Coinbase CEO Brian Armstrong Breaks Down the Three Biggest Trends in Crypto + More from Davos!

What It Covers

The All-In podcast broadcasts from Davos 2025 with interviews featuring Coinbase CEO Brian Armstrong on crypto regulation and stablecoin legislation, Cerebras Systems CEO Andrew Feldman on AI chip architecture and data center infrastructure, and Gecko Robotics CEO Jake Loosararian on industrial robotics deployment. Topics span the Genius Act implementation, wafer-scale computing advantages, and AI-driven infrastructure inspection automation.

Key Questions Answered

• •Stablecoin Regulation Framework: The Genius Act mandates US-regulated stablecoins maintain 100% reserves in short-term US treasuries with maximum thirty-day maturity, creating safer alternatives to fractional reserve banking. This eliminates run-on-bank scenarios like Silicon Valley Bank's collapse while enabling rewards programs based on customer activity beyond balance holdings. Five of the top twenty global systemically important banks now use Coinbase infrastructure to integrate crypto products.
• •Crypto Market Structure Transformation: Coinbase identifies three dominant trends reshaping the industry - all asset classes moving on-chain for trading including equities, prediction markets experiencing explosive growth through partnerships like Kalshi integration, and stablecoin payments accelerating especially for business-to-business cross-border transactions. The company processes massive backlogs for Coinbase Business serving small and medium enterprises seeking faster settlement and lower foreign exchange fees.
• •Wafer-Scale Computing Architecture: Cerebras builds chips fifty-six times larger than NVIDIA's B200 containing four trillion transistors, delivering twenty to fifty times faster performance for AI workloads. Large chips process more information simultaneously and deliver results faster because memory access speed creates the primary bottleneck in GPU-based inference. Systems cost one to one-point-five million dollars on-premise or rent by token starting at fifty cents per million tokens in cloud deployments.
• •AI Infrastructure Power Economics: Data centers now measure capacity in megawatts rather than square footage because power availability represents the limiting constraint for deployment. The cheapest power sources rank as hydroelectric first, then natural gas especially from petroleum flare-off previously wasted. Cerebras secured a seven-hundred-fifty megawatt deal with OpenAI requiring multi-year buildout across locations with abundant natural gas or geothermal resources in regions like West Texas and The Nordics.
• •Memory Supply Chain Dynamics: The AI industry faces an eighteen-month memory shortage caused by demand signal confusion rather than actual production constraints. Companies escalated orders from six months to eighteen months of inventory when receiving uncertain delivery timelines, creating artificial scarcity. High-bandwidth memory consumption by GPUs particularly strains DRAM supply for consumer electronics. Manufacturing capacity remains unchanged from four months prior despite price increases and extended lead times.
• •Industrial Robotics ROI Model: Gecko Robotics generates thirty percent revenue from defense applications including submarine manufacturing and destroyer maintenance, achieving ninety percent speed improvements in production cycles. Purpose-built robots with specialized sensor arrays inspect metal fabrication quality, welds, and structural integrity to build comprehensive databases of infrastructure health. The company plans to purchase humanoid robots from Tesla, Figure, or Boston Dynamics to create an application layer rather than building general-purpose hardware.
• •Job Displacement Timeline Assessment: Current tech layoffs primarily result from delayed impacts of SaaS tools expanding management scope rather than AI automation. Middle management roles focused on information transfer and small team coordination diminish in value as leaders gain direct visibility through integrated data systems. True AI-driven displacement remains three to five years away when foundation models for specialized tasks like welding reach production readiness, requiring human supervision of multiple automated systems rather than direct field work.