Top Tech News Today, August 14, 2026: Apple, Anthropic, DeepSeek, Google, IBM, Pony.ai, OpenAI, SpaceX, Uber & More

It’s Friday, August 14, 2026, and the last 24 hours just reminded everyone why tech never sleeps. Google dropped a sharper, cheaper AI workhorse overnight. Apple quietly trained its own model for China with Alibaba’s help.

The AI boom is entering a more consequential phase. What started as a race to build smarter models is becoming a global contest for chips, data centers, energy, autonomous systems, cybersecurity, and the capital to fund it all. Big Tech’s AI purchase commitments are approaching $1.5 trillion, SMIC is raising chip prices as factories run near capacity, and Uber and Pony.ai are preparing to put more than 2,000 robotaxis on European roads.

At the same time, AI economics are beginning to shift. OpenAI and Anthropic are cutting prices while DeepSeek is raising them, faster inference is becoming a competitive weapon, and companies are discovering that the real cost of AI may depend as much on infrastructure and orchestration as on the model itself.

The bigger story: AI is no longer just changing software. It is beginning to reshape the physical and financial infrastructure of the global technology economy. These aren’t incremental updates. They’re the moves shaping the next wave of AI infrastructure, Big Tech strategy, cybersecurity, consumer hardware, and global regulation.

Here are the top tech stories that matter most today.

Technology News Today

Apple Builds Its Own AI Model for China With Alibaba as iPhone AI Strategy Shifts

Apple has trained a custom artificial intelligence model for China with help from Alibaba, marking a significant shift in how the iPhone maker plans to bring Apple Intelligence to one of its largest and most tightly regulated markets. The China-focused large language model was trained with Alibaba’s support and would give Apple greater control over the AI running on devices sold in the country. Apple has previously relied more heavily on models developed by Chinese partners because services such as OpenAI’s ChatGPT are unavailable in mainland China.

The strategy matters beyond Apple. China requires generative AI services offered to the public to clear regulatory requirements, creating a very different operating environment from the U.S. and Europe. Apple recently registered its on-device generative AI service with Chinese regulators, clearing an important hurdle ahead of a planned Apple Intelligence rollout. A proprietary China model could help Apple compete more directly with Huawei and other domestic smartphone makers that have made AI central to their products. It also illustrates how geopolitical fragmentation is creating separate technology stacks: global companies may increasingly need different models, infrastructure partners, and compliance strategies for different markets.

Why It Matters: Apple’s China-specific model shows how AI regulation and geopolitics are beginning to reshape even the underlying software architecture of global consumer technology.

Source: The Verge.

Google Unveils Gemini 3.7 Flash AI Model for Coding and Agent Workflows

Google launched Gemini 3.7 Flash on Thursday, describing it as its most intelligent workhorse model yet for software engineering, knowledge work, and autonomous agent tasks. Released just three weeks after Gemini 3.6 Flash, the new version delivers measurable gains in coding benchmarks, including a jump from 34.4% to 43.6% on FrontierCode 1.1 Main and from 49% to 65.3% on DeepSWE v1.1. It supports a 1-million-token context window, improved first-pass code accuracy, and better production-ready generation. Pricing starts at an introductory rate of $0.75 per million input tokens and $3.75 per million output tokens through year-end—half the previous Flash cost—to encourage broader adoption by developers building multi-step agents. The model is available immediately via the Gemini API, AI Studio, Gemini Enterprise, and the Spark agent for AI Pro and Ultra subscribers.

This rapid iteration reflects Google’s push to close the gap with rivals in practical agentic AI while keeping costs low enough for production-scale deployment. By focusing on coding and workflow automation rather than a full Pro-tier leap, Google is targeting the everyday developer and enterprise use cases that drive real revenue in the current AI infrastructure boom. The price cut also pressures competitors on value, potentially accelerating migration of coding assistants and internal agents onto Google’s stack.

Why It Matters: Faster, cheaper agent-ready models lower barriers for startups and enterprises racing to deploy production AI systems.

Source: Reuters.

OpenAI and Anthropic Cut AI Prices as Chinese Models Intensify Global Competition

The AI model business is moving into a serious price war. OpenAI and Anthropic are lowering prices on some models as lower-cost Chinese competitors including DeepSeek and Moonshot AI gain users among companies looking to control increasingly large inference bills. OpenAI has cut pricing for GPT-5.6 Luna substantially, while Anthropic is positioning Claude Opus 5 at roughly half the price of its higher-end Fable 5 model. Data cited by the Financial Times indicates that prices customers pay for leading U.S. models have declined materially since mid-July.

