Investors should maintain exposure to NVIDIA (NVDA) due to relentless compute demand while targeting emerging opportunities in silicon photonics and optical networking that resolve critical data-routing bottlenecks.

Enterprise software leaders like Salesforce (CRM) and Snowflake (SNOW) are prime rebound opportunities as they leverage proprietary data repositories to power autonomous AI workflow agents.

Allocate capital to defense technology and domestic manufacturing suppliers benefiting from increased government spending on autonomous systems and counter-drone defense.

Investors should also target physical computing bottlenecks by acquiring exposure to companies building out power infrastructure and specialized hardware for AI data centers.

Finally, seek opportunities in AI data infrastructure providers that supply the proprietary datasets and simulation environments required to train robotics foundation models.

Hard Tech & Physical Computing Infrastructure

• Venture capital investment in "atoms" (physical hardware and infrastructure) has surged, rising from 8% to 20% of recent Y Combinator batches.
  • Sub-sectors experiencing significant expansion include industrial manufacturing (up from 4% to 10%), robotics (up from 1% to ~7%), defense tech (up from 1.5% to 5%), semiconductors and silicon photonics (up to ~4%), and power infrastructure (up to ~3%).
  • AI-assisted coding and advanced scientific reasoning models are reducing development timelines and engineering overhead, enabling smaller teams to tackle complex hardware and supply-chain challenges.
  • Key structural tailwinds include the reshoring of American manufacturing, satellite infrastructure expansion following the SpaceX IPO, and government modernization efforts across defense.

Takeaways

• Look for investment opportunities in companies solving physical bottlenecks for AI—specifically advanced power generation, custom silicon, optical switching, and domestic industrial supply chains.

NVIDIA (NVDA) & AI Hardware Alternatives

• Demand for AI compute continues to outpace supply, causing older hardware like NVIDIA A100 GPUs to appreciate in cost per hour rather than depreciate.
• Hardware efficiency trends show LLM architectures require less floating-point precision, moving down from FP32 and FP16 toward FP8, FP4, and even FP2 / ternary representations.
  • Startups are developing alternative silicon architectures (such as Lam Labs and Bot) and photonic optical interconnect switches (such as Dipole Labs) to resolve routing bottlenecks between GPUs.

Takeaways

• While high demand supports premium pricing for legacy and current GPU architectures, monitor competitive developments in optical networking, lower-precision custom silicon, and photonic interconnects designed to bypass traditional electronic switching limitations.

Enterprise SaaS & Systems of Record: Salesforce (CRM) & Snowflake (SNOW)

• After a broader software pullback, legacy SaaS platforms have rebounded, highlighted by strong earnings reports from Salesforce and Snowflake.
• The competitive dynamic for software has shifted from human-operated point solutions to automated, full-stack workflow agents.
  • Core systems of record maintain strong defensive moats if they serve as the central repository of structured enterprise data used by autonomous agents.
  • Salesforce is leveraging Slack as an AI harness to embed collaboration data directly into autonomous agent workflows, defending its data moat against standalone AI tools.

Takeaways

• Focus on software businesses that possess proprietary system-of-record data and can successfully deploy their own agent harnesses, as autonomous agents are expected to consume and interact with enterprise software at much higher frequencies than human workers.

Defense Technology & Domestic Manufacturing

• Modern defense procurement is shifting away from traditional cost-plus defense primes toward agile, venture-backed startups leveraging computer vision, autonomous systems, and advanced communications.
  • Startups like Icarus (solar-powered overwatch aircraft) and Nine Mothers (counter-drone defense systems) have secured seven-figure contracts with defense agencies.
  • Dual-use industrial suppliers, such as Knox Metals in Detroit, are growing at software-like rates by supplying raw components to the rapidly expanding defense tech ecosystem.

Takeaways

• The convergence of autonomous systems, counter-drone technologies, and domestic industrial supply chains represents a high-growth sector supported by non-cyclical government and enterprise demand.

AI Data Infrastructure & Reinforcement Learning (RL) Environments

• Providing specialized training data and Reinforcement Learning (RL) simulation environments to frontier AI labs has evolved into a multi-hundred-million-dollar annual market.
  • Top frontier AI labs are spending an estimated $1 billion collectively on custom domain-specific data and simulation environments.
  • Demand is expanding rapidly into physical world training, including egocentric video, teleoperation, and industrial manufacturing datasets designed to train robotic foundation models (e.g., fine-tuning physical models like Pi).

Takeaways

• Proprietary, high-fidelity datasets in specialized domains (finance, physical robotics, enterprise workflows) represent an essential and monetizable layer in AI model training.