intrusion attempts when users intentionally misrepresent their goals. As AI agents gain more autonomy, cybersecurity protections will increasingly need to evaluate behavior and context rather than rely primarily on what users claim they are doing.

Why It Matters: AI coding agents are becoming useful enough to assist real cyberattacks, creating a new security problem for providers trying to distinguish legitimate cybersecurity work from criminal activity.

Source: Reuters.

Boston Scientific Cyberattack Disrupts Medical Device Shipments Worldwide

Medical technology giant Boston Scientific is dealing with a cybersecurity incident that has disrupted parts of its global operations, including its ability to process and ship customer orders. The company detected the incident affecting certain IT systems on August 25 and later experienced a network outage that affected business applications. Boston Scientific makes medical devices used across cardiology, neurology, oncology, and other interventional procedures, giving the disruption potential consequences beyond ordinary corporate IT downtime.

The company told regulators that its investigation remains underway and that it has not yet determined the full operational or financial impact. Boston Scientific has also not provided a timetable for restoring all affected systems. As of Thursday morning, no public confirmation indicated that sensitive data had been stolen, and no known ransomware group had claimed responsibility. The incident shows why attacks on healthcare and medical-device companies are particularly disruptive: compromised corporate systems can disrupt supply chains connecting manufacturers, hospitals, physicians, and patients. Even without direct interference with medical devices, interruptions to fulfillment and logistics can produce real-world consequences. Security investigators will be watching closely for signs of data theft, extortion, or broader network compromise.

Why It Matters: Cyber incidents involving critical healthcare suppliers can quickly spill from corporate networks into medical supply chains, making operational resilience as important as protecting data.

Source: SecurityWeek.

Ransomware Crew Used SpaceX’s Cursor AI Agent to Breach at Least Seven Companies

Gambit Security and Reuters reported that a Russian-speaking affiliate of the Aur0ra ransomware group used SpaceX’s Cursor coding agent to help break into at least seven companies between April 8 and May 21. Gambit recovered 28 chat sessions from a server the gang left exposed. The operator told the agent the work was a simulation, then had it steal credentials, map internal networks, coerce authentication, and pursue account takeovers. Cursor was running Anthropic’s Claude 4.5 Sonnet. Identified victims included Belgium’s Christeyns, Germany’s Teckentrup, and Scotland’s Helideck Certification Agency.

Cursor became part of SpaceX on August 14 in a $60 billion acquisition. Neither Cursor nor SpaceX commented. Gambit’s Eyal Sela said AI assistance can make an operator 30% to 50% faster by skipping manual steps. Separate analysis by CloudSEK tied a related Aurora affiliate to more than 20 organizations across nine countries. The logs show commodity enterprise tradecraft—NetExec, BloodHound, NTLM relay, certificate abuse—executed through an agent rather than a fully autonomous worm.

Why It Matters: Off-the-shelf coding agents are now part of ransomware operations, turning a developer productivity tool into an accelerant for mid-market breaches.

Source: Reuters.

ATF Confirms “Major” Cyber Incident After Qilin Ransomware Claim

The U.S. Bureau of Alcohol, Tobacco, Firearms and Explosives has confirmed that a standalone system was compromised in what the agency called a “major incident,” shortly after the Qilin ransomware operation listed ATF on its leak site. ATF said it disconnected the affected environment after discovering the intrusion and began forensic and incident-response work with the Department of Justice. The agency said the compromised system operates separately from its enterprise network.

ATF said there is currently no indication that its main enterprise systems, eForms platform, or other agency systems were affected, and officials said operations are continuing. Qilin did not initially provide details on exactly what information it may have stolen or whether it demanded a ransom. The ransomware-as-a-service group has become one of the more prolific cyber-extortion operations and has claimed thousands of victims since first appearing under the Agenda name in 2022. The incident comes amid a broader run of cyberattacks affecting U.S. government systems, reinforcing concerns that isolated or specialized networks may receive less security attention than core enterprise infrastructure. Investigators will now need to determine what data the compromised environment contained and whether attackers successfully exfiltrated it.