The shift changes the competitive equation in AI. For several years, model developers competed primarily on benchmarks, context windows, coding performance, and reasoning capabilities. Enterprise buyers are increasingly asking a simpler question: how much useful work does each dollar of inference buy? That favors efficient models and gives Chinese developers an opening even when they do not lead every benchmark. It also pressures frontier labs carrying enormous computing, staffing, and data-center expenses. If model intelligence becomes increasingly interchangeable for routine workloads, margins could compress much faster than many AI business models assume. For startups, however, lower inference costs could unlock applications that previously failed basic unit-economics tests.

Why It Matters: AI is shifting from a benchmark race toward an economics race, where cost per completed task may matter as much as raw model intelligence.

Source: Financial Times.

DeepSeek Launches V4 Pro and Raises AI API Prices by More Than 1,000% on Some Workloads

Chinese AI startup DeepSeek has formally introduced its V4 Pro flagship while sharply increasing what customers pay to use its higher-performance model. Caixin reported that some API pricing is rising by as much as 1,100%, while DeepSeek’s published pricing shows V4 Pro costing up to $1.32 per million cache-miss input tokens and $3.96 per million output tokens during peak periods. Lower off-peak rates remain available, and the company continues offering its cheaper V4 Flash model.

The increase is notable because DeepSeek became globally prominent largely by demonstrating that strong AI performance could be delivered at dramatically lower prices than many Western competitors. V4 Pro suggests the company now sees room to monetize higher-value workloads rather than competing exclusively through rock-bottom pricing. Even after the increase, DeepSeek remains inexpensive relative to several frontier alternatives, but the move highlights an emerging reality: inference prices may not continually move in one direction. Better models consume valuable compute, and providers with meaningful demand will test customers’ willingness to pay for reliability, coding performance, reasoning, and throughput. DeepSeek’s move also arrives as U.S. model companies cut selected prices, creating an unusual market where premium Chinese models are getting more expensive while some American alternatives become cheaper.

Why It Matters: DeepSeek’s pricing shift suggests leading Chinese AI companies are moving from disruption through cheap inference toward monetizing premium model performance.

Source: Reuters.

China’s Z.ai Unveils GLM-5.3 AI Model With Near-Frontier Cybersecurity Performance

Chinese AI company Z.ai has unveiled GLM-5.3, an open-weights model that the company says approaches Anthropic’s Mythos 5 on some cybersecurity tasks. Z.ai reported an 84.5% score on the CyberGym vulnerability-detection benchmark, slightly above the 83.8% it cited for Mythos 5. The gap was considerably wider on exploit development, where GLM-5.3 scored 54.4% against 78% for Anthropic’s model. Z.ai says the model also delivers a roughly 50% improvement over GLM-5.2 on its internal coding benchmark.

The more important development may be how Z.ai plans to release it. The company intends to make GLM-5.3 publicly available after additional security testing while limiting some higher-risk capabilities through a trusted-access system. That approach attempts to preserve the advantages of open models while acknowledging that advanced cybersecurity skills can be dual-use. Models that can identify software vulnerabilities can help defenders patch systems faster, but the same capabilities can lower the barrier to offensive operations. The announcement also suggests that Chinese AI labs are narrowing capability gaps in specialized domains rather than competing only on price. Cybersecurity may become one of the most consequential tests of how governments and developers balance open access against misuse risks.

Why It Matters: GLM-5.3 shows that open AI models are gaining increasingly sophisticated cyber capabilities, forcing developers to rethink how advanced models can remain accessible without making offensive tools universally available.

Source: South China Morning Post.

French Tax Agency Cyberattack Exposes Data From Individual and Business Taxpayers

France’s Finance Ministry has confirmed that attackers accessed and extracted taxpayer information from the country’s tax administration systems. The breach affects both individuals and professional taxpayers. According to the ministry, a malicious actor claimed responsibility after gaining access to the General Directorate of Public Finances, and investigations confirmed unauthorized consultation and extraction of taxpayer data. Reuters reported that a French breach-tracking service estimated data from close to 700,000 taxpayers may have been stolen, although the government had not independently confirmed that figure.

The incident is particularly sensitive because tax agencies hold unusually rich collections of identity, financial, employment, property, and business information. Even without passwords or payment credentials, stolen government records can become valuable inputs for identity theft, targeted phishing, fraud, and social-engineering attacks. French authorities said affected users would receive individual notices describing what information may have been accessed and what precautions they should take. The breach also adds to a growing cybersecurity challenge for governments: public agencies increasingly operate large digital services containing information that cannot simply be replaced after theft, as a compromised password can. Protecting these databases therefore becomes a long-term identity-security issue as much as an incident-response problem.

