DALL-E GPT, Gemini Robotics ER 1.6, GitHub Spark, and the AI power bottleneck (2026)

The coverage window runs from Aug. 30 through the morning of Aug. 31, 2026 (Asia/Dhaka). The clearest changes are operational: three developer products reach retirement or shutdown dates, while two industry updates show how AI deployment still depends on trained workforces and physical power.

Quick scan

Development	What changed	Scale or status	Why it matters
OpenAI DALL-E GPT	The official GPT retired from ChatGPT on Aug. 30; ChatGPT Images is the replacement for image creation and editing. 1	User-created GPTs with image generation remain available.	Users who need old outputs should download them; image workflows need a product switch.
Google Gemini Robotics ER 1.6	The `gemini-robotics-er-1.6-preview` endpoint reaches its shutdown date on Aug. 31. 2	Google points API users to ER 2 and ER 2 Streaming preview endpoints.	Robotics integrations face an immediate migration and testing deadline.
GitHub Spark	Spark stopped accepting new users and new apps on Aug. 4, and existing users get through Aug. 31 to export apps. 3	Deployed apps continue running after Spark retires.	Exporting preserves editable code; apps using `llm()` also need a new inference provider.
Caterpillar's AI deployment	Caterpillar is applying mining-automation practices to field support, manufacturing, and software work. 4	About 1.6 million connected assets and more than 16 petabytes of structured data.	The bottleneck is changing jobsite workflows and training people, not only building a model.
SpaceX turbine foundry	Elon Musk said SpaceX is building in-house turbine-blade and vane casting capacity. 5	Musk said the approach could bring natural-gas turbines online up to 18 months sooner.	Faster data-center power would come with added emissions and public-health scrutiny.

DALL-E GPT reaches its retirement date

OpenAI's official DALL-E GPT retired from ChatGPT on Aug. 30. The date came from a July 31 release-notes entry, so the event is a product change taking effect in this window rather than a new announcement made on Aug. 30. 1

OpenAI directs users to ChatGPT Images for creating and editing images. Users who need to keep images made through the DALL-E GPT should download them before the retirement date. GPTs created by users remain available when those GPTs have image generation enabled. 1

The practical checkpoint is simple: check saved prompts, outputs, and team instructions for a dependency on the official GPT. A custom GPT with image generation is a separate case from the retired official GPT, so the owner should test that custom workflow independently.

Gemini Robotics ER 1.6 reaches shutdown

Google's Gemini API changelog says the gemini-robotics-er-1.6-preview model will shut down on Aug. 31. Google lists gemini-robotics-er-2-preview and gemini-robotics-er-2-streaming-preview as the migration targets. 2

The ER 2 preview supports text, image, video, and audio input, along with function calling, spatial reasoning, and multi-step tool orchestration. The streaming preview is intended for lower-latency text streaming through the Live API. 2

For API users, the important change is the endpoint lifecycle. An integration that still names the 1.6 preview needs a replacement model ID and a compatibility test before the shutdown date. The changelog gives the destination; each application still has to verify its own tool calls, input handling, and latency requirements.

GitHub Spark reaches its export deadline

GitHub's Aug. 4 notice set Aug. 31 as the last day for existing Spark users to export their apps. GitHub stopped accepting new Spark users and new app creation on Aug. 4. Deployed apps continue to run after Spark retires, while the export path preserves the code for continued editing in a repository. 3

GitHub says users can open the Spark workbench, choose the ellipsis menu, and select Create repository. Apps that call Spark's llm() function have a second migration task: the function depends on GitHub Models, which retired on July 30. Those apps need their own inference provider, API key, and billing arrangement. Apps that do not use llm() are unaffected by the GitHub Models retirement. 3

The next action differs by app. Exporting is enough for an app whose runtime dependencies are already independent. An app using llm() needs both a code export and a provider change, even though its deployed version can keep running.

Caterpillar treats deployment as a workflow problem

Caterpillar is carrying lessons from autonomous mining equipment into more ordinary worksites and enterprise software. CTO Jaime Mineart told TechCrunch that the company's Cat AI Assistant lets field technicians use voice commands to retrieve repair procedures, troubleshoot faults, and identify parts while standing beside a machine. 4

The assistant draws on data from about 1.6 million connected assets and more than 16 petabytes of structured data, according to Mineart. Caterpillar is also using AI for site scanning, digital twins, legacy-code modernization, software testing, and earlier defect detection. 4

The deployment constraint is organizational. Caterpillar has about 118,000 employees, and Mineart said the company plans to spend $100 million over five years training workers in AI, autonomy, and robotics. The company reported $20.5 billion in second-quarter revenue, while power-generation sales rose 72% to $3.10 billion, helped by data-center demand. 4

The checkpoint for other industrial companies is whether a deployment changes the work around the tool. Caterpillar's example puts operator knowledge, remote supervision, and employee training alongside the assistant itself.

SpaceX targets the turbine-manufacturing bottleneck

Elon Musk said SpaceX is building the ability to cast its own natural-gas turbine blades and vanes. TechCrunch reported that Musk said the move could accelerate turbines coming online by up to 18 months. 5

The manufacturing problem is specific. Turbine blades operate in extreme heat and depend on internal cooling channels, thermal-barrier coatings, and single-crystal casting. TechCrunch reported that only four companies currently have industrial-scale capability for this process, leaving data-center builders exposed to a narrow supplier base. 5

The trade-off arrives with the power. Gas turbines can shorten the wait for electricity beside a data center, while the emissions create regulatory and health questions. A Piedmont Environmental Council-commissioned study using the EPA's COBRA health-impact model estimated that eight turbines at one Virginia facility could affect more than 2.5 million people and produce 3.4 to 6.5 additional premature deaths per year, with estimated annual health damages of $53 million to $99 million. 6

SpaceX's next observable checkpoint is whether the foundry can produce power-plant turbine parts at the claimed speed and scale. The claim describes a proposed manufacturing advantage; it is not yet evidence that the foundry has delivered that capacity.

What to watch

Migration deadlines: DALL-E GPT users, Gemini Robotics API users, and Spark app owners each have a different replacement or export action clustered around Aug. 30-31.
Runtime dependencies: Spark's deployed apps may continue running, but llm() users still need to replace the GitHub Models dependency.
Physical constraints: Caterpillar's training plan and SpaceX's turbine foundry both point to work outside the model as the next test: people must adapt workflows, and infrastructure suppliers must manufacture enough power equipment.