On September 3, 2026, OpenAI launched GPT-6 Astra, presented as "the world's most intelligent and aligned model." The news made the rounds of every feed within 24 hours, like every launch for the past three years. But focusing on the model is looking at the finger. What deserves your attention is what changed underneath: the way these systems actually work now, and the industrial scale of what's being built to run them. Together, they sketch the market your business will operate in for the next decade. Here is the state of play, in three layers: how it works, the infrastructure, the market.
Layer 1: AIs no longer answer, they act
The mental image most people have of AI dates from 2023: a chat box that produces text, sometimes brilliant, sometimes invented. That image is three years out of date. The current generation, with Astra at the top, works differently on three levels:
It reasons in steps. Today's models break a problem down, plan, check their own results and retry when needed. That's what lets them handle complete workflows, producing a finished analysis, spreadsheet or presentation rather than fragments of text.
It uses a computer the way you do. The hinge capability of this generation is called computer use: the model sees a screen, clicks, types, navigates, fills out forms. Astra is roughly twice as fast as the previous generation on this front. OpenAI had already shut down its Atlas browser in August to fold those capabilities directly into ChatGPT, a story we covered in a dedicated article: the browsing agent is no longer a separate product, it's a feature of the tool hundreds of millions of people already use.
It keeps the thread. Persistent memory, context that carries from one session to the next, notes the model takes and consults on its own: the amnesiac assistant becomes a collaborator who knows the file.
The consequence of these three properties fits in one word: agents. Systems you hand an objective, shop this for me, compare these suppliers, prepare this report, and that execute the steps alone. That's the change in kind: AI moves from a tool you consult to an actor that transacts.
Layer 2: the infrastructure of an industrial revolution, not a fad
If you doubt how serious this transition is, follow the money, not the demos.
In 2026, the five cloud giants, Microsoft, Alphabet, Amazon, Meta and Oracle, have committed between US$660 and 690 billion in capital expenditures, nearly double 2025. Amazon leads with about $200 billion; Google follows between $175 and 185 billion. Analysts estimate roughly three quarters of those sums go directly to AI infrastructure: data centers, chips, energy.
Add the Stargate project, the OpenAI, SoftBank and Oracle consortium committing $500 billion over four years for roughly 7 gigawatts of capacity in the United States, and the sector total passes one trillion dollars for 2026 alone.
For scale: this is one of the largest private capital deployments in history, comparable to the great infrastructure waves of the last century, railroads, electrification, telecoms. Nobody spends a trillion dollars on a chat feature. They spend it on a new layer of the economy.
Two practical readings of those numbers:
- Intelligence becomes a utility. Like electricity, it's metered (in tokens), billed by usage (Astra's API prices are public: about US$10 per million input tokens, $50 per million output) and its unit cost falls as capacity gets built. What used to cost a team now costs a subscription.
- Power concentrates. A handful of companies owns the power plants. Everything your business builds with AI will rest on someone else's infrastructure, with their prices, their terms and their outages. That has a direct strategic consequence, coming up.
Layer 3: what it means for the market
Cross the two layers, agents that act, infrastructure that makes using them cheap and ordinary, and three economic shifts take shape. They concern every business, including yours.
1. The cost of average cognitive work collapses. Generic writing, summaries, standard analyses, ordinary code: everything "correct but unoriginal" trends toward near-zero marginal cost. It's not that these tasks disappear; it's that they stop being an advantage. Value migrates toward what the machine doesn't supply: first-hand data, lived experience, relationships, accountability. For marketing, we've said it bluntly in our other articles: generic content is now worth nothing, and lived proof becomes the last defensible edge.
2. Part of your customers become machines. When agents shop, compare and fill carts on behalf of their users, your site is no longer just a storefront for humans: it's an interface for machines. Three requirements follow, and they're concrete: complete, readable information (prices, delays, policies, reviews), up-to-date structured data, and a frictionless purchase path, because an agent forgives no broken form. The whole checkout CRO checklist from our dedicated article now serves two audiences at once.
3. Platform dependence rises a notch, and your owned assets gain value. In a market where discovery increasingly flows through AI answers operated by a few players, what you own outright, your brand, your email list, your community, your customer data, becomes your only unrented position. It's the same reasoning that makes email the most underrated channel in ecommerce, and brand identity the shield against automated comparison: when an agent compares spec sheets, only the brand asked for by name escapes the price war.
The action plan, without panic
For a Quebec SMB, the current window looks like the web around 2005: early enough to get ahead, late enough that the tools are mature. In order:
- Test your AI visibility. Ask ChatGPT to recommend your category in your region. Who shows up, and why?
- Make your site machine-readable: structured data, complete information, frictionless checkout.
- Reinvest the generic-content budget into proof: case studies, real numbers, reviews, credible partnerships.
- Equip your team now. Astra ships inside existing ChatGPT plans; the productivity gains on internal work (analyses, drafts, spreadsheets) are available this week, not in five years.
- Strengthen what you own: the list, the brand, the customer relationship. That's the part of the game that depends on no platform.
The honest caveat
Two reservations. First, the infrastructure numbers describe commitments, and the history of great capital waves always includes excess: some of these billions will be badly invested, some promises won't hold, and OpenAI itself shut down Atlas nine months after launch. Judge products on usage, not announcements. Second, this generation's power cuts both ways: Astra is the first model rated "Critical" for cybersecurity capabilities under OpenAI's preparedness framework. Caution about what you connect to these tools, and with what access, is part of any serious strategy.
But the direction is no longer in question. Intelligence is becoming a utility, agents are becoming market actors, and the trillion invested this year guarantees the movement won't reverse. The question for your business isn't "if", it's "which role": a supplier that AIs can read and recommend, or an invisible brand that will never understand where its customers went.
FAQ
What is GPT-6 Astra? The model OpenAI launched on September 3, 2026, presented as a generational leap in professional work, science, software engineering and cybersecurity. It rolls out in ChatGPT (paid plans), through the API and on AWS.
How much are the giants investing in AI infrastructure? Between US$660 and 690 billion in committed 2026 capital expenditures from Microsoft, Alphabet, Amazon, Meta and Oracle, plus the Stargate project ($500 billion over four years from OpenAI, SoftBank and Oracle). The sector total passes one trillion for 2026.
What is an AI agent? An AI system you hand an objective to, and that executes the steps itself: navigating, comparing, filling, producing. The current model generation, able to use a computer like a human, makes these agents usable at scale.
Concretely, what should an SMB do this year? Check how AI recommends it, make its site and purchase path readable and frictionless for agents, shift content budget toward proof (real cases, reviews, data), and strengthen owned assets: email list, brand, community.
Want a clear picture of what this transition changes for your business specifically? Book a call