About this episode AI Unfiltered delivers your daily essential briefing on the artificial intelligence landscape, cutting through the hype to highlight what truly matters. We expertly analyze new models, groundbreaking research, key product launches, and significant funding rounds. Tune in to gain a seasoned researcher's perspective, empowering you to distinguish real industry-shifting signals from the ever-present noise.
0:00 Alphabet just announced an eighty-five billion dollar equity raise.0:12 Last time we talked about NVIDIA’s hardware moves, but today Google’s parent company showed us the check they’re writing to buy all of it, and then some.0:39 This isn't just funding.0:43 It's a declaration of a new Cold War in computing infrastructure.0:55 That’s the headline that changes the map.1:02 But the ground is shifting everywhere.1:09 In China, MiniMax just released the M3 model.1:17 It's open-source, with a one million token context window.1:28 The claim is that it kills closed-source models.1:36 We’ll see if the weights, when they drop, back that up.1:46 Meanwhile, Hasbro launched a new division, Sixth Wall AI Studio.1:58 They're not building models.2:03 They're building personalities.2:08 Using a new system called CharacterOS, they’re creating official, licensed AI versions of characters like Optimus Prime.2:30 It’s a new business model: Behavioral Licensing.2:38 And in Europe, the first commercial robotaxi service just launched today in Zagreb, Croatia.2:55 It's run by Project 3 Mobility, powered by Pony-dot-ai.3:05 It’s a huge milestone for the continent, even with a human safety driver still behind the wheel.3:22 Finally, the CEOs of OpenAI, Anthropic, and others signed a joint letter.3:35 They are urging the U.S.3:39 Congress to pass laws preventing AI from being used to develop biological weapons.3:54 A necessary, if sobering, reminder of the stakes.4:03 Let’s go back to that Alphabet number.4:10 Eighty-five billion dollars.4:15 It’s the largest equity raise of its kind, and it’s for one purpose: AI infrastructure.4:30 To understand the scale of this shift, look at their capital expenditures.4:43 In 2022, Alphabet spent thirty-one billion.4:51 For 2026, they are now projecting one hundred and eighty to one hundred ninety billion dollars.5:08 That is a six-fold increase in four years.5:16 CEO Sundar Pichai’s explanation was simple.5:23 Demand is "meaningfully exceeding available supply." They physically cannot build data centers and manufacture their eighth-generation TPUs fast enough.5:51 The good news for them is that the cost to serve an AI query dropped seventy-eight percent last year.6:09 So the unit economics are improving.6:15 The problem is, the volume is exploding.6:22 This eighty-five billion dollars isn’t for moonshots.6:32 It’s to keep the lights on for a world that just became addicted to AI.6:45 Now, contrast that brute-force spending with what MiniMax just did.6:57 Their new M3 model isn't just another open-source release.7:07 The key is its architecture—something called MiniMax Sparse Attention, or M-S-A.7:21 For years, the problem with long context windows—a million tokens, in this case—was cost.7:37 To generate one new word, the model had to re-read everything.7:49 It was computationally insane.7:54 M-S-A changes that.7:57 It allows the model to pay attention to only the most relevant parts of that million-token history.8:15 The result?8:17 MiniMax claims they can run a one million token context window at five to ten percent of the cost of models like GPT-5.5.8:39 Let that sink in.8:42 A 90 percent cost reduction on the single most expensive part of inference.8:55 On the SWE-Bench Pro coding test, it scored 59 percent, beating Gemini 3.1 Pro and just behind Anthropic's Opus 4.7.9:16 This isn't a toy.9:19 So here are the two forces shaping today.9:26 On one side, you have Alphabet, spending the equivalent of a small country's GDP just to meet baseline demand for centralized AI.9:50 They are building a bigger pipe.9:55 On the other side, you have a breakthrough like MiniMax M3, which doesn’t need a bigger pipe because it uses less water.10:17 One strategy is about owning the infrastructure.10:25 The other is about making the infrastructure irrelevant.
