AI News – February 15: Amazon Targets 50 Billion, GPT-5.3 Codes by Itself and the "Surgery Case"
A week of strong contrasts (February 9-15, 2026). While Amazon shakes the market with a potential $50 billion investment in OpenAI and the immensely powerful GP
If last week we focused on Google's targeted acquisitions, this week (February 9-15, 2026) enters history for the scale of investment and the depth of the ethical questions raised. While Amazon prepares a $50 billion check for OpenAI (shaking the exclusive alliance with Microsoft), labs are releasing models like GPT-5.3 Codex and Claude Opus 4.6, capable of long-term reasoning that makes code automation an operational reality. However, a Reuters investigation casts a disturbing shadow on the use of AI in medicine, documenting surgical errors due to "visual hallucinations" by machines.
Here is the reasoned chronicle of a week where computing power challenges the limits of physics and human responsibility.
1. The Megadeal: Amazon, OpenAI, and the 50 Billion
The financial news of the year may have arrived in mid-February. Microsoft's hegemony over OpenAI is no longer exclusive.
🔍 What happened:
- Increasingly solid rumors confirm that Amazon is in advanced negotiations to invest up to $50 billion in OpenAI.
- The goal is not to acquire the company, but to tie frontier models (GPT-5 and beyond) to the AWS infrastructure. This would enhance enterprise services like Bedrock and automated contact centers.
- The economic context: As reported by ReadAboutAI (readaboutai.com), OpenAI and Anthropic are burning through cash at unsustainable rates for training. Break-even is estimated no earlier than 2030. OpenAI needs cash and, above all, chips; Amazon needs the world's best model to not lose the cloud war.
💡 Why it matters: This deal establishes that AI is an "Infrastructure War." No company, not even OpenAI, can survive without a partner that owns data centers. For user companies, this could mean native integration of GPT-5 into AWS, simplifying corporate adoption.
🎯 Our take: Diversification is Sam Altman's new strategy. Being tied to a single supplier (Microsoft) was a risk; having Amazon and Microsoft as "bankers" guarantees survival until AGI.
Sources: ReadAboutAI Also read: AI and CRM: Complete Guide for Effective Sales
2. The Code War: GPT-5.3 Codex and Claude Opus 4.6
While managers look at billions, developers look at tokens. It was the week of "Long-Horizon Reasoning."
🔍 What happened:
- Within hours of each other, GPT-5.3 Codex and Claude Opus 4.6 were released.
- GPT-5.3 Codex: Unifies the programming stack with the generalist model. Supports inputs up to 400k tokens (entire code repositories) and outputs of 128k. It is 25% faster, making real-time refactoring possible.
- Claude Opus 4.6: Anthropic is betting everything on agents. The model doesn't just write code, but "reasons" about complex architectures, planning changes that impact dozens of files simultaneously without losing context.
💡 Why it matters: We are moving from "Copilots" (which suggest a line) to "Architects" (which design modules). For software houses, this means halved development cycles, but requires much higher human architectural supervision to avoid structural bugs.
🎯 Our take: The ability to handle long contexts (400k tokens) eliminates the AI's "short memory" problem. Now the algorithm sees the entire project, not just the function it's writing.
Sources: StemGeeks, ReadAboutAI Also read: Generative Artificial Intelligence and Creativity: Tool or Threat?
3. OpenClaw: Open Source Takes Back the Stage
It's not just proprietary software. The open-source world responds with tools that "do things," not just "say things."
🔍 What happened:
- The OpenClaw framework (evolution of Moltbot) has surpassed 150,000 stars on GitHub, becoming the global standard for building local autonomous agents.
- Unlike ChatGPT which lives in the browser, OpenClaw connects to enterprise APIs (CRM, ERP, Finance) and acts: moves money, closes tickets, updates databases.
- It supports on-chain (blockchain) integrations and Asian messaging clients like Feishu/Lark, demonstrating a global versatility that closed models struggle to match.
💡 Why it matters: It marks the transition from experimentation to operational production. Companies are beginning to trust open-source agents running on their own servers (on-premise), ensuring privacy and control that OpenAI's APIs cannot fully offer.
