
Credits: European Commission – Photographer / Wikimedia Commons — CC BY 4.0.
On October 6, 2025, at its developer conference in the United States, Sam Altman’s OpenAI unveiled AgentKit and opened ChatGPT to third-party ChatGPT apps, while a strategic deal with AMD promises up to 6 GW of MI450 GPUs starting in 2026. The stated goal: democratizing AI agents and securing supply in the face of Nvidia are goals of the Stargate project. This project is backed by Washington and partners such as Samsung.
AgentKit, When the Tool Becomes the Story
In the shadow of San Francisco’s glass walls, OpenAI lifted the veil on October 6, 2025 on AgentKit, a toolkit designed for anyone who wants to build artificial intelligence agents from the first sketch all the way to production use. The announcement reads like a manifesto: bringing developers closer to ChatGPT-like apps and democratizing agent creation. In addition, it aims to reduce friction and provide a guided path from prototype to deployment.“We are pursuing a long-term vision where AI will transform the global economy,” Sam Altman, OpenAI’s CEO, says. The narrative shift is clear: moving from hands-on experimentation to controlled industrialization.
In practical terms, AgentKit assembles ready-to-use modules. Agent Builder provides a drag-and-drop interface that organizes workflows, versions paths, and exposes guardrails. ChatKit makes it possible to embed agent-powered conversations in an app or website without reinventing the interface or session management. Evals for Agents brings datasets, tracing, and systematic scoring to measure how well these agents perform in real-world conditions. This end-to-end toolkit offers a bridge between research and product, where scattered scripts and makeshift hacks used to pile up.
The philosophy is visible between the lines: offer simple interoperability. OpenAI touts standardized API connectors and a central registry to let the agent talk to the services of a sprawling ecosystem. The connectors are based on a common protocol for ChatGPT apps on macOS, iOS, and the web. That way, everyone can build their app while respecting security and privacy rules. The ambition is twofold: broaden the developer base and ensure stable service quality. In a world that is becoming agentified, the kit plays the role of a common grammar.
ChatGPT Apps That Talk Inside ChatGPT
Another piece of the puzzle: the arrival of ChatGPT apps. The interface no longer just produces text. On Mac, the ChatGPT apps interface no longer just produces text. It summons, right in the middle of a conversation, services that can act and display. Spotify, Canva, Booking.com, Expedia, Figma, and Zillow appear as companions you call by name. On iOS, ask for a Friday-night playlist or a presentation mockup. The app then appears in the conversation with its cards and thumbnails. In addition, it offers interactive controls to make your user experience easier. You can also look for a hotel near Buttes-Chaumont …appear as companions, optimized even for MacBook Air, right inside ChatGPT apps.
OpenAI is releasing an Apps SDK and making openness a guiding principle. The connectors rely on a common protocol so everyone can build their app while following security rules. The goal is not only to bring together disparate building blocks but to make them speak the same language. This needs to happen at the right time to ensure efficiency and interoperability. ChatGPT is thus becoming a software platform where everyone is looking for the best ChatGPT app for their use case… On Windows too, the ChatGPT apps interface stops being a simple text box and becomes a stage where cards, forms, readers, and workshops parade by.
A Power Pact: 6 GW of AMD GPUs Starting in 2026
Technology poetry does not hold without energy. At the same time, OpenAI and AMD are signing an unprecedented partnership. The deal calls for up to 6 GW of installed capacity based on Instinct MI450 chips, with an initial one-gigawatt tranche in the second half of 2026. The goal is to anchor OpenAI’s infrastructure on a second source of large-scale compute. This is necessary in the face of Nvidia’s dominance.
Several media outlets put the contract at tens of billions of dollars. It includes warrants that could allow OpenAI to hold up to 160 million AMD shares. This is possible if technical and commercial milestones are reached. The numbers are circulating and the assumptions vary. However, it should be remembered that they depend on an industrial timeline. In addition, they rely on the delivery of successive batches and a chain of approvals.
On Wall Street, the reading is immediate: AMD gains a showcase and a multi-year revenue stream. OpenAI buys supply sovereignty and a financial lever to support scaling up. Beneath the waterline, one quiet issue looms large: the operating cost of generative AI at a global scale, where the kilowatt-hour becomes the new unit of storytelling.
Stargate, Geography of an American Dream
This material pact fits into a broader fresco called Stargate. The project is backed by Washington and led by a صنعتی coalition. It aims to build giant data centers in the United States. In addition, it seeks to build bridges with Asia. The White House presents itself as a catalyst for massive private investment. Donald Trump, President of the United States, makes it the banner of a strategy for technological leadership and jobs.
