As AI inference is projected to represent 80% of all AI compute, investors should target key public infrastructure and memory providers like Micron Technology (MU), SanDisk, and Cerebras to capture rising demand. In private semiconductor markets, revenue-generating Positron at a sub-$10 billion valuation offers a safer, more fundamentally grounded entry point than pre-revenue peer Etched at $21 billion. Mega-cap private asset Anthropic remains a premier investment target, with run-rate revenue rapidly accelerating to $65 billion ahead of an anticipated $2.0 trillion IPO. High-growth vertical software leaders like Higgsfield in generative video and Harvey AI in legal automation illustrate massive enterprise monetization and high switching costs. To effectively capture this cycle, investors should build a diversified basket focused on AI infrastructure and memory hardware rather than betting on a single early-stage chip designer.

Etched (Private)

• Valuation Surge: Etched doubled its private valuation from $10 billion to $21 billion in just one month.
• Technology Focus: The company builds specialized semiconductor chips dedicated to AI inference compute (running live models) rather than training.
• Operational Progress:
  • TSMC manufactured and delivered their initial chip batch, which was running inference models within four months with solid fab yields.
  • The company has reportedly poached notable engineering talent from NVIDIA (NVDA).
  • Famous investor Michael Burry reportedly claimed the company could eventually be bigger than NVIDIA.
• Key Risks:
  • Etched is currently pre-revenue and has not yet delivered completed products to paying customers at scale.
  • Investors face execution risk around whether the company can manufacture and deliver against a backed-up order book to justify a $21 billion+ valuation.

Takeaways

• Etched offers high-upside exposure to dedicated inference hardware, but investors should weigh the steep $21 billion pre-revenue valuation against execution and delivery risks.

Positron (Private)

• Valuation & Fundamentals: Operating in the same AI inference semiconductor space as Etched, Positron is currently valued at sub-$10 billion.
• Commercial Traction: Unlike pre-revenue peers, Positron is already shipping silicon and generating revenue in inference compute.

Takeaways

• Positron represents a potentially lower-risk, fundamentally grounded alternative to higher-valuation inference chip competitors due to its existing revenue generation and lower entry valuation.

Anthropic (Private)

• Revenue Velocity: Anthropic’s run-rate revenue reached $65 billion, up from $10 billion in December and $47 billion in May (adding $18 billion in roughly 2.5 months).
• Valuation Expectations: Private secondary market pricing ranges around $1.5 trillion, with discussion of a potential future IPO valuation reaching $2.0 trillion.
• Market Dynamics & Risks:
  • Growth is driven by accelerating enterprise AI adoption, always-on AI agents, and expanding autonomous bots.
  • The primary competitive risk is the rapid development of low-cost, open-source AI models (particularly from China), though overall AI market expansion continues to drive massive top-line growth regardless of market share shifts.

Takeaways

• Anthropic is demonstrating unprecedented top-line growth, making it a premier mega-cap private AI asset, though investors must monitor how open-source competition impacts long-term pricing power and margins.

Higgsfield (Private)

• Growth Metrics: The AI video generation platform reached 30 million users and quadrupled its valuation to $5.4 billion in eight months.
• Revenue Performance: Annual recurring revenue (ARR) hit $700 million, representing a 35x year-over-year run-rate increase.
• Market Disruption: The platform enables low-cost, high-fidelity commercial and cinematic video creation, drastically reducing the production costs traditionally required for commercial advertising and movie making.

Takeaways

• Higgsfield highlights the rapid monetization potential of generative video tools, showing that generative creative media is moving quickly from consumer novelty to scalable enterprise and commercial production.

Harvey AI (Private)

• Financial & Usage Milestones:
  • ARR expanded to $350 million, with company valuation increasing 41% in approximately five months.
  • Platform usage scaled from 1 trillion tokens in January to 14.5 trillion tokens in June.
• Product Development: Introducing enterprise and user-level "memory," allowing models to retain historical context, client nuances, and law firm data over time.
• Business Model Strength: Harvey utilizes forward-deployed engineers to embed custom workflows directly into enterprise clients, creating high switching costs and recurring, automated usage.

Takeaways

• Harvey illustrates that vertical AI software tailored to specific high-value industries (like legal services) can achieve rapid lock-in and compounding usage by integrating directly into client operational workflows.

AI Infrastructure: Inference & Memory Hardware

• Shift Toward Inference: As AI applications, robotics, and personal multimodal devices (such as smart glasses and camera-equipped earbuds) proliferate, inference compute is projected to represent 80% of all AI compute demand.
• The Memory Bottleneck: Adding long-term context and memory to models creates a massive requirement for storage and high-bandwidth memory hardware.
• Public Market Mentions:
  • Micron Technology (MU) and SanDisk stand to benefit from long-term memory and data retention requirements in AI systems.
  • Cerebras has seen upward momentum (up 18% over the past month) amid renewed demand for specialized AI silicon.

Takeaways

• Rather than attempting to pick a single winner among early-stage chip designers, investors should consider a diversified basket approach across inference hardware and memory infrastructure providers to capture secular AI compute demand.