Episodes

  • Why AI Startups Are Racing to Build Their Own Chips

    Why AI Startups Are Racing to Build Their Own Chips
    Jun 29 2026
    Episode 82 of AI Business with Fexingo explores the accelerating trend of AI startups designing custom chips. Lucas and Luna break down how companies like Cerebras and Groq are challenging NVIDIA's dominance, with a focus on the recent $550 billion South Korean memory investment to ease 'RAMageddon'. They discuss the economics of in-house silicon, citing NVIDIA's 3.1% dip to $193.90 and AMD's 3.1% gain to $536.10 as market signals. The hosts explain why inference costs per query are driving this shift, how custom chips reduce latency for real-time applications, and what it means for enterprise adoption. A specific case study on Cerebras's wafer-scale engine illustrates the trade-offs between flexibility and performance. This episode provides concrete insights for builders and operators navigating the AI hardware landscape, with no fluff, just numbers and real-world decisions. #CustomChips #AIHardware #NVIDIA #AMD #Cerebras #Groq #Semiconductors #Inference #RAMageddon #SouthKorea #EnterpriseAI #Business #Technology #FexingoBusiness #BusinessPodcast #AIStartups #ChipDesign #AIModels Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    7 mins
  • How AI Companies Are Using Micron Memory Chips

    How AI Companies Are Using Micron Memory Chips
    Jun 29 2026
    Amid a broad AI stock selloff in late June 2026, memory maker Micron has bucked the trend, with Wall Street calling it 'the next Nvidia.' In this episode, Lucas and Luna explore why high-bandwidth memory is becoming the bottleneck for AI inference and training, how Micron's HBM4E chips fit into the picture, and what the shift from compute-bound to memory-bound AI workloads means for enterprise adoption. They reference the recent five-day drop in AI stocks like Nvidia and AMD, the 18% plunge in ARM, and contrast that with Micron's relative strength. The conversation digs into the economics of inference cost per token, the role of memory bandwidth in reducing latency, and why hyperscalers are rethinking their hardware stacks. A concrete look at the memory layer that powers the AI stack. #Micron #AI #Semiconductors #Memory #HBM #Inference #Nvidia #AMD #ARM #EnterpriseAI #WallStreet #ChipBottleneck #AIHardware #Latency #DataCenters #Business #Technology #FexingoBusiness Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    8 mins
  • How AI Companies Are Measuring Inference Costs per Query

    How AI Companies Are Measuring Inference Costs per Query
    Jun 28 2026
    Episode 80 of AI Business with Fexingo dives into the economics of inference—the real cost of running an AI model every time it responds. Lucas and Luna break down why inference cost per query has become the key metric for AI companies, from OpenAI to startups deploying small language models. They discuss the surprising numbers: how a single GPT-4 class query can cost a fraction of a cent at scale, and why companies like NVIDIA and AMD are seeing their stock wobble as the market rethinks 'GPU demand equals revenue.' The hosts also explore how inference optimization—like quantization, speculative decoding, and model distillation—is reshaping hardware spend and cloud contracts. With concrete examples and a nod to recent market data (ARM down 18% in five days, SMCI down 13%), this episode connects the engineering trenches to the balance sheet. If you're building or funding AI, this is the metric you need to track. #InferenceCost #AIEconomics #GPU #NVIDIA #AMD #ARM #SMCI #CloudCompute #ModelOptimization #Quantization #SpeculativeDecoding #Distillation #LLM #TechBusiness #BusinessAndTechnology #FexingoBusiness #BusinessPodcast #AI Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    8 mins
  • How AI Companies Are Buying Their Own Data Center Power

    How AI Companies Are Buying Their Own Data Center Power
    Jun 28 2026
    Episode 79 of AI Business with Fexingo dives into the emerging trend of AI companies directly acquiring or building their own power generation assets, from natural gas plants to small modular reactors. Lucas and Luna discuss why firms like Microsoft, Amazon, and OpenAI are moving beyond PPA contracts to own energy infrastructure, driven by surging compute demands and grid constraints. They break down the economics, the risks, and what this means for the future of data center location strategy. Specific references to recent moves by major cloud providers and the role of nuclear restart projects are explored. The conversation also touches on how this shift affects utility stocks and power markets. A must-listen for anyone tracking AI infrastructure and energy policy intersections. #AICompanies #DataCenterPower #EnergyInfrastructure #SmallModularReactors #NaturalGas #CloudCompute #Microsoft #Amazon #OpenAI #NuclearEnergy #GridConstraints #PowerPurchaseAgreements #UtilityStocks #Business #Technology #FexingoBusiness #BusinessPodcast #AIInfrastructure Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    10 mins
  • How Asian AI Startups Are Filling the Anthropic Export Gap