🎯 Our take: Automation is no longer the exclusive domain of Big Tech. OpenClaw democratizes the creation of customized "digital employees."
Sources: AI Forum UK, StemGeeks Also read: AI Sales Automation: Intelligent CRM and Tools
4. The Dark Side: Errors in the Operating Room
The week's investigation brings us abruptly back down to earth. When AI makes a mistake on a spreadsheet, it's corrected; when it makes a mistake on a patient, it's another story.
🔍 What happened:
- An in-depth investigation by Reuters (reuters.com) documented real cases of robot-assisted surgeries where AI made identification errors.
- In some cases, computer vision systems erroneously segmented organs or identified critical body structures as "tissue to be removed."
- The investigation highlights a dangerous gap between vendor marketing ("zero-error surgery") and clinical reality, where AI can suffer from visual biases or hallucinations in non-standard anatomical contexts.
💡 Why it matters: It raises the issue of responsibility. If the robot suggests the wrong cut and the surgeon trusts it (Automation Bias), who is to blame? The doctor, the algorithm, or the hospital that bought the technology?
🎯 Our take: AI in medicine must remain a decision support system (DSS), not an autonomous decision-maker. Human supervision is not optional; it's a lifesaver.
Sources: Reuters Investigation Also read: Who Judges the Algorithm? Ethics and Responsibility in AI Decisions
5. Geopolitics and Energy: The "Data Center Compact"
AI is not ethereal; it's made of iron, silicon, and electricity. A lot of electricity.
🔍 What happened:
- The Trump administration is working on a new "AI Data Center Compact". The goal is to strike deals with Big Tech to ensure data center expansion does not collapse the national power grid and respects physical and cybersecurity standards.
- In parallel, Fortune (fortune.com) describes this moment as the "February 2020" of AI: we are on the threshold of an exponential change that most people perceive but do not yet fully understand, similar to the days before the global lockdown.
💡 Why it matters: Energy is the real bottleneck of AI (more than chips). Without new energy sources (SMR nuclear, renewables), training GPT-6 could be physically impossible. Politics is stepping in to regulate the critical infrastructure of the 21st century.
🎯 Our take: Digital sovereignty goes through energy sovereignty. You cannot have a national AI if you don't have the energy to power it.
Sources: Reuters Policy, Fortune Also read: AI, Energy, and Sustainable Smart Grids
📊 The Point of the Week
The week of February 9-15, 2026 teaches us that AI is becoming "heavy." Heavy in terms of capital ($50 billion), heavy in terms of responsibility (surgery), and heavy in terms of infrastructure (data centers). The era of lightweight apps and fun chatbots is over. We are in the era of industrial AI, with all the risks and opportunities that entails.
Stay tuned to La Bussola to navigate beyond the hype.
FAQ: Frequently Asked Questions of the Week
1. Why is Amazon investing in OpenAI if it already has Anthropic? Amazon wants to be the "Switzerland of AI": neutral and infrastructural. It wants all the best models to run on its AWS servers. Investing in OpenAI allows it to offer AWS customers native access to GPT-5, preventing them from migrating to Microsoft Azure.
2. What is OpenClaw and why is it different from GPT? GPT is a "brain in a box." OpenClaw is a "brain with hands." It is an open-source framework designed to connect to real enterprise systems (APIs, databases). It is meant to act (e.g., make a bank transfer, update a CRM), not just talk.
3. Is it safe to be operated on by an AI robot today? Robot-assisted surgery is generally safe and precise, but it is always controlled by a human surgeon. The Reuters investigation highlights risks in experimental semi-autonomous systems or when the surgeon trusts the AI's graphical visualization too much without direct verification.
4. What does "Long-Horizon Reasoning" mean in the new models? It means the AI doesn't forget what it said 5 minutes ago. Models like GPT-5.3 can maintain coherence over huge projects (e.g., writing an entire software or analyzing an entire book), planning next steps based on distant goals in time, instead of responding sentence by sentence.
5. Will the "Data Center Compact" affect Europe? Yes. If the USA imposes energy and security standards for AI data centers, global tech companies (Google, Amazon, Microsoft) will likely apply those standards everywhere for uniformity, also affecting European infrastructure.