Why It Matters: Government data breaches can expose permanent identity information at enormous scale, giving attackers material that can support fraud and targeted attacks for years.

Source: Reuters.

China’s SMIC Raises Chip Prices as AI Demand Pushes Factories Toward Full Capacity

Semiconductor Manufacturing International Corp., China’s largest chip foundry, is raising prices for some of its most sought-after manufacturing capacity as AI-related demand keeps factories near capacity. SMIC’s utilization rate climbed to 93.7% in the second quarter, while the company shipped about 2.9 million 8-inch-equivalent wafers. Revenue topped $3 billion for the first time, and quarterly profit more than tripled to $479.2 million. DIGITIMES also reported that SMIC plans to begin breaking out AI chip revenue separately as demand becomes significant enough to warrant its own reporting category.

The story extends beyond advanced GPUs. AI infrastructure is consuming capacity across memory, networking, controllers, power-management chips and mature semiconductor nodes, creating spillover demand through much of the supply chain. SMIC has added 12-inch wafer capacity and is accelerating production-line ramp-ups, but building semiconductor capacity takes time and billions of dollars. The company also remains constrained by U.S. export controls that limit access to some advanced manufacturing equipment. Strong domestic demand nevertheless gives SMIC a large captive market as Chinese technology companies try to reduce reliance on foreign semiconductor suppliers. Higher foundry pricing also adds another source of cost pressure that can eventually work its way into servers, industrial systems, and consumer electronics.

Why It Matters: AI demand is tightening semiconductor capacity well beyond Nvidia-style accelerators, strengthening Chinese foundries while pushing chip manufacturing costs higher.

Source: DIGITIMES.

Pony.ai and Uber Plan More Than 2,000 Robotaxis Across Five European Cities

Pony.ai and Uber are significantly expanding their autonomous-driving partnership, with plans to deploy more than 2,000 Pony.ai robotaxis across five European cities. The companies already operate together in Zagreb, Croatia, and intend to expand into four additional European markets. Pony.ai said the partnership will also extend into the Middle East, although it has not yet provided a full deployment timetable or named the additional European cities.

Scale matters more than the headline vehicle count. Robotaxi companies have spent years conducting tightly controlled pilots containing dozens or hundreds of vehicles. Deployments measured in thousands begin testing whether autonomous driving can operate as transportation infrastructure rather than a technology demonstration. Uber’s strategy is also becoming clearer: instead of developing a single proprietary autonomous-driving stack, it is positioning its ride-hailing network as the distribution layer for multiple robotaxi providers. Pony.ai gains immediate access to customers, dispatch software, payments, and demand density, while Uber avoids bearing the full technical cost of developing autonomous vehicles. Europe remains a challenging market because regulations differ between jurisdictions, but a successful multi-city rollout could provide a template for faster international expansion by autonomous-driving companies.

Why It Matters: A 2,000-vehicle deployment would move European robotaxis closer to commercial fleet scale and strengthen Uber’s position as the marketplace connecting autonomous-driving companies with riders.

Source: Pony.ai.

Big Tech’s AI Purchase Commitments Approach $1.5 Trillion as Infrastructure Obligations Surge

The financial footprint of the AI infrastructure race is becoming far larger than headline capital-expenditure numbers suggest. Alphabet, Microsoft, Amazon, Nvidia, Oracle and Meta have accumulated close to $1.5 trillion in purchase commitments tied heavily to computing infrastructure, chips, data-center capacity and energy, according to Financial Times analysis. Those obligations are separate from roughly another $1.5 trillion in lease commitments identified by Goldman Sachs. Alphabet alone reported purchase commitments rising dramatically between the first and second quarters as it secured long-term technical infrastructure and energy agreements.

These commitments matter because they represent future cash obligations that may not appear on conventional balance sheets like debt. AI companies and hyperscalers are locking in GPUs, servers, electricity, construction, networking and data-center capacity years before they know exactly how profitable future AI demand will become. That strategy can protect companies from shortages, but it also reduces their ability to pull back quickly if returns disappoint. The economics increasingly resemble heavy industry: massive upfront commitments, long-lived contracts and high fixed costs must eventually be supported by revenue. For investors, evaluating an AI company may therefore require looking beyond annual capital expenditures toward leases, supply agreements and other contractual obligations.

Why It Matters: AI infrastructure risk increasingly sits in long-term contractual commitments, meaning the true financial exposure of Big Tech’s compute race may be much larger than standard capex figures imply.

Source: Financial Times.