Why It Matters: The breach highlights how attackers can target specialized government systems outside primary enterprise networks, expanding the cybersecurity challenge facing federal agencies.

Source: BleepingComputer.

Apple Sets September 9 Event as Foldable iPhone Era Approaches

Apple has scheduled its next major hardware event for September 9 at Apple Park, setting the stage for what could be one of the company’s biggest iPhone lineup changes in years. The event is expected to feature the iPhone 18 Pro and iPhone 18 Pro Max, along with Apple’s long-anticipated foldable iPhone, which could carry the iPhone Ultra name. Apple is expected to delay the standard iPhone 18 until spring 2027, potentially splitting its flagship phone launches across two annual windows.

The expected devices reportedly use Apple’s A20 Pro processor manufactured on a 2-nanometer process, while the Pro models could feature an upgraded variable-aperture camera. A foldable iPhone would place Apple directly into a category where Samsung, Google, and Chinese manufacturers have already spent years refining devices. Apple has historically entered emerging hardware categories later than rivals and focused on reducing compromises before committing at scale. The September event will therefore be closely watched for evidence that foldables are moving from a premium niche into a mainstream smartphone category. Developers will also watch how Apple adapts iOS and app interfaces to a screen that can move between phone- and tablet-like dimensions.

Why It Matters: Apple entering foldable phones could push the category toward mainstream adoption and trigger a new hardware and software cycle across the smartphone industry.

Source: MacRumors.

Salesforce Puts Its CRM Inside Claude With New “Claudeforce” AI Partnership

Salesforce and Anthropic have expanded their partnership with Claudeforce, an initiative that puts Salesforce data, business rules, workflows, and actions directly inside Claude. The first product, Salesforce in Claude, launches with 37 prebuilt sales skills covering tasks such as meeting preparation, pipeline analysis, and deal-health reviews. Users can query live Salesforce information and take permitted actions without opening the traditional Salesforce interface. Pilot customers have access now, with an open beta planned for September.

The bigger story is what this says about the future of enterprise software interfaces. Salesforce spent decades building increasingly sophisticated dashboards, menus, and workflows. AI agents make it possible for those interfaces to matter less, because users can communicate their intent directly to an AI system that operates the underlying software. Salesforce is responding by making its platform accessible through Claude rather than insisting that customers remain inside its own interface. Claude is also being integrated across Agentforce and Slack, while Salesforce says it will make Claude Code and Claude Enterprise broadly available to developers and knowledge workers. If this model works, enterprise software could increasingly become infrastructure that agents operate on users’ behalf.

Why It Matters: Salesforce is preparing for a future where AI agents, rather than application screens, become the primary interface workers use to access enterprise software.

Source: VentureBeat.

Nvidia Shows Off Groq 3 AI Inference Architecture After $20 Billion Acquisition

Nvidia used the Hot Chips conference to detail the Groq 3 LPX architecture acquired through its $20 billion purchase of Groq and presented an early third-party benchmark suggesting significant performance gains for long-context AI inference. Artificial Analysis measured a Groq 3-based system at 3,431 output tokens per second on a 100,000-context Gemma 4 31B reasoning workload, roughly four times the reported performance of the next-fastest public endpoint in that comparison.

The development shows why inference has become a strategic battleground. Training giant models gets much of the attention, but once millions of users begin querying those models, inference cost and speed determine whether AI products can operate economically at scale. Groq built its reputation around specialized language-processing hardware optimized for predictable, high-speed inference. By absorbing that architecture, Nvidia gains another tool alongside GPUs, CPUs, networking, and software as it tries to control more of the AI computing stack. Nvidia said an LP30-based rack is already in production. Real-world deployment data will matter more than early benchmarks, but the architecture could strengthen Nvidia’s ability to defend its position as inference workloads become an increasingly large share of AI infrastructure spending.

