Top Tech News Today, August 17, 2026: Anthropic, GE, Microsoft, Nvidia, OpenAI, Stripe, Unitree & More

It’s Monday, August 17, 2026. In the last 24 hours, the tech world moved at the speed of capital and silicon: multi-billion-dollar AI infrastructure deals locked in, a reported $7 billion AI gateway deal surfaced, carbon math for the next wave of data centers was published, quantum fabs powered on, and frontier models briefly went dark.

AI’s infrastructure race is getting bigger, more expensive, and harder to hide. Microsoft is wrestling with the physical limits of data centers and electricity, Nvidia is reportedly preparing to back roughly $100 billion in OpenAI financing, and Big Tech may have trillions more tied up in AI commitments than balance sheets suggest. At the same time, robotics is heading to public markets, AI video startups are reaching multibillion-dollar valuations, and fresh cybersecurity breaches are hitting major companies and governments.

Here are the tech stories boardrooms, investors, founders, and policymakers will be watching this week.

Technology News Today

Nvidia Nears Deal to Guarantee Roughly $100B in Credit for OpenAI Data Center Project

Nvidia is reportedly nearing an agreement to guarantee roughly $100 billion in credit supporting OpenAI’s plans for another enormous data center, according to The Information. The financing would help shore up OpenAI’s ability to develop the computing infrastructure required for increasingly expensive model training and inference workloads. The report landed alongside another Nvidia infrastructure move: talks of investing about $3 billion in SB Energy, underscoring the chipmaker’s expanding role well beyond selling GPUs.

The potential guarantee illustrates how intertwined the AI economy has become. Nvidia benefits when AI companies build more data centers because those facilities consume its processors and networking equipment. OpenAI, meanwhile, needs extraordinary amounts of capital and power to keep expanding. Financing structures that connect chip suppliers, AI labs, energy developers, cloud providers, and lenders can accelerate construction, but they also create circular relationships that investors are watching closely. Nvidia is increasingly functioning as an ecosystem financier as well as a semiconductor supplier, using its balance sheet and market position to help customers expand the infrastructure that ultimately drives demand for Nvidia hardware.

Why It Matters: Nvidia’s reported credit support shows the company evolving from the AI boom’s main chip supplier into one of the financial anchors helping fund the infrastructure itself.

Source: The Information.

Stripe Reportedly Agrees to Acquire AI Gateway Startup OpenRouter for More Than $7 Billion

Stripe has reportedly finalized a deal to acquire OpenRouter for more than $7 billion, a dramatic valuation jump for an AI infrastructure startup that was reportedly worth about $1.3 billion when it announced a $113 million Series B in May. OpenRouter gives developers a single gateway to access hundreds of AI models and route requests based on factors such as price, performance, and workload requirements. Its investors include Sequoia, Andreessen Horowitz, Menlo Ventures, and Alphabet’s CapitalG.

The strategic logic goes beyond adding another software product to Stripe. AI applications increasingly need to move among multiple model providers rather than lock themselves into a single vendor. That creates demand for a neutral orchestration layer handling model selection, billing, reliability, and switching. OpenRouter has said it serves millions of users and provides access to hundreds of models. For Stripe, such a layer could eventually sit alongside its payments infrastructure as another piece of developer plumbing. Stripe has not publicly confirmed the reported transaction and told TechCrunch that it does not comment on rumors or speculation, so the deal should still be treated as reported rather than formally announced.

Why It Matters: A multibillion-dollar OpenRouter acquisition would validate AI model routing as a major infrastructure layer, not a temporary feature between developers and model providers.

Source: TechStartups via Bloomberg.

AI Video Startup Higgsfield Raises $400M at $5.4B Valuation as Enterprise Revenue Surges

AI video generation startup Higgsfield has raised $400 million at a $5.4 billion valuation, backed by investors including Goldman Sachs, Intel, DST Global, and Liberty Global. Founded by former Snap executive Alex Mashrabov, the two-year-old company has grown at a striking pace: its annualized revenue reportedly climbed from roughly $20 million a year ago to $700 million in August. Higgsfield now counts more than 30 million users across 238 countries and territories, with the U.S. as its largest market.