Startup Boom Accelerates as 40 New Unicorns Join the Ranks in a Single Month

Forty privately held companies reached valuations of at least $1 billion in July, the highest monthly addition to Crunchbase’s Unicorn Board in more than four years. Three newcomers were valued above $10 billion, while financial services, robotics, AI orchestration, multimodal AI, energy, and semiconductors ranked among the strongest categories.

The data offers a broader signal about venture markets after several years of more selective fundraising, as many startups avoided raising capital at lower valuations. AI has clearly reopened the upper end of the private market, but sector composition matters. Investors are increasingly funding businesses connected to physical infrastructure, robotics, chips and energy alongside software models and applications. That suggests the current funding cycle is spreading through the AI supply chain rather than remaining concentrated in foundation-model companies. At the same time, a surge in unicorn creation can be a warning as well as a sign of strength. Private valuations depend heavily on growth assumptions, future financing conditions and public-market exit opportunities. The next test is whether the expanding crop of billion-dollar startups can produce enough revenue and durable margins to justify those prices.

Why It Matters: The strongest burst of unicorn creation in four years shows risk capital returning aggressively, with AI now pulling funding into robotics, chips, energy and financial technology alongside software.

Source: Crunchbase News.

Quantum Tech Company Infleqtion Doubles Revenue as NASA Work Drives Commercial Growth

Quantum technology company Infleqtion reported second-quarter revenue of $12.6 million, up 116% from a year earlier and ahead of Wall Street expectations, with NASA project milestones contributing significantly to the increase. The company raised its full-year revenue outlook to roughly $43 million, although operating losses widened as it continued investing in technology development and absorbed higher stock-based compensation expenses.

Infleqtion is unusual among publicly followed quantum companies because its business extends beyond the race to build general-purpose quantum computers. The company develops neutral-atom quantum systems as well as quantum sensors and atomic clocks, technologies with potential applications in navigation, communications, aerospace and national security. It already works with agencies including NASA, the Department of Energy, DARPA and the U.S. Air Force research ecosystem. That government revenue gives Infleqtion a path to commercialization that does not depend entirely on waiting for fault-tolerant quantum computing to become economically useful. The broader quantum sector has attracted enormous investor enthusiasm despite limited present-day commercial revenue at many companies. Infleqtion’s results offer a different yardstick: whether quantum physics can deliver valuable sensing, timing, and government applications while full-scale quantum computing continues to develop.

Why It Matters: Quantum companies may find meaningful businesses in sensing, timing and government systems long before universal quantum computers reach mass commercial adoption.

Source: Barron’s.

Enterprise AI Startup Writer Launches Palmyra X6 and New Harness to Cut Agent Costs

Enterprise AI company Writer has introduced its Palmyra X6 model alongside an upgraded agent harness aimed at reducing one of the fastest-growing costs in production AI: token consumption. Writer estimates that combining the new model with changes to its orchestration infrastructure can reduce customer costs by as much as 50% on basic tasks. The company’s underlying research argues that the software controlling how agents retrieve context, invoke tools, retry failed tasks, and manage conversation history can influence economics as much as the choice of model itself.

That idea could become increasingly important as companies move from AI chat interfaces to agents that perform multi-step work. A single agent task may involve dozens of model calls, tool invocations, and repeated context windows, causing token usage to compound quickly. Writer’s research found that changing the orchestration layer while holding models constant reduced tokens per task and cost per task while maintaining comparable completion quality. If those findings translate broadly into production systems, competition in enterprise AI may shift away from simply selecting the smartest model toward engineering efficient systems around multiple models. That could create a new infrastructure category around routing, memory, context management, and agent execution.

Why It Matters: As AI agents become more complex, orchestration efficiency may become one of the biggest determinants of whether enterprise AI produces attractive unit economics.

Source: TechCrunch.

IBM and OpenAI Expand Enterprise AI Alliance Across Business Operations and Cybersecurity

IBM and OpenAI have announced a broad strategic partnership to deploy OpenAI models and tools across enterprise workflows, application modernization, software development, and cybersecurity. IBM said its consultants and engineers will help customers integrate OpenAI technology into core operations, initially focusing on industries including financial services, government, telecommunications, and retail. The partnership also connects IBM more deeply with OpenAI’s enterprise partner ecosystem and cybersecurity initiatives.

The alliance reflects a major shift in how the enterprise AI market is taking shape. Foundation-model developers can build highly capable technology, but many large corporations still depend on consulting firms and systems integrators to connect new models with decades-old databases, business applications, security policies and regulatory requirements. IBM brings those relationships and integration capabilities; OpenAI brings models, coding systems and AI products. The combination also illustrates why the next phase of AI adoption may be less visible than the chatbot boom. Large organizations increasingly want AI embedded into accounting, compliance, customer support, software maintenance and security workflows rather than deployed as standalone experimental tools. Consulting firms may therefore become a primary distribution channel through which frontier AI reaches traditional businesses.

