🤖 AWS + AI = Amazon’s Profit Engine

🎰 AWS + AI = Amazon’s Profit Engine

🪽 Teaching AI to Admit When It Doesn’t Know

📍 6 AI Stock Sectors

🛠️ 5 Trending Tools

🥪 4 Brief Bites

💰 3 Funding Frontlines

💼 4 Job Opportunities

On Wednesday, Amazon announced earnings for Q1 FY2026, with net sales increasing 17% to $181.5 billion, operating income increasing 30% to $23.9 billion, and net income increasing 77% to $30.3 billion when compared to Q1 FY2025. For context, Amazon outperformed Wall Street’s net sales expectations, beating the $177.2 billion projection by $4.3 billion.

It’s important to note that Amazon’s reported net income is inflated because it includes a $16.8 billion pre-tax valuation gain from Amazon’s investment in Anthropic. In simple terms, because Anthropic’s FMF increased, Amazon recorded that increase as a net gain on the income statement.

AWS contributed about $37.6 billion in net sales, meaning the cloud computing business accounted for around 21% of the online retail giant’s Q1 revenue. AWS remains the primary driver of profitability, making up over 59% of Amazon’s total operating income.

AWS and OpenAI recently expanded an existing $38 billion multi-year compute capacity agreement by $100 billion over eight years. This expansion includes Amazon investing $50 billion in OpenAI in exchange for its consumption of 2 GW worth of Trainium capacity through AWS. Amazon also announced a strategic collaboration with Anthropic that includes investing up to $20 billion and supplying up to 5 GW of compute in exchange for Anthropic spending $100 billion over the next ten years on AWS.

What’s notable isn’t just AWS’s impressive sales growth. It’s how that sales growth is being engineered. By anchoring Amazon CEO Andy Jassy’s $200 billion AI spending spree to customer commitments with frontier AI firms like OpenAI and Anthropic, Amazon effectively locks in demand before supply is fully brought online.

Today’s language models are statistical systems or estimation engines designed to predict the probability of a sequence of words. In simple terms, they’re essentially sophisticated autocomplete machines trained on the entire internet.

For example, when given: “The cat chased the {BLANK}!” language models ask themselves, given the words so far, what’s the most likely next word? In this case, it might predict: “{MOUSE}!”

The fundamental issue is the “mismatch of objectives.” In other words, a language model is trained to process diverse inputs and generate plausible outputs that sound human.

That means if the most “probable” next word within a sentence is factually incorrect but grammatically correct, the language model will pick it because it lacks grounding: it can’t double-check a fact against the real world. It can only check if the next word within a sentence fits the statistical patterns it learned from billions of training examples.

To address this, frontier AI firms pioneered reasoning models like “OpenAI o1,” which leverage RL: a training technique that rewards reasoning models for getting the right answer. But, this still has a critical flaw.

With RL, reasoning models earn the same reward for right answers whether they carefully reason or correctly guess. Over time, this teaches them to confidently answer every question they’re asked, whether supported by strong evidence or pure guesswork.

The world’s best frontier AI models, such as OpenAI’s “GPT-5.5,” aren’t just passive next-word predictors in practice. They’re often wrapped in retrieval and reasoning. Even so, “GPT-5.5” can still hallucinate when it lacks reliable external verification because the core algorithmic architecture still functions as a sophisticated autocomplete machine that confidently guesses when it doesn’t know.

The world’s best frontier AI models don’t reliably know when they might be wrong. “RLCR” is engineered to fix miscalibrated confidence. In simple terms, it isn’t trying to make them better at getting answers right. It’s trying to make them better at learning when not to answer.

Firestorm Labs raised an $82M Series B for AI-assisted drone factories.

Scout AI closed a $100M Series A to train AI for unmanned warfare.

Rogo raised a $160M Series D to help bankers with finance work.

Figure AI (San Jose, CA): Electrical Engineering Intern, Fall 2026

NVIDIA (Santa Clara, CA): Hardware Applications Engineer, Entry-Level

Google DeepMind (London, UK): Gemini Diffusion Scientist, Mid-Level

OpenAI (San Francisco, CA): AI Success Engineer, GTM, Senior-Level