Editor's note
The week's pattern is consistent: AI is moving from product to infrastructure. Nvidia's record quarter, DeepSeek's near-close funding round, and Anthropic's hardware standard are all signals of underlying buildout — compute, capital, and physical integration. Meanwhile, the shutdown of Amazon's Mechanical Turk is a quiet but telling milestone: the human-labor platform that once powered AI training is being retired by the industry it helped create. For web and marketing teams, the most time-sensitive item is tomorrow's Google AI Max auto-migration — there is still time to act before September 1.
AIOpenAI Help Center
OpenAI Retires the Official DALL-E GPT on August 30 — Image Generation Consolidated Into ChatGPT Images
OpenAI retired its official DALL-E GPT from ChatGPT on August 30. One of the original custom GPTs maintained directly by OpenAI since their late-2023 launch, it has been replaced by ChatGPT Images, which runs on the newer gpt-image-1 and gpt-image-1-mini models. ChatGPT Images is available to all plans including free. Users were advised to download stored images before the cutoff. User-created GPTs with image generation are unaffected.
Context
The DALL-E GPT was among the first custom GPTs that OpenAI itself published and maintained after introducing the custom GPT feature in 2023. Its retirement is part of a product-consolidation move: OpenAI is integrating image generation into the main ChatGPT interface rather than routing it through a standalone GPT. ChatGPT Images, which replaces it, runs on more capable updated models and offers broadly equivalent functionality.
HaLVision's take
For teams with workflows built around specific OpenAI-maintained GPTs, now is a good time to audit dependencies. GPTs published by OpenAI can be retired with relatively short notice. The underlying image generation capability is unchanged — the practical task is updating any internal links or documentation that pointed to the DALL-E GPT, directing users to ChatGPT Images instead.
WebSearch Engine Land
Google Auto-Upgrades Search Campaigns to AI Max Starting September 1 — Manual Opt-Out Required
Starting September 1, Google will automatically upgrade Search campaigns using campaign-level broad match or Automatically Created Assets (ACA) to AI Max for Search. There is no passive opt-out: advertisers must manually disable the relevant features before September 1 to avoid migration. Dynamic Search Ads migration to AI Max remains delayed to February 2027.
Context
AI Max for Search is Google's consolidated automated optimization layer, combining automated ad copy, destination selection, and query expansion under a single setting. Advertisers already using standalone broad match or ACA were effectively using components of AI Max; Google is now formally unifying those settings into the AI Max framework. For advertisers who want to retain manual controls over targeting and assets, disabling affected features before September 1 is the only way to avoid the upgrade. The change is not reversible in a straightforward way once applied.
HaLVision's take
If you manage Google Ads accounts for clients, today is the day to check affected campaigns and decide whether AI Max aligns with each campaign strategy. Clients with strict targeting controls, brand safety requirements, or carefully maintained exclusion lists are the priority review — AI Max expands query matching and generates assets in ways that can conflict with manual settings. Act before the deadline closes tomorrow.
AIAnthropic
Anthropic Opens Research Preview of the Model Hardware Standard — a Shared Spec for AI Agents to Operate Physical Devices
Anthropic opened a research preview of the Model Hardware Standard (MHS) on August 27, an open specification enabling AI agents to safely discover, operate, and troubleshoot physical equipment such as microscopes, robotic arms, and liquid handlers. MHS is model-agnostic and works alongside the existing Model Context Protocol. Partners include AWS, Universal Robots, and Hugging Face. Open-source release is planned after safety evaluation.
Context
Where MCP established a standard for connecting AI agents to external software services, MHS extends that approach to physical hardware. Originally developed with HHMI Janelia Research Campus, it standardizes device characteristics — weight limits, safety parameters, adjustable settings — that previously existed only as tacit expertise or paper manuals. The research preview is limited to scientific labs and advanced manufacturers, who collaborate on safety benchmarks and best practices for AI systems operating physical equipment before the standard is made publicly available.