Why It Matters: The IBM-OpenAI partnership shows that enterprise AI adoption is moving from experimentation toward integrating models directly into the systems that run large organizations.

Source: IBM.

Space Startup Astra Seeks $250 Million at $1 Billion Valuation in Comeback Attempt

Astra Space is seeking $250 million in fresh capital at roughly a $1 billion valuation as the rocket company tries to rebuild itself after launch failures and a dramatic collapse in market value. CEO Chris Kemp told Reuters that the fundraising is expected to close this quarter. Astra once traded above $2 billion after going public in 2021 but was taken private in 2024 for just $11.25 million after its original launch program struggled.

The company’s second act centers on Rocket 4, satellite propulsion, and a strategy built around inexpensive, rapidly deployable launch systems. Astra is targeting Rocket 4 launches beginning in 2027 and says the expendable vehicle could cost roughly $5 million per launch. Rather than competing directly with SpaceX on large reusable rockets, Astra is emphasizing mobile launch capabilities that could let military customers place satellites into orbit from dispersed locations. Meanwhile, its spacecraft-propulsion business has delivered hundreds of thrusters, giving Astra another revenue stream while rocket development continues. The comeback remains highly speculative: launch vehicles are capital intensive, technically unforgiving and dominated by better-funded rivals. Yet national-security demand for responsive space access is creating opportunities for companies that can launch smaller payloads quickly.

Why It Matters: Astra’s comeback tests whether national-security demand and low-cost responsive launch can revive a space startup that nearly disappeared after its first rocket program failed.

Source: Reuters.

OpenAI Launches Ultrafast GPT-5.6 Sol Mode Running Up to 14 Times Faster

OpenAI has introduced an early preview of Ultrafast, a new API service tier that runs GPT-5.6 Sol up to 14 times faster than standard processing. Initially available to a limited group of API customers, the service is powered by Cerebras and can generate up to 750 output tokens per second. OpenAI is targeting workloads where latency matters enough that developers will pay for substantially faster inference.

The announcement highlights another frontier in AI competition that receives less attention than model benchmarks: speed. Long reasoning times are acceptable for research reports or asynchronous coding jobs, but they become a serious limitation when AI is embedded inside real-time software, customer interactions, trading workflows, interactive coding environments or autonomous agents waiting for a model before taking their next action. Faster inference can therefore expand the set of economically practical AI applications even when the underlying model itself does not become smarter. Cerebras said tests on economically valuable knowledge-work tasks showed large end-to-end speed gains without corresponding quality degradation. Specialized inference hardware is also emerging as a stronger competitive layer alongside Nvidia GPUs, giving model companies additional options for serving workloads optimized for latency rather than training.

Why It Matters: The AI infrastructure battle is increasingly about latency as well as intelligence and price, opening opportunities for specialized inference hardware and real-time AI applications.

Source: OpenAI.

Applied Materials Plans Major Chip Equipment Expansion as AI Demand Pushes Revenue to $9.1 Billion

Applied Materials reported fiscal third-quarter revenue of $9.12 billion, up 25% year over year, as semiconductor manufacturers continued spending heavily on equipment needed for AI processors, memory and advanced packaging. Adjusted earnings reached $3.50 per share, while semiconductor systems revenue rose to approximately $7.04 billion. The company forecast another increase in the current quarter and said it intends to substantially expand manufacturing capacity, with management targeting roughly double its semiconductor-system output by 2028.

Applied Materials occupies a crucial position several layers beneath the companies normally associated with AI. Nvidia, AMD and custom-chip developers design processors, but those processors cannot exist without fabrication equipment and materials engineering supplied by companies such as Applied Materials, ASML, Lam Research and KLA. AI has also increased demand for DRAM, high-bandwidth memory, advanced packaging and increasingly complicated transistor structures, requiring chipmakers to invest more heavily in manufacturing tools. Investors nevertheless pushed Applied Materials shares lower following the results, showing how high expectations have become across the semiconductor supply chain. Strong earnings are no longer sufficient when valuations already assume years of extraordinary AI spending. That tension will be one of the central questions facing semiconductor companies if hyperscaler infrastructure budgets remain elevated.

Why It Matters: Applied Materials’ growth shows how AI spending is flowing far beyond GPU designers into the specialized equipment required to manufacture the next generation of chips.

Source: Wall Street Journal.