On the Asian side, Samsung and SK join the dance. The message is clear: the memory ecosystem is getting in step with demand and securing HBM supplies. At this scale, the supply chain becomes drama. Each wafer, each slab of silicon tells part of the story. Stargate sketches a map where Texas, New Mexico, Ohio, and a site in the Midwest stand alongside Seoul in the same story of steel and electricity.
A Valuation That Makes Your Head Spin
OpenAI reaches a $500 billion valuation according to several equity transactions carried out in the fall. The symbolic threshold cements the company as the world’s most valuable startup. The figure is striking, but editorial caution is in order. These amounts reflect private transactions, revenue expectations, and adoption curves that are still moving. Valuation is not a prophecy. It is a collective bet on the texture of the future.
In this harsh light, the deal with AMD becomes more than a contract line. It turns into a clear capex. It converts enthusiasm into machines, and a flood of promises into tangible megawatts.
Sam Altman, The Cautious Architect
The saga could have been nothing but a triumphant march. Sam Altman, a discreet figure with a very steady gaze, injects a note of realism. He acknowledges the speculative intensity of the moment, warns of the risk of a bubble, and reaffirms a patient vision: “the transformation will be lasting, provided talent, energy, and rigor are aligned.” His role is to hold together the idealism of a beneficial AI and the hard reality of a column of numbers.
Altman does not wear the cape of a demigod. He plays more the part of the organizer of a project where researchers, engineers, regulators, and artists must find a common measure. AgentKit then looks as much like a set of requirements as it does a piece of software. It should be read as a form of discipline: evaluate, trace, correct. A factory for agents that does not sacrifice rigor on the altar of speed.
Bezos, The Chorus of the Moderns
On the neighboring stage, Jeff Bezos comments on the wave. He speaks of a useful bubble, like a rising foam. However, that should not mask the underlying swell. He sees AI as a force with the same momentum as the great shifts in digital technology. He is already imagining data centers powered by space solar energy in ten or twenty years. His voice is not an oracle. Still, it speaks to the confidence of a builder seasoned in long-term investment.
This outside view sheds light on the moment. It underscores a paradox: too much money can fund both the best and the unnecessary. Responsibility means sorting, concentrating capital on uses that solve real problems and respect the environment. This requirement echoes Altman’s and, in a strange counterpoint, the restraint of some investors who fear the frenzy.
Tool Culture and the Daily Shift
What does AgentKit change in everyday life? Developers get a grammar that avoids rewriting the same functions. Companies gain a foundation for experimenting without excessive risk. Creators can use agents to dig through corpora and summarize information. In addition, they can run monitoring, put together a mockup, or plan an itinerary. At every step, built-in evaluation promises to separate the useful from the gimmick.
With ChatGPT apps integrated into ChatGPT, the boundary between service and conversation disappears. The plane ticket is bought in the same sentence as the reading recommendation. The playlist is born from a prompt, then refined over a few exchanges. The interface stops being a simple text box and becomes a stage where cards, forms, readers, and workshops parade by.
Energy, Climate, Sovereignty: The Other Side of the Picture
The fresco has its grave note. Six gigawatts promised with AMD is an admission of an unprecedented appetite for electricity. The environmental management of such facilities cannot be a marginal afterthought. Water is needed for cooling, decarbonized power contracts, and optimization engineering that tracks waste at every software and hardware layer. OpenAI’s promise is measured as much by its models as by its emissions curves.
Technological sovereignty carries just as much weight. Multiplying GPU suppliers and spreading out data centers responds to geopolitical logic. In fact, working with Samsung in Asia while consolidating American sites is strategic. In this context, the federal government, with Donald Trump at the helm, promotes an industrial narrative. In fact, jobs and defense sit alongside innovation in this vision.
A Silicon Valley Epic, Without Emphasis or Self-Indulgence
It would be tempting to elevate OpenAI to the rank of instant myth. The magazine takes a different line. On one side, an old-school success story: a lab turned global platform, a brand that moved from researchers’ jargon into everyday language. On the other, a head-spinning picture: colossal sums, a silicon battle, and a democratic debate about the use and control of these tools.
The beauty of AgentKit and the ChatGPT apps that live within ChatGPT lies in their simplicity. Their power rests on a more prosaic alliance: energy, supply chain, governance, and ethics. The deal with AMD and the scale of Stargate bring these dimensions together in one move. If the story holds, it will tell how AI left the lab. In fact, it is reshaping the material reality of our lives.