The bigger shift is where that revenue is coming from. Higgsfield initially gained traction among creators generating social-media visuals, but Mashrabov told the Financial Times that businesses now account for most of its revenue, compared with less than a quarter in January. That transition puts Higgsfield into a much larger contest over enterprise content production, where AI-generated video could lower production costs and dramatically increase the volume of advertising and marketing material companies can produce. The funding also shows investors are still willing to assign multibillion-dollar valuations to AI applications showing rapid revenue growth rather than betting exclusively on foundation-model companies.

Why It Matters: Higgsfield’s growth suggests AI-generated video is moving from creator experimentation into a serious enterprise software category.

Source: Financial Times.

Microsoft’s Massive AI Buildout Faces Questions Over Chip and Data Center Capacity

Microsoft’s enormous AI infrastructure expansion is facing new scrutiny after a Guardian investigation identified an apparent gap between the company’s reported data center capacity and the number of advanced AI chips believed to be operating inside those facilities. Internal documents reviewed by the Guardian indicate Microsoft has about 2.2 million AI chips installed globally, despite spending roughly $280 billion since 2022 on land, buildings, computing infrastructure, and related AI expansion. Microsoft disputed the publication’s calculations but did not specify which numbers it considered incorrect.

The issue may be less about acquiring GPUs than actually connecting them to usable facilities and electricity. Microsoft CEO Satya Nadella has previously described power availability and completed data center shells as major bottlenecks, noting that chips can sit unused when there is nowhere to plug them in. The Guardian also pointed to delays surrounding the company’s Fairwater facilities and questions about how much announced capacity is truly online. That distinction matters across the AI industry: billions of dollars of chips, land, leases, and electricity agreements do not automatically translate into active compute. For investors and startups relying on hyperscale cloud capacity, operational megawatts increasingly matter as much as headline capital expenditure.

Why It Matters: The AI infrastructure race is shifting from buying GPUs to the harder challenge of bringing enough power, networking, and physical data center capacity online.

Source: The Guardian.

Big Tech’s AI Spending Commitments May Be $3 Trillion Higher Than Balance Sheets Suggest

The AI infrastructure boom may be significantly larger than conventional capital-spending numbers indicate. A Wall Street Journal analysis found that nine major technology companies had roughly $3 trillion in commitments, many related to AI infrastructure, that do not appear as traditional debt on their balance sheets. The obligations include long-term data center leases, chip purchasing agreements, financing arrangements, and other commitments being used to secure scarce computing capacity.

That matters because the economics of the AI race cannot be measured by quarterly capital expenditures alone. Cloud giants and AI companies increasingly use partnerships, leasing structures, special-purpose financing, and long-term purchase contracts to obtain infrastructure without immediately recording the entire economic commitment as conventional corporate debt. The result is an AI investment cycle whose true scale may be less visible than headline spending suggests. It also raises the stakes around utilization: trillions of dollars in commitments ultimately require sufficient demand for AI inference, training, agents, and enterprise applications to justify them. If demand continues rising, these arrangements could secure the computing backbone of a new software economy. If utilization disappoints, long-term infrastructure obligations could become far more consequential.

Why It Matters: AI’s financial footprint extends far beyond reported capex, making infrastructure commitments an increasingly important measure of Big Tech’s real exposure to the boom.

Source: The Wall Street Journal.

OpenAI Reportedly Disbands Preparedness Team as AI Safety Responsibilities Shift

OpenAI has reportedly disbanded the team responsible for evaluating whether increasingly capable AI models could create severe risks. The preparedness team was dissolved at the end of July, with responsibility for specific areas such as biological and cybersecurity risks distributed into other teams, according to reporting cited by The Verge. Team leader Dylan Scandinaro will reportedly shift his attention to the implications of recursively self-improving AI systems.

