AI Price Wars Shift to Cost Management, Alibaba Exits Gaming

The AI industry is moving fast, and this week's news shows a clear pattern: companies are getting serious about costs, security, and practical deployment. DeepSeek's new pricing model, Alibaba's decision to sell its gaming division, and OpenAI's broader Codex access are all signs that the hype phase is over. Now it's about building sustainable businesses.

DeepSeek's Time-Based Pricing Reshapes the Cost Game

DeepSeek officially rolled out its time-based pricing on August 17. The new system charges more during peak hours and offers discounts during off-peak times, nudging developers to shift their workloads. This is a big change from the relentless price cuts we saw earlier. When you have millions of calls, GPU queues, and enterprise customers demanding stability, you can't just compete on being the cheapest.

The pricing shift means developers and companies need to rethink how they schedule batch jobs, caching, and retries. It's no longer just about the sticker price; it's about when you run your model. This move signals that Chinese AI model providers are moving from customer acquisition to capacity management. The real cost of serving AI is finally coming to the surface.

Alibaba Sells Gaming Arm to Double Down on AI

Alibaba is reportedly selling its gaming business, Lingxi Interactive Entertainment, to private equity firm Trustar Capital for at least $1.5 billion. In an internal letter, Lingxi's CEO said the move lets Alibaba focus on its strategic priorities. Gaming was a solid cash cow for Alibaba, but the company has been pouring resources into cloud, AI models, enterprise services, and smart e-commerce.

This isn't Alibaba abandoning content; it's a strategic reallocation. The money, management attention, and talent freed up from gaming will likely flow into AI infrastructure and commercialization. It's a clear sign that Alibaba sees AI as its future, even if it means giving up a profitable side business.

Codex Opens Million-Token Context to All ChatGPT Users

OpenAI's Tibo announced that Codex's GPT-5.6 Sol model now supports a 1 million token context window for ChatGPT account users, not just API key holders. This is huge for developers. You can now dump an entire codebase or a massive document into the context and have the model work with it without constantly trimming and re-pasting.

The move lowers the barrier for non-API developers to use large-context models in their daily coding, code review, and refactoring. AI coding tools are no longer just about speed; they're about context capacity and how easily you can access that power. By pushing this capability into the account tier, OpenAI is making it more accessible to a wider audience.

Xiaomi's AI Hiring Surge Signals Hardware's Shift

Xiaomi's AI-related job postings have jumped by over 50%. That's a clear indicator that hardware companies are scrambling to build AI capabilities into their products. From voice assistants and on-device models to smart driving and home automation, Xiaomi needs engineers, data scientists, and product managers who can weave AI into the user experience.

Hardware firms used to compete on supply chain and price. Now they need to integrate AI into their system-level experience. Xiaomi's hiring spree shows that AI isn't just a feature; it's a structural investment that will shape the next generation of phones, cars, and IoT devices.

OpenAI Disbands Preparedness Team, Raising Safety Questions

Reports say OpenAI disbanded its Preparedness team, which was responsible for assessing severe risks from frontier models and developing mitigations. This is a red flag for many observers. As AI models get more powerful and start interacting with tools and executing actions, safety evaluation can't just be a pre-launch checklist. It needs to cover unauthorized operations, deceptive outputs, cyber risks, and abuse prevention.

The team's dissolution could be a reorganization, but it raises concerns about the independence of risk governance. For OpenAI, the more capable its models, the more it needs to convince enterprises, regulators, and the public that safety isn't an afterthought.

Anthropic CEO Calls AI Backlash a Trust Crisis

Anthropic's CEO, Dario Amodei, responded to criticism about his AI risk warnings, saying the backlash is fundamentally a trust crisis. He's pushing the conversation beyond pessimism versus optimism to whether people believe AI companies are honest about limitations and deployment consequences. As Claude gets embedded in coding, office work, and enterprise workflows, the demand is shifting from promises to verifiable mechanisms.

Trust isn't about a single statement; it's about consistent behavior when models fail, jobs change, content is mislabeled, or privacy is at stake. AI companies need to establish stable rules around these issues to earn and keep trust.

ChatGPT's Computer History Feature Learns Your Workflow

The ChatGPT macOS app now includes a Computer History feature that records your clicks, keystrokes, and work patterns. It's designed to learn how you work, suggest automations, and help you pick up where you left off. This moves AI assistants from chat windows into the operating system level, aiming to understand your actual workflow across multiple apps.

