No‑BS AI Briefing
No‑BS AI Briefing is for builders who don’t have time for hype. Each episode focuses on a handful of high‑signal stories in AI and AGI, unpacked in simple language with a builder’s perspective. You’ll hear what changed, why it matters, and how you can experiment with the tools, ideas, or strategies yourself—whether you’re leading a team, shipping a startup, or exploring AI side projects.
No‑BS AI Briefing
AI Compute Crunch: Google Throttles Meta, Builders Face Scarcity & Cost
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Google just throttled Meta's AI access. Dhamti. If it can happen to them, it can happen to you. We're also talking about how China's reaching parity in AI cybersecurity, why AI search is ditching big brands for independent creators, and what it all means for your product roadmap and budget. Nobs AI briefing brought to you by Proactive AI. Welcome back. I'm your host, Vikash, and this is where builders get straightforward AI news without the fluff. Alright, let's dive into what's been happening. We've got some big stories that really highlight the practical realities of building with AI right now. First up, a bombshell that should make every builder sit up straight. Google has capped Meta's access to Gemini. According to reports from the Financial Times, Google has been limiting Meta's access to their powerful Gemini models since March of this year. Why? Because they simply can't meet the compute demand. Think about that for a second. Meta, one of the biggest tech giants in the world, is being told by Google, another Titan, that they just don't have enough GPUs for them. This compute crunch has forced Meta to tell its own staff to use AI tokens more efficiently. And it's actually disrupted and delayed several of their internal AI projects. And Google's taking this seriously, securing new capacity, including a massive $920 million per month deal with SpaceX just to lease compute resources. For builders, this isn't just a corporate spat, it's a stark warning. If meta can get throttled, your startup absolutely can. It signals that scaling isn't just about model quality anymore, it's about physical compute access. This scarcity could push prices way up. We're talking 2-3x price increases, so budgeting for AI is about to get a lot harder. Next, something interesting from the cybersecurity front. Chinese firms Jeep AI and 360 Security have launched new cybersecurity tools that are reportedly matching what Anthropics mythos can do. Zipu AI, a major player in China, just released GLM 5.2. This is an open weight model, meaning you can download it and run it locally, and they're positioning it directly against Anthropic Smythos for highly specialized cybersecurity tasks. At the same time, 360 security unveiled their own suite called Yithian Toolong, which includes Thulong Funk for vulnerability detection and Yithian Zhen for defense and incident response. Now, this isn't just about another set of tools. It's a competitive shift. It means Chinese firms are rapidly reaching domain parity in sensitive areas like security. The fact that GLM 5.2 is open weight and can be deployed locally is huge because it reduces the reliance on cloud APIs and provides more control. But on the flip side, it also means the potential for more sophisticated attacks is increasing, which tells us we all need to revisit our threat models and security postures. Are we ready for AI-powered adversaries? Also in the news, the Bank for International Settlements or BIS has issued a pretty stark warning. They believe the current AI investment surge could end in a prolonged bust. We're talking over a trillion dollars in AI investment projected for 2025 and 2026, primarily from hyperscalers. BIS is drawing parallels to historical manias like the dot-com bubble and even the railway boom, suggesting that if these massive investments don't deliver sufficient returns, we could be looking at a drawn-out downturn. Alliance, another major financial player, is also flagging the current environment as bubble territory. The sheer exuberance is highlighted by things like SpaceX's recent $86 billion IPO and a $25 billion bond sale. For founders and builders, this means a few things. Expect venture capital to get tighter if ROI starts to slip. So this isn't just about getting funded, it's about surviving. We need to prioritize measurable impact over flashy narratives. Your unit economics need to be solid, and you should probably plan for harder fundraising rounds, uh biasing towards profitability and sustainable growth rather than uh just chasing the next funding round. And finally, something fascinating for anyone thinking about content and marketing AI search is reportedly favoring independent YouTubers over brand content. The studies are showing that when people ask AI assistants for information in high-intent categories, think how-to guides or product reviews. Those AI assistants are citing independent YouTube creators in roughly half of their responses. What's even more surprising is that 41% of these cited videos have fewer than a thousand views, and 94% of them are 10 minutes or longer. Plus, the traffic coming from AI sourced YouTube videos converts at a staggering 14.2% compared to just 2.8% from traditional Google search. This is a massive shift, isn't it? For product marketers and founders, this means your polished short form brand content might not cut it anymore. You need to rethink your content strategy. Long-form structured videos that are designed to be easily summarized and cited by AI are the new gold standard. This isn't just an SEO tweak, it's a fundamental change in how your content can reach high-intent users and drive conversions at five times the rate of traditional search. It's time to optimize for AI citation, not just keywords. Now, out of all these stories, the one that I think is the most critical for every single builder, every founder, every engineer listening is that first one. Google's compute crunch and why infrastructure is now the bottleneck. What happened here is pretty straightforward but deeply impactful. Google, one of