The organizational change comes amid a broader reshaping of OpenAI’s safety structure. Previous groups focused on superalignment and AGI readiness have also been dissolved or reorganized, while several prominent safety and research employees have departed. Moving preparedness work into individual operating groups does not necessarily mean OpenAI is abandoning risk assessment; distributed ownership can sometimes bring safety work closer to the teams actually building products. But the move will inevitably draw scrutiny because frontier models are becoming more capable at coding, cyber operations, biological research, and autonomous tasks, even as OpenAI prepares for an eventual public-market push. How companies govern those risks is becoming as important to regulators as model benchmarks themselves.

Why It Matters: OpenAI’s reorganization raises a fundamental governance question: whether a dedicated independent team or product and research teams should handle AI safety.

Source: The Verge.

ChatGPT’s New Computer History Feature Tracks User Activity Without Taking Screenshots

OpenAI is developing a feature called Computer History that lets ChatGPT recall activity on a user’s computer. In a demonstration highlighted by The Verge, OpenAI developer-experience team member Dominik Kundel showed the system locating the last document he edited, checking whether he had shared it through Slack, and generating a summary of his morning activity. Unlike Microsoft’s controversial Recall feature, OpenAI says Computer History does not continuously capture screenshots, video, or audio. Instead, it records structured activity described as “events.”

The feature points toward a larger change in personal computing. AI assistants become significantly more useful when they have context about what users have already done, which files they touched, who they communicated with, and what tasks remain unfinished. That same context creates obvious privacy and security concerns, particularly if an attacker, malicious extension, or compromised AI agent gains access to the historical record. The design challenge is therefore not simply giving assistants more memory but establishing strict boundaries around collection, retention, permissions, and deletion. Personal computing may increasingly revolve around AI-maintained activity histories, making those controls a new layer of operating-system security.

Why It Matters: Computer History moves ChatGPT closer to a persistent personal computing assistant, while making privacy controls around AI memory increasingly critical.

Source: The Verge.

French Tax Authority Data Breach Exposes Information on 678,000 People

France’s Ministry of Economy and Finance has disclosed a cybersecurity breach affecting approximately 678,000 individuals after an attacker gained unauthorized access to systems belonging to the General Directorate of Public Finances, or DGFiP. The agency manages some of France’s most sensitive tax and financial administration systems, making any unauthorized access significant even when the compromised dataset does not include complete taxpayer records.

The incident adds to a growing list of attacks targeting government agencies whose centralized databases contain information useful for identity theft, phishing, financial fraud, and more targeted social-engineering campaigns. Public-sector breaches can have unusually long consequences because citizens cannot simply change many of the underlying identifiers connected to government records. The attack also demonstrates that digital modernization creates a larger concentration of risk: centralized online tax systems improve efficiency but become attractive targets because compromising one platform can expose information on hundreds of thousands of people at once. European governments have been increasing cybersecurity spending, yet public agencies remain under pressure from both financially motivated attackers and state-linked groups.

Why It Matters: The French breach shows how government digital infrastructure remains a high-value target because a single intrusion can expose sensitive information belonging to hundreds of thousands of citizens.

Source: BleepingComputer.

Anthropic Experiences 42-Minute Claude Outage Affecting Login and Multiple Services

Anthropic’s Claude platform suffered a service disruption on the evening of August 16 that began with authentication failures and spread to degraded performance across claude.ai, Claude Code, Claude Cowork and related interfaces. The company’s status page logged the start at 21:58 UTC and marked full restoration by approximately 22:40 UTC. The Claude API remained largely available while user-facing web and desktop products went dark for the duration.

No root cause was disclosed, and the incident adds to a series of recent stability issues the company reported in the weeks leading up to its anticipated IPO preparations. Enterprise customers relying on agentic workflows experienced temporary interruptions in production use cases.

Why It Matters: Even brief outages at frontier AI providers highlight the operational fragility of systems now embedded in critical business processes.

Source: Unite.AI.

GE and Philips Investigate Clop Ransomware Claims as Supply-Chain Attacks Spread

General Electric and Philips are investigating claims by the Clop ransomware group that it breached their systems and stole data, according to BleepingComputer. Both companies are major multinational technology and industrial organizations, so any confirmed compromise could have wide implications across manufacturing, healthcare equipment, corporate systems, suppliers, and customers. The companies’ investigations are ongoing, so claims made by the attackers should not be treated as independently verified evidence of what data, if any, was stolen.