The potential is huge: fewer repetitive tasks, better task memory, and proactive suggestions. But it also raises serious privacy concerns. Click and key data is highly sensitive. OpenAI will need to be transparent about how this data is used, stored, and controlled. Users need clear opt-in, visibility, and the ability to turn it off.

Stripe to Acquire OpenRouter for Over $7 Billion?

Stripe is reportedly in talks to acquire OpenRouter, an AI gateway startup, for more than $7 billion. OpenRouter, which describes itself as the "Stripe of AI," helps developers route requests to different models based on price, latency, and capability. If Stripe buys it, it's a bet that AI applications need not just models but also billing, risk management, settlement, and model-switching infrastructure.

This would merge Stripe's payment network with AI call routing, making the "router" and the "cash register" one business. It's a sign that AI's commercial infrastructure is maturing, and whoever controls the gateway could capture significant value.

Microsoft's New AI Evaluation Framework: Human Capital vs. Token Capital

Microsoft CEO Satya Nadella proposed a new way to evaluate AI investments, framing them in terms of "human capital" and "token capital." This shifts the conversation from server costs and software subscriptions to productivity accounting. Organizations need to decide which tasks are done by employees and which by model tokens, and how they combine for better output.

This narrative helps justify Microsoft's Copilot and Azure AI spending. But for customers, the real challenge is putting model usage, employee time, process quality, and business outcomes on the same ledger. It's not just about buying AI; it's about measuring its impact.

Meta's AI Vision Under Fire

Meta's AI ambitions are meeting skepticism. While Zuckerberg pushes open-source models, AI assistants, and smart glasses, critics question whether massive capital spending translates into user growth, ad efficiency, or new hardware wins. The issue isn't lack of investment; it's that the superintelligence narrative feels distant from current revenue and user experience.

Meta needs to show tangible product changes: better creator tools, smarter ad targeting, improved content distribution, and wearable devices that people actually want. AI vision needs to be grounded in measurable results, not just long-term promises.

Intel Arc Pro B65: A New Option for Local AI Workstations

Intel's Arc Pro B65 TF 32G professional GPU is now available for 8,999 yuan. It packs 32GB of VRAM, 20 Xe cores, a 2400MHz clock, and 197 TOPS of INT8 performance, all within a 200W power envelope. This isn't a gaming card; it's aimed at local inference, professional graphics, and small workstation setups.

As more companies and creators want to run models on-premises, they need alternatives to NVIDIA. The Arc Pro B65 offers a compelling mix of memory, performance, and price for certain workloads, though driver maturity and ecosystem support remain open questions.

Tencent Doubles Down on AI While Gaming Revenue Hits $65B

Tencent's gaming revenue for Q2 exceeded 65 billion yuan, with domestic growth up 17% year-over-year, though overseas growth slowed. Gaming remains a cash cow that funds Tencent's AI bets. The company is investing heavily in AI across cloud, ads, gaming, enterprise collaboration, and WeChat, but executives like Martin Lau admit they're still waiting for the "explosive" moment when AI clearly boosts revenue and efficiency.

Tencent's strategy is multi-pronged: model capabilities, scenario density, and commercial loops. The breakthrough might come in ads, office tools, or social features, but it's not here yet.

DeepSeek Harness: Self-Evolving Agent Toolchain

Details emerged about DeepSeek Harness's "self-evolution" blueprint, which aims to make AI development as modular as building with LEGO. This is part of the broader trend in agent toolchains: models are just the base, and the real value is in task orchestration, tool calling, evaluation, plugins, and continuous improvement.

If DeepSeek can standardize the development, testing, deployment, and self-improvement loop, it could offer more than just an API—it would deliver engineering capability. But as agents become more autonomous, developers will demand robust frameworks, clear permission boundaries, and observability.

AI Power Demand Keeps Rising: Bloom Energy Ups Forecast

Bloom Energy raised its full-year electricity demand forecast, citing AI infrastructure needs. Data centers are gobbling up power, and that's pushing energy companies, fuel cells, and storage solutions into the spotlight. AI isn't just about GPUs; it's about reliable, scalable power.