HaLVision's take
Web and software studios have no immediate workflow action here. The forward-looking relevance: if you serve manufacturing, biotech, or research clients, AI-to-hardware integration is moving from concept to planning-horizon. Familiarity with MHS alongside MCP positions you to advise on agent architectures as physical-AI projects begin appearing in client briefs.
AIChina Money Network
DeepSeek Nears $7.4B Funding Round at $74B Valuation, Eyes 2027 IPO on Shanghai's Star Market
Multiple reports indicate Chinese AI startup DeepSeek is close to closing a funding round of approximately $7.4 billion at a $74 billion pre-money valuation, with signing expected before the end of August. Investors include JD.com, NetEase, IDG Capital, and China's national AI fund. Proceeds target R&D and large-scale compute infrastructure expansion; an IPO on Shanghai's Star Market is targeted for 2027.
Context
DeepSeek gained international attention in early 2025 when its R1 model matched leading Western benchmarks at a fraction of the compute cost. The funding round was paused in July after founder Liang Wenfeng's frustration over leaked remarks, then restarted. At $74 billion, the valuation places DeepSeek among the most valuable AI companies globally and reflects sustained institutional interest in Chinese AI infrastructure capacity. A 2027 IPO timeline remains subject to regulatory conditions and US-China tech friction.
HaLVision's take
No near-term pricing changes for SMBs using AI APIs. The strategic read is that DeepSeek is building toward a public market, which typically brings more API offerings, pricing transparency, and commercial-tier services. If your stack includes or could benefit from cost-efficient inference, DeepSeek's expanding service catalog is worth monitoring as they approach IPO.
AINVIDIA Newsroom
Nvidia Reports Record $96.2B Q2 Revenue, Up 106% Year-Over-Year — Data Center Delivers $89B
Nvidia reported Q2 FY2027 (quarter ended July 26, 2026) revenue of $96.2 billion, up 106% year-over-year and 18% quarter-over-quarter, well above the $92 billion consensus estimate. Data center revenue reached $89 billion, up 117% annually, driven by continued hyperscaler and enterprise capital expenditure on AI compute. Q3 guidance is $108 billion, plus or minus 2%, again above analyst expectations.
Context
The results confirm that enterprise and hyperscaler GPU spending remains in an accelerating phase. Separately, Nvidia was reported to have quietly paused cloud-financing deals it had offered to smaller AI data-center operators, suggesting tighter capital allocation even amid record growth. The year-over-year revenue doubling underscores the scale of concentrated capital flowing into AI infrastructure — primarily through cloud-provider GPU purchases.
HaLVision's take
Nvidia's revenue growth is a leading indicator of AI inference costs: hyperscaler GPU capex flows through to the price of compute that AI APIs run on. For SMBs, near-term API pricing is unlikely to fall sharply while hardware demand stays at this pace. Monitoring multi-vendor API pricing and locking favorable tier agreements when available is sound ongoing practice.
WebCNBC
Amazon Mechanical Turk to Close on September 30 After 21 Years
Amazon announced on August 25 that it will shut down Mechanical Turk (MTurk), its crowdsourced data-labeling marketplace, on September 30, 2026. Launched in 2005 and once serving over 500,000 workers, the platform had been in steady decline. SageMaker Ground Truth and Amazon Augmented AI close alongside it. New customer enrollment stopped July 30; existing users have until September 30 to export data and migrate workflows.
Context
Mechanical Turk was the canonical model for using human crowd labor on tasks AI could not handle reliably — Jeff Bezos called it 'artificial artificial intelligence.' Its decline reflects two concurrent shifts: specialized data-labeling platforms such as Scale AI, Prolific, and Mercor offering higher quality and accountability, and AI models that can now generate, annotate, and synthesize their own training data at scale. The shutdown signals that the human-crowdsourcing layer of AI development is being replaced by automated pipelines.
HaLVision's take
Organizations relying on MTurk for annotation, UX research, or content classification tasks should prioritize data export and identify an alternative before the September 30 deadline. For studios advising on AI workflows, the practical note is that data-labeling costs continue to fall as automation improves — budget assumptions built on MTurk pricing should be updated with current alternatives.