Clop has become known for exploiting vulnerabilities in widely deployed enterprise software and using stolen information for extortion. That model can be particularly damaging because a single flaw in a shared product can expose dozens or hundreds of organizations simultaneously. The broader lesson for enterprises is that cybersecurity risk increasingly comes from software dependencies and third-party systems rather than direct attacks on a company’s perimeter. Industrial and healthcare companies face additional pressure because operational disruption can extend beyond ordinary IT systems into equipment, manufacturing processes, or critical services. As investigations continue, key questions include how attackers gained access, what information they accessed, and whether the incidents stem from a shared technology supplier.

Why It Matters: The GE and Philips investigations highlight how one compromised enterprise technology layer can potentially expose distributed companies across multiple industries globally.

Source: BleepingComputer.

DoiT Acquires AI Cost-Management Startup Attribute in Deal Estimated at $65 Million

Cloud optimization company DoiT has acquired Israeli startup Attribute in a deal estimated at roughly $65 million. Founded in 2023 by Izhak Zimmermann and Liad Tropp, Attribute built software that gives organizations a real-time view of spending across AI tokens, models, agents, cloud services, products, teams, and even individual customers. The startup previously raised about $13.5 million in seed funding.

The acquisition points to a problem emerging as companies move from AI pilots into production: many organizations have surprisingly little visibility into what their AI workloads actually cost. Traditional cloud FinOps tools were built around virtual machines, storage, networking, and database consumption. Generative AI introduces another layer of variable spending tied to tokens, model selection, agents, inference frequency, and external APIs. DoiT said its data suggests monthly AI spending among customers could triple over the next year, while only 15% of executives in one survey said they could calculate AI return on investment. Attribute gives DoiT a way to connect AI consumption directly to teams, features, products, and customers rather than treating model costs as a single undifferentiated cloud bill.

Why It Matters: As AI spending scales, tools that show exactly which agents, models, products, and customers are consuming money could become essential enterprise infrastructure.

Source: CTech.

OpenAI Funds 14 Global Policy Projects Studying AI’s Economic and Social Impact

OpenAI is funding 14 policy and research projects examining how artificial intelligence could reshape employment, economic opportunity, public policy, and societal resilience. The selected organizations will collectively receive $1 million in funding plus up to another $1 million in OpenAI model credits. Recipients span the U.S. political spectrum and include organizations in Europe, Brazil, Singapore, and South Korea.

Among the projects, the American Enterprise Institute and Urban Institute will collaborate on research examining AI’s effects on the U.S. workforce, while the Progressive Policy Institute plans to explore a person-based benefits system for an economy where work may become more fluid. The initiative reflects growing recognition inside the AI industry that the next phase of competition will involve public policy as much as model performance. Governments are already grappling with job displacement, copyright, safety, energy use, data centers, and national competitiveness. OpenAI also has a direct interest in shaping that debate, so outside research funded by the company will inevitably be examined for independence and transparency alongside its findings.

Why It Matters: Frontier AI companies are increasingly investing in the policy ecosystem that will help determine how their technology is regulated and absorbed into the economy.

Source: Semafor.

AI Boom Drives New Push to Rebuild U.S. Copper Processing and Modernize the Power Grid

The massive electricity buildout required for AI is creating opportunities far beyond data centers and semiconductor factories. A new Axios report highlights growing investment in domestic copper processing as demand rises from AI facilities, transmission infrastructure, electric vehicles, and grid modernization. The United States has significant copper resources but relies heavily on overseas processing, including facilities in China. Startups such as Red Metals and Still Bright are developing new approaches to expand domestic processing capacity.

At the same time, utilities are rebuilding parts of the electric grid to improve wildfire resilience, expand transmission, build microgrids, and meet rising power demand. Thousands of miles of new transmission infrastructure may ultimately be required to connect data centers, renewable energy projects, storage systems, and population centers. Copper sits underneath much of that buildout because electrical systems require huge volumes of conductive wiring. The result is a reminder that AI infrastructure is ultimately physical infrastructure. Models may exist in software, but scaling them depends on mines, metals, transmission lines, substations, transformers, cooling equipment, and electricity generation.