the world's largest cloud providers, is telling Meta, one of the world's largest tech companies, that it doesn't have enough GPUs to meet its AI compute demands for Gemini. This isn't just a minor delay. We're talking about internal projects at Meta being disrupted and pushed back. Google itself is scrambling to acquire more capacity, even signing a monumental nearly billion dollar a month deal with SpaceX for compute. This isn't a hypothetical future problem, it's happening right now with major players. Why does this matter right now? Because for years the narrative has been about who has the best model, who has the smartest AI. But this news completely flips the script. The real constraint on scaling AI products isn't the models anymore. It's the sheer physical infrastructure required to run them. We've hit a data center ceiling and the demand for inference for actually using these models is exploding faster than new hardware can be deployed. This has immediate market implications. Expect prices for AI compute to rise significantly and expect lead times for scaling up your AI infrastructure to get much longer. On the product side, it means reliability can't be taken for granted. If your primary provider can't meet demand, your product roadmap can stall for development, it forces a re-evaluation of every model call. Can we be more efficient? Can we use smaller models? Can we run things locally? So who should really care about this? Honestly, everyone building anything with AI. Yeah, founders need to understand this is a strategic risk that affects their runway, their time to market, and their competitive edge. It's no longer just about building a great product, it's about securing the raw materials for that product. Product managers need to factor compute availability and cost into their feature planning and pricing models. Can your product absorb a 2x or 3x increase in inference costs? Can you guarantee service levels if compute becomes unreliable? Infrastructure engineers and made SREs are now on the front lines. They need to explore multi-cloud strategies, evaluate open weight models for local deployment, and become experts in compute optimization. And yes, even indie hackers should care. If you're building a side project that relies heavily on a single API, what happens if that API gets more expensive or rate limited or even less reliable? Your lean operation could suddenly become a very expensive hobby. How I'd think about it as a builder, given this reality, is that we need to shift our mental model from compute is a commodity to compute is a precious, scarce resource. Think of it like a critical component in your supply chain that suddenly has a massive global shortage. You wouldn't just keep building as usual. You'd look for alternative suppliers, you'd design your product to use less of that component, or you'd explore manufacturing it in-house. For us, that means evaluating our AI dependencies with the critical eye. Are we single-threaded on one provider? Can we architect for redundancy? Can we leverage open weight models that might run on cheaper, more available hardware, or even on the edge? The opportunity here is for companies that can build highly efficient models or those who can innovate in alternative compute solutions. The risk is becoming completely reliant on a supply chain that's showing clear signs of strain. What to ignore? Don't get caught up in the hype that the models will just get better and cheaper. The models are getting better, but the physical infrastructure needed to run them at scale is a very real, very physical bottleneck that can't be wished away. My no-estake here is simple. This isn't just about Google or Meta. This is a fundamental shift in the AI landscape. Compute scarcity is real, it's here, and it's going to impact every single one of us. Plan for it or get left behind. If you want one practical takeaway from today's episode that you can act on this week, here it is. Audit your compute dependencies in 30 minutes. Here's how to try it in under 30 minutes. First, grab a quick coffee and list out every single AI provider you currently use across your products or internal tools. Think about all your OpenAI calls, Anthropic APIs, Google Cloud AI services, even any third-party services that abstract AI for you but are themselves relying on a major provider. Just a simple list. Don't overthink it, just get it down. Second, for each item on that list, flag any single points of failure. Are you exclusively using one vendor for a critical function? What would happen if that vendor suddenly capped your access or raised prices by 3x? This isn't to panic but to identify where your biggest risks lie. Are there any parts of your product that would break if one API went dark? And third, identify just one self-hostable fallback option for one of those flagged critical dependencies. For example, if you're using GPT-4 for text generation, could you identify an open weight model like GLM 5.2 as we just heard about from China? Or another open source model that could potentially handle a portion of that workload if your primary API became unavailable or too expensive. You don't need to implement it, just identify it as a potential alternative. Why is this specific experiment worth your time right now? Because what happened to Meta isn't an anomaly, it's a bellwether. Compute scarcity isn't going away overnight. By taking just 30 minutes, you're not just identifying risks. You're starting to build a mental map for resilience. You're giving yourself options or at least the awareness that you need to start thinking about them. This isn't about being paranoid, it's about being proactive and ensuring your product's future isn't solely in the hands of a single cloud provider's GPU allocation. It's about building defensively, which is crucial in this new era of AI. That's it for today's No BS AI briefing. If this helped, follow the show in your podcast app and share it with one builder you know. And if you've got questions or topics you want covered, connect with me on LinkedIn and send them over. See you in the next briefing.