Why It Matters: Nvidia is extending beyond general-purpose GPUs with specialized inference hardware as serving AI models becomes one of the industry’s largest computing workloads.

Source: Tom’s Hardware.

Hugging Face Unveils Microduck, a $399 Desktop Biped Built for AI Training

Hugging Face’s robotics team at Pollen Robotics opened pre-orders for Microduck, a 25-centimeter, 800-gram biped priced at $399 before tax and shipping. The robot has 15 motors, a camera, LiDAR, two inertial sensors, Wi-Fi, Bluetooth, and NFC for accessories. It ships with a controller and seven trained behaviors and runs an onboard policy loop at 50 hertz. Seeed Studio in Shenzhen will manufacture it. First deliveries are targeted before Christmas in North America and Europe, with a planned run of about 20,000 units.

The software, simulator, and reinforcement-learning stack are open source. Buyers can train new skills in simulation and deploy them on the hardware. Demo footage shown to Bloomberg included roller skating, carrying small objects, and singing. Pollen, founded in Bordeaux by former Inria researchers, joined Hugging Face in 2025. Microduck follows the lab’s Reachy Mini line and aims to put a programmable walking robot on the same price shelf as a game console.

Why It Matters: A sub-$400 open-source walker lowers the cost of physical-AI experimentation for students and startups that cannot buy industrial humanoids.

Source: Bloomberg

Kioxia and Sandisk Plan More Than $31 Billion AI Memory Investment in Japan

Kioxia and Sandisk plan to invest more than $31 billion in Japan through 2032 to expand flash-memory production and develop next-generation semiconductor technology. The companies’ joint investment would include new production infrastructure at Kioxia facilities, including its Kitakami plant in Iwate Prefecture, and is contingent on support from the Japanese government. The two companies have already invested more than $50 billion together in Japan over the past quarter-century.

The announcement reflects another bottleneck emerging from the AI infrastructure boom: memory. GPUs and specialized AI accelerators receive most of the attention, but increasingly large training datasets, inference workloads, vector databases, and storage-heavy applications require enormous amounts of high-performance memory and NAND flash. Nvidia has simultaneously warned that memory costs are rising and disclosed huge future supply commitments. Japan also sees advanced semiconductors as a strategic economic and national-security priority and has been supporting domestic manufacturing as global supply chains become more politically sensitive. The Kioxia-Sandisk plan follows similarly large investments elsewhere in Asia and suggests that AI infrastructure spending is propagating through nearly every layer of the semiconductor supply chain.

Why It Matters: AI’s infrastructure boom is spreading from GPUs into memory and storage, driving tens of billions of dollars in new semiconductor capacity across Asia.

Source: The Wall Street Journal.

AI Chip Startup Architect Labs Says AI Designed a New Chip in Just Two Weeks

Architect Labs says artificial intelligence helped two human chip architects develop and verify a processor design called Redwood in roughly two weeks, compressing a process that can normally take large engineering teams more than a year. The startup used AI systems to handle much of the detailed design work while humans directed the overall architecture. Redwood has been tested through FPGA simulations but has not yet been manufactured, an important distinction as the company prepares to send the design to TSMC.

Architect Labs emerged from stealth earlier this summer with $24 million in seed funding and backing from investors that include Google DeepMind’s Jeff Dean and executives associated with OpenAI and Nvidia. The startup argues that AI-assisted semiconductor engineering could eventually do for chip design what foundries such as TSMC did for manufacturing: reduce the resources required to create custom silicon. That claim still needs validation through fabricated hardware, where issues involving power, timing, yield, reliability, and manufacturing constraints can expose problems simulations miss. But if AI substantially reduces chip-development cycles, startups and specialized industries could design processors for narrower workloads without needing enormous semiconductor engineering organizations.

Why It Matters: AI-assisted chip design could lower one of the biggest barriers to custom silicon, potentially allowing far more startups and companies to build processors optimized for specific workloads.

Source: Business Insider.

