Overview of The Jaeden Schafer Podcast
This episode covers major AI industry moves: Google’s multi-layer push at Google Cloud Next (new TPUs, Chrome as an “AI coworker,” and a large compute deal with Thinking Machine Labs), startup fundraising and new labs, an enterprise security leak involving Anthropic’s Mythos, and an OpenAI–Infosys enterprise distribution deal. The host also highlights two startups (10x Science and Neocognition) and gives practical advice for enterprises using AI tools.
Key stories (top-level)
- Google announced TPU 8T (training) and TPU 8i (inference), a Chrome “auto-browse” AI coworker powered by Gemini, and a multi-billion-dollar compute deal with Thinking Machine Labs at Cloud Next.
- OpenAI signed a distribution/integration deal with Infosys to push ChatGPT/Codex into Infosys’ enterprise clients across 60+ countries.
- Bloomberg reports an unauthorized group accessed Anthropic’s Mythos cybersecurity model via a third-party credential and predictable endpoint pattern.
- Two startups raised seed rounds: 10x Science ($4.8M) and Neocognition ($40M). Both aim to solve operational problems in AI-to-biotech and AI agents specialization respectively.
Startups & research labs covered
10x Science
- Background: Stanford spinout from Nobel laureate Carolyn Bertozzi’s lab; $4.8M seed led by Initialized Capital.
- Problem addressed: Triage and analysis of thousands of AI-generated drug candidates (bottleneck in pharma). Current triage (mass spectrometry) is slow, expert-dependent, and hard to interpret for regulators.
- Solution: SaaS layer combining deterministic chemistry + AI agents to make results traceable, explainable, and regulator-friendly (avoid black-box claims).
- Why it matters: Fills the “picks-and-shovels” tooling gap under generative biotech models.
Neocognition
- Raised ~$40M seed; founder is an academic (AI agent lab at Ohio State).
- Thesis: Current agents are unreliable generalists. Neocognition aims to build agents that self-specialize quickly in new domains—mirroring human rapid specialization—so vertical scaling doesn’t require custom engineering per use case.
- Team: Small (~15), PhD-heavy; notable early angel/backer interest from high-profile industry figures.
- Why to watch: If successful, could significantly improve agent reliability and scalability across verticals.
Anthropic Mythos breach (summary & implications)
- Report: An unauthorized group in a private Discord gained access to Mythos (Anthropic’s enterprise cybersecurity tool) by obtaining credentials from a third-party contractor and guessing the model endpoint via predictable URL patterns.
- Anthropic’s stance: Investigating; claims no evidence of internal system compromise—incident was in vendor environment.
- Key lesson: Many security incidents arise from vendor/contractor credential exposure and predictable endpoint patterns rather than exotic model exploits.
- Immediate advice for enterprises: Audit vendor access, tighten credential practices, and limit predictable, globally accessible endpoints.
OpenAI + Infosys deal (enterprise expansion)
- Deal: Infosys will embed OpenAI models (ChatGPT, Codex) into its Topaz AI platform and client transformations across 60+ countries. Financial terms undisclosed.
- Strategic value: Infosys provides distribution to Fortune-tier accounts and large enterprise deals where Microsoft/Azure may not be the integrator. This can accelerate OpenAI’s enterprise revenue growth.
- Context: Infosys already earns significant AI service revenue and acts as a system integrator for large transformations.
Google Cloud Next — announcements & analysis
Major product news
- TPU chips: TPU 8T (training) and TPU 8i (inference). Google claims:
- 3x faster training vs. NVIDIA alternatives (per Google).
- ~80% better performance-per-dollar for inference.
- Ability to scale >1 million TPUs in a single cluster.
- Chrome “auto-browse” (AI coworker):
- Powered by Gemini; reads context across open tabs to automate workplace tasks (CRM updates, quote comparisons, candidate summaries, competitor research).
- Workflow “skills” can be saved and triggered; U.S. Workspace users getting it first.
- Enterprise prompts will not be used to train Google models (privacy promise).
- Compute partnership: Multi-billion-dollar deal to give Thinking Machine Labs access to high-end NVIDIA GB300 systems (on Google Cloud) plus training and deployment services for their custom frontier-model product.
Strategic analysis: Google’s three-layer strategy
- Layer 1 (Silicon): Google offers its TPUs and resells NVIDIA GPUs—giving customers chip choice.
- Layer 2 (Compute host): Google Cloud hosts frontier labs (Anthropic already runs on TPUs; Thinking Machine now on Google Cloud).
- Layer 3 (Agent & application): Chrome + Workspace integration (auto-browse) places Google at the user-facing agent layer.
- Why this matters: Google is one of the few companies credibly operating across silicon, cloud compute for frontier labs, and the agent/application layer. That full-stack position could give structural advantages (cost, integration, distribution) beyond pure model benchmark leadership.
Risks & caveats
- Benchmarks: Google’s TPU performance claims are internal; independent, real-world benchmarks are still needed.
- Model quality: Gemini still trails leading consumer-facing models on some benchmarks; real-world cost and integration advantages may outweigh leaderboard positions for many customers.
- Regulatory scrutiny: Integrating Chrome and Workspace with deep AI capabilities will attract antitrust / privacy attention (DOJ, EU). This could trigger additional oversight.
Notable quotes & insights
- “Google is running a three-layer strategy — silicon, compute hosting for frontier labs, and the agent layer — and it’s the only player hitting all three credibly right now.”
- Security pattern: “It’s not really a model exploit. It’s just a contractor credential plus a predictable URL pattern.”
Watchlist (what to monitor next)
- Independent TPU 8T / 8i benchmarks and real-world cost comparisons vs NVIDIA.
- Auto-browse rollout and whether it moves from pilot into broader enterprise usage.
- Whether additional frontier labs migrate to Google Cloud after Thinking Machine Labs.
- Anthropic’s investigation outcome and any enterprise/IPO PR impact.
- How the OpenAI–Infosys deal affects OpenAI’s enterprise revenue growth trajectory.
Practical recommendations (for enterprises & practitioners)
- Audit vendor and contractor access to AI systems; enforce least privilege and rotate credentials.
- Avoid relying on model claims alone—ask for independent benchmarks and cost-per-inference projections.
- When choosing cloud/AI partners, compare stack-level integration (cost, data flow, governance) in addition to raw model accuracy.
- If you’re integrating AI into workflows, evaluate solutions that support automation, skills, and enterprise data privacy commitments.
Final takeaway
Google’s Cloud Next announcements signal a strategic push to own more of the full AI stack (silicon → compute → agent/user layer). That structural positioning could be decisive even if model benchmark leadership isn’t yet clear. Meanwhile, enterprise AI adoption accelerates via SI partnerships (Infosys + OpenAI), startups are building critical tooling and agent-specialization research, and operational security (vendor access) remains a top short-term risk.