Cloud providers and model companies now face two big costs: chip depreciation and long-term electricity supply. The energy bill for AI is going to keep climbing.

Amazon Updates Terms: Mandatory Arbitration and Class Action Waiver

Amazon has updated its terms of service, introducing mandatory arbitration and a class action waiver. This is a classic platform move to limit legal exposure. With millions of users, disputes are inevitable, and arbitration can reduce the burden of large-scale lawsuits.

But it also raises questions about user rights and access to justice. For internet companies, terms of service are more than legal text; they signal how the platform allocates risk, cost, and bargaining power.

Huawei's Smart Driving and Cockpit Tech Expands to SUVs

Deepal's new G318 SUV is set to feature Huawei's Qiankun ADS 5 Pro and HarmonySpace 5 cockpit system. This shows how Huawei's intelligent driving and cockpit solutions are moving into more mainstream vehicles. The competition is no longer about having basic driver assistance; it's about making high-level autonomous driving and smart cabin experiences standard.

As Huawei partners with more automakers, those companies must rethink their hardware design, software updates, and sales pitches around intelligence.

Huawei Leads Global Wearables, Smartwatches Outperform

In Q2 2026, global wearable band shipments fell 2%, but smartwatch shipments grew 6%. Huawei led the overall wearable market, while Apple took nearly half of the smartwatch segment. Basic fitness bands are declining, indicating a shift toward health monitoring and premium features.

Huawei's strength in high-end sports watches is paying off. The industry's future lies in better sensors, health algorithms, and keeping users engaged over time.

China Issues First Auto Chip Certification Standards

China's market regulator approved five certification standards for automotive chips, creating the world's first national evaluation system for the entire auto chip supply chain. Auto chips are critical for smart cockpits, ADAS, battery management, and security, and they require much higher reliability than consumer electronics.

A unified certification system could lower barriers for domestic chips and help automakers and suppliers build trust. As EVs and smart cars grow, chips are not just about supply chain; they're about safety, compliance, and national competitiveness.

Tencent Cloud Partners with CPE Source Peak

Tencent Cloud and asset manager CPE Source Peak announced a digital ecosystem partnership. This goes beyond cloud procurement, touching data governance, risk management, investment research efficiency, and AI tool deployment. Tencent Cloud is packaging infrastructure, databases, security, and model services into industry-specific solutions.

For financial firms, such partnerships can improve internal operations and potentially connect portfolio companies to Tencent's ecosystem. Cloud competition is moving from generic compute to deep industry expertise.

Douyin Sets Up Tech Company in Kunming with $700M Capital

Douyin (TikTok China) founded a tech company in Kunming with a registered capital of 700 million yuan. This signals a push into regional operations, local services, and e-commerce. With short video traffic as a base, Douyin is expanding into local life, delivery, and merchant tools, and regional entities help it get closer to local industries.

The high capital suggests this isn't a shell company; it's a serious commitment to expanding its business network in lower-tier cities.

JD.com's Delivery-and-Installation Service Covers Millions of Homes

JD.com's "delivery and installation" service now covers multiple countries and has served over 10 million households. For large appliances and complex electronics, the challenge isn't just shipping; it's scheduling, installation, debugging, and after-sales support. By integrating these, JD builds a service moat that pure e-commerce players struggle to match.

This capability is especially valuable in high-ticket, service-heavy categories, where logistics networks, technician teams, and after-sales processes directly impact repeat purchases and brand trust.

TechCrunch Eyes Electric Air Taxi Routes

TechCrunch's mobility newsletter this week focused on electric air taxi route changes, highlighting the ongoing adjustments in eVTOL commercialization. Air taxis need to solve certification, landing sites, noise, safety, pricing, and energy supply. Any route change reflects a rebalancing of business models and regulations.

The industry is still far from high-frequency, affordable service. Low-altitude economy hype is real, but safety and operating costs will keep progress in check.

Valve's European Logistics Partner Hacked, Customer Data at Risk

Valve's European logistics partner suffered a cyberattack, exposing potential customer data related to Steam hardware. This is a reminder that hardware sales involve third-party logistics, warehousing, and delivery systems, each a potential weak link. While payment details may not be exposed, names, addresses, and orders can fuel phishing scams.

Platforms need to audit and secure their entire supply chain, not just their own systems. Third-party risk is a growing concern in the digital age.