U.S. Labor Department Taps OpenAI, Google, Meta and Amazon for AI Jobs Data

The U.S. Labor Department is working with major technology companies, including OpenAI, Google, Meta, and Amazon, to better understand how artificial intelligence is changing employment and hiring. The initiative is intended to supplement traditional government labor statistics with private-sector data that could provide a faster view of AI adoption and its effects on occupations. Acting Labor Secretary Keith Sonderling is leading the effort as policymakers debate whether AI will primarily increase productivity, eliminate jobs, or reshape work in less predictable ways.

The initiative addresses a genuine measurement problem. Traditional employment surveys were built for an economy where technological change generally unfolded more slowly than today’s AI deployment cycle. Companies are introducing coding agents, customer-service automation, AI research tools, and autonomous workflows in months rather than years. That makes it difficult for government statistics to distinguish ordinary economic changes from displacement or productivity gains caused by AI. Access to private-sector signals could improve that picture, although questions will remain about methodology, representativeness, transparency, and how government agencies validate proprietary datasets. Better measurements will matter more as lawmakers consider workforce programs, education policy, taxation, and other responses to AI-driven labor changes.

Why It Matters: Governments cannot craft credible AI labor policy without reliable evidence about which jobs are disappearing, changing, or being created, making better real-time employment data increasingly important.

Source: Axios.

Amazon’s Ring Introduces TAKE Encryption to Limit Access to Home Security Video

Amazon-owned Ring is rolling out a new encryption system called TAKE, short for Throw Away the Key Encryption, as the default protection for customer video worldwide. Under the system, encryption keys rotate every five minutes and are discarded within 24 hours. Ring says the architecture is meant to preserve cloud features such as smart alerts and AI-powered video capabilities while reducing the company’s long-term ability to access customer recordings.

TAKE is different from conventional end-to-end encryption because Ring can temporarily access keys inside protected AWS Nitro Enclaves when cloud processing is required. Once the keys are destroyed, Ring says it cannot subsequently recover the associated footage without a user authorizing access again. That could limit what the company can provide in response to later law-enforcement requests. Ring’s relationship with police agencies has been a recurring privacy concern, making the design technically and politically significant. End-to-end encryption remains available as an option for customers seeking stronger separation from Ring itself. The broader challenge reflects a growing tension across consumer AI products: cloud-based intelligence often requires data access, while customers increasingly expect providers to minimize it.

Why It Matters: Ring is experimenting with a middle ground between end-to-end privacy and cloud AI features, a tradeoff more consumer technology companies will face as devices become increasingly intelligent.

Source: The Verge.

AWS and Nvidia Plan to Deploy 2 Million More GPUs in Massive AI Infrastructure Expansion

Amazon Web Services and Nvidia are dramatically expanding their infrastructure partnership, with AWS planning to deploy an additional 2 million Nvidia GPUs across its global infrastructure in 2027 and 2028. The deployment will include Blackwell Ultra, Rubin, and Rubin Ultra systems and builds on AWS’s previously announced plan to add more than 1 million Nvidia GPUs beginning in 2026. The companies say customer demand has exceeded their earlier expectations.

The partnership extends far beyond GPU purchases. Nvidia’s Vera CPUs will come to AWS, while the companies are expanding collaboration around networking, data processing, open models, robotics, and AI factories. They also plan secure AI infrastructure for the U.S. government, including systems containing 100,000 GPUs for federal and national-security workloads. Amazon Robotics will use Nvidia’s physical-AI technology for warehouse automation and next-generation robots. The sheer number of processors involved illustrates the industrial scale at which hyperscalers now plan capacity. It also complicates the narrative that Amazon’s Trainium chips and Nvidia GPUs are purely competitive products: hyperscalers increasingly build their own silicon while simultaneously buying enormous quantities of Nvidia hardware to satisfy customers demanding multiple computing options.

Why It Matters: AWS committing to another 2 million Nvidia GPUs provides one of the clearest signals yet that hyperscalers expect AI computing demand to remain enormous for years.

Source: Amazon Web Services.