    How Asian AI Startups Are Filling the Anthropic Export Gap
    Jun 27 2026
    With Anthropic's Mythos model under an export ban that has dragged into mid-2026, Asian AI startups are releasing their own 'Mythos-like' models to fill the void. In this episode, Lucas and Luna examine three specific startups — Tokyo's Kizuna AI, Seoul's Hanbit Intelligence, and Singapore's Merlion Labs — that have each released large language models trained on region-specific data and optimized for local languages. They discuss the technical choices these startups made, such as using sparse mixture-of-experts architectures and training on smaller but higher-quality datasets, and what this means for AI sovereignty and enterprise adoption in Asia. They also touch on how NVIDIA's stock at $192 and AMD at $521 reflect investor bets on this decentralized AI build-out. The episode closes with a reflection on whether we're seeing the end of the 'one model to rule them all' era. #AI #Business #Technology #Anthropic #Mythos #ExportBan #AsianAI #KizunaAI #HanbitIntelligence #MerlinLabs #SparseMoE #AISovereignty #LLM #EnterpriseAI #NVIDIA #AMD #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    8 mins
  • How AI Companies Are Using Synthetic Data to Train Models

    How AI Companies Are Using Synthetic Data to Train Models
    Jun 27 2026
    In this episode of AI Business with Fexingo, Lucas and Luna explore how artificial intelligence companies are increasingly relying on synthetic data — artificially generated datasets — to train their models. They discuss why companies like Anthropic, OpenAI, and Meta are turning to this approach, touching on the release of Anthropic's Mythos model to over 100 US companies and agencies. The conversation covers the economics of synthetic data, quality control challenges, and the implications for enterprise adoption. Lucas breaks down how synthetic data can reduce costs and privacy risks while citing a Gartner prediction that 60% of AI training data will be synthetic by 2028. Luna brings up the recent debate around model collapse and why human-generated data still matters. Tune in for a nuanced look at the data strategies powering the next wave of AI. #SyntheticData #ArtificialIntelligence #Anthropic #Mythos #AITraining #MachineLearning #DataGeneration #ModelCollapse #EnterpriseAI #OpenAI #GPT56 #DataPrivacy #BusinessTechnology #TechPodcast #FexingoBusiness #BusinessPodcast #AIModels #DataStrategy Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    10 mins
  • Why AI Companies Are Buying Game Studios for Synthetic Data

    Why AI Companies Are Buying Game Studios for Synthetic Data
    Jun 26 2026
    Episode 76 of AI Business with Fexingo: Lucas and Luna explore a surprising AI data strategy—buying game studios. With NVIDIA down 7.1% in five days and Super Micro plunging 14%, the chip narrative is shifting. But behind the scenes, leading AI labs are acquiring game development teams to generate synthetic visual data for training foundation models. Lucas breaks down the economics: a single AAA game engine can produce millions of labeled frames cheaper than real-world data collection, while circumventing privacy and copyright issues. Luna pushes back on quality concerns, asking whether synthetic data can replicate edge cases like rare car accidents or unusual weather. They point to recent deals—including a major acquisition by a stealth startup—and cite research showing models trained on 80% synthetic data match pure-real performance on certain benchmarks. The episode closes with a question about regulatory scrutiny as synthetic data becomes a critical, unregulated input to the AI stack. A quick behind-the-scenes note: listener support via buy me a coffee dot com slash fexingo keeps this show ad-free and independent. #SyntheticData #AI #GameStudios #NVIDIA #SuperMicro #DataStrategy #FoundationModels #ComputerVision #AIInfrastructure #Business #Technology #Podcast #FexingoBusiness #BusinessPodcast #LucasAndLuna #GenerativeAI #DataPrivacy #EnterpriseAI Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    8 mins
  • How AI Agents Are Stress-Testing Safety Before Launch

    How AI Agents Are Stress-Testing Safety Before Launch
    Jun 26 2026
    Episode 75 of AI Business with Fexingo explores the emerging practice of 'agent stress-testing' — building simulated digital worlds to probe AI agents for dangerous behaviors before deployment. Lucas and Luna discuss Patronus AI's recent $50 million raise, the White House asking OpenAI to slow-roll a model release over safety concerns, and how companies like NVIDIA and Palantir are investing in adversarial simulation. They unpack why traditional red-teaming falls short for autonomous agents and what this means for enterprise adoption. A concrete look at how the industry is trying to catch failures before they cause real-world harm, anchored to the June 26, 2026 market and policy landscape. #AI #AISafety #AgentStressTesting #PatronusAI #OpenAI #WhiteHouse #NVIDIA #Palantir #RedTeaming #AIAlignment #EnterpriseAI #AdversarialSimulation #DigitalWorlds #AIPolicy #Business #Technology #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
    Show More Show Less
    10 mins