Why It Matters: AI’s infrastructure bottleneck is expanding downstream from GPUs and electricity into basic industrial materials such as copper and the grid equipment needed to deliver power.

Source: Axios.

AI Data Center Optical Interconnect Market Projected to Reach $144 Billion by 2030

The market for optical interconnect technology inside AI data centers is projected to exceed $144 billion by 2030, more than ten times its 2024 level, according to new industry projections reported by Tom’s Hardware. Silicon photonics is expected to capture a large share of that growth as data centers move toward higher-speed connections and co-packaged optical systems. Demand for older 400 Gbps technology is expected to flatten as 1.6 Tbps networking becomes increasingly important for next-generation AI clusters.

Networking has become one of the less visible constraints in scaling AI. Training a frontier model does not simply require thousands of GPUs; those processors must exchange enormous quantities of data with extremely low latency. As clusters grow into tens or hundreds of thousands of accelerators, electrical connections become increasingly difficult to scale because of power consumption, heat, signal loss, and physical distance. Optical technologies move more data using light and can reduce some of those constraints. Nvidia, Broadcom, Marvell, hyperscalers, networking companies, and photonics startups are all positioning around this transition.

Why It Matters: The next AI infrastructure battle will increasingly be about connecting GPUs together efficiently, making photonics and high-speed networking strategic technologies alongside the processors themselves.

Source: Tom’s Hardware.

China’s Unitree Robotics Sets August 19 Shanghai IPO After Record Investor Demand

Chinese humanoid robotics company Unitree is scheduled to begin trading on Shanghai’s STAR Market on August 19, becoming the first general-purpose robotics company to list on mainland China’s public markets. Reuters reports that Unitree is already the world’s largest humanoid robot maker by sales, while its IPO was more than 8,000 times oversubscribed by retail investors, a record for the technology-focused STAR Market.

Unitree has attracted international attention with humanoid and quadruped robots capable of running, dancing, navigating environments, and performing complex movements. The listing comes as AI investment increasingly shifts from software to “physical AI,” where machine-learning systems control robots operating in factories, warehouses, homes, and public environments. China has made robotics a strategic industrial priority and benefits from an enormous domestic manufacturing supply chain that could help lower hardware costs faster than Western rivals. The IPO will provide an unusually public test of investor appetite for humanoid robotics as an operating business rather than a private venture-capital story.

Why It Matters: Unitree’s listing marks an important transition for humanoid robotics from heavily funded experimentation toward public-market scrutiny, commercialization, and industrial scale.

Source: Reuters.

Trump-Backed Crypto Firm Partners With AI Platform Offering Models From Restricted Chinese Tech Companies

World Liberty Financial, the crypto venture backed by President Donald Trump and his family, is collaborating with Hong Kong-based WorldClaw, an AI platform offering access to models from both U.S. and Chinese developers. A Reuters review found that 43 of the 90 models available through WorldClaw were developed by Chinese companies including Alibaba, Baidu, Z.ai, DeepSeek, and Moonshot. Several of those companies have been designated or criticized by the U.S. government over alleged national-security, military, or intellectual-property concerns. The companies have disputed several of Washington’s allegations.

WorldClaw accepts World Liberty’s USD1 stablecoin for payment, and a World Liberty executive has served as an outside adviser to the AI venture. Reuters stressed that the collaboration itself is not illegal. WorldClaw also offers models from U.S. companies including OpenAI and Anthropic, reflecting the increasingly global market for model-routing platforms. The situation nevertheless exposes a tension between national AI policy and the commercial reality of open and inexpensive Chinese models gaining adoption worldwide. It also raises questions about data routing because WorldClaw says user inputs may be shared with underlying model providers.

Why It Matters: The partnership shows how Chinese AI models are penetrating global developer ecosystems even as Washington attempts to restrict Chinese technology and strengthen U.S. leadership in AI.

Source: Reuters.