Welcome to the 72nd edition of Deep Tech Catalyst, the channel by The Scenarionist, where science meets venture!
Whether you're building AI infrastructure, developing semiconductor technology, or navigating the path from prototype to pilot, the road for deep tech founders is anything but simple.
Hardware-heavy startups face long cycles, complex supply chains, and often need to prove traction before having a product in hand. But with the right strategic moves—and the right questions—early-stage teams can still stand out.
To help decode this journey, we’re joined by Henry Huang, Investment Director at Micron Ventures!
In particular, we explore:
The core components of AI hardware and where innovation is headed
What makes networking—and especially optical interconnects—an exciting space for startups
How to think about tape-outs, pilots, and ecosystem design as a founder in semiconductors
What early-stage investors actually look for when there’s no product or customer yet
Let’s get into it. 🚀
✨ For more, see Membership | Deep Tech Briefing | Insights | VC Guides
BEYOND THE CONVERSATION — REFLECTIONS & STRATEGIC INSIGHTS FROM THE EPISODE
Framing AI Hardware Infrastructure
When it comes to identifying high-potential areas for innovation in AI hardware, it’s helpful to frame the ecosystem around three core pillars: compute, memory and data storage, and networking. Each of these plays a critical role in supporting scalable, efficient AI infrastructure.
1. Compute
Over the past decade, the venture capital industry has invested heavily in AI processor technologies. While startups have made significant advancements, it’s important to note that NVIDIA continues to dominate the space, capturing the majority of the market value.
Because of this, achieving venture-scale returns in the compute domain is increasingly challenging for new entrants. The market is highly competitive, and the bar for differentiation is extraordinarily high.
2. Memory and Data Storage
The second pillar—memory and data storage—also presents a significant market opportunity, particularly as High Bandwidth Memory (HBM) emerges as a critical enabler of AI performance.
That said, from a venture perspective, this space remains challenging for startups.
The primary barrier is the capital-intensive nature of semiconductor manufacturing. Established players control large-scale fabrication infrastructure, resulting in high entry costs and few successful startup exits in the memory space over the past 10 to 15 years.
3. Networking
The third pillar, networking, is where attention is rapidly shifting—and where the greatest potential may lie for startups today.
As AI models become more distributed and data-intensive, interconnect technologies are increasingly critical. The need to efficiently move data between processors, memory units, and systems within data centers is now a central challenge.
Interestingly, one of the most successful startups in the AI hardware space isn’t a processor company at all—it’s a firm focused on high-performance networking solutions.
Technical Pressures at the Core of AI Hardware
Founders working at early stages—particularly from TRL-4, where prototypes begin to take shape—need a clear view of the market structure, key challenges, and potential opportunities.
One of the primary challenges in the semiconductor sector—particularly as it applies to AI—is efficiency. This challenge can be divided into 2 major dimensions:
1. Power Efficiency
A significant concern is the power consumption associated with AI workloads. Data centers are well known for their high energy usage, especially when training large models such as ChatGPT and other generative AI systems.
However, the power draw doesn't come solely from the processors.
A large portion of energy is consumed in moving data—between processors and memory, and across nodes within clusters and data centers. This data movement is an increasingly important area of focus for innovation.
2. Resource Utilization Efficiency
Another underappreciated issue is the underutilization of computational resources.
For example, Meta has published data showing that over 30% of AI model training time is spent waiting for network responses. This implies that substantial infrastructure resources sit idle during those periods.
From the perspective of memory providers, addressing this inefficiency—especially through innovations in memory technologies—represents a critical opportunity. Reducing latency and improving data throughput can unlock significant gains in both performance and energy efficiency.
The Rise of Optical Interconnects
As AI workloads continue to scale, the networking layer of data center infrastructure is undergoing a major transformation—one that presents compelling opportunities for startups.
Among the most promising areas is optical interconnect technology, which is gaining momentum due to its ability to deliver higher speeds, lower latency, and greater energy efficiency.
What makes this space especially attractive is that barriers to entry are relatively lower, making it more accessible to early-stage innovators. These opportunities are being driven by two main factors:
1. The Shift from Electrical to Optical Networking
A significant shift is taking place in data center architecture: networking systems are moving from electrical-based to optical-based designs.
This transformation is largely driven by the superior efficiency and bandwidth that optical technologies offer compared to traditional electrical interconnects.
Technologies such as silicon photonics—including both micro-LED and laser-based solutions—are central to enabling the next generation of high-speed, energy-efficient connections between processors and memory units.
2. More Flexible and Modular Innovation Pathways
Unlike traditional semiconductors, where manufacturing demands massive capital expenditure and access to proprietary fabs, the optical networking space offers more flexible and modular innovation paths.
This makes it significantly easier for startups to enter the market and develop differentiated solutions.
To Recap:
Given the underutilization of networking resources in AI training environments, any solution that optimizes data movement—regardless of whether it's electrical or optical—has the potential to unlock significant value.
Improving network efficiency has a direct impact on training and inference throughput, making this a particularly attractive area for investment and technical innovation.
From Lab to Pilot: Navigating Early-Stage Milestones in Semiconductor Startups
Transitioning from a prototype to a successful pilot—especially in the highly complex semiconductor and AI hardware sectors—requires both technical execution and strategic business engagement. Based on insights from venture-backed case studies, several key milestones and best practices can guide early-stage founders through this journey.
3 Key Development Milestones to Know
In semiconductor development, progress is typically marked by tape-out milestones, which refer to various phases of chip readiness:
Engineering Tape-Out: This marks the first iteration of working silicon. It enables internal testing and provides a base for system-level demos. At this stage, startups should begin engaging with potential customers by showcasing early performance and collecting feedback.
Customer Sample Tape-Out: At this point, the prototype is stable enough to be shared with select customers, even in small volumes. This phase is ideal for deepening technical discussions and piloting limited evaluations.
Production Tape-Out: The final milestone signals readiness for broader deployment. Chips are validated and manufactured at higher volumes for full-scale commercial adoption.
Don’t Underestimate the Role of Software
One of the most critical yet often overlooked factors in early customer adoption is software readiness. Hardware alone is rarely enough to win a customer engagement. Increasingly, success depends on the quality and usability of the software layer that supports the silicon.
💡 Core Insight: Consider a software-first approach. Providing early software tools—such as simulators—can enable potential customers to evaluate and integrate your solution long before hardware is fully available. This builds confidence and shortens adoption timelines once your product is production-ready.
Tailor Engagement Strategies to Your Customer Segments
Founders should recognize that customer expectations and engagement strategies vary significantly depending on whether the startup targets cloud or edge markets.
1. Cloud Segment: Fewer, Larger Customers
Startups targeting cloud infrastructure often deal with a handful of hyperscalers. Winning one of these clients can transform a company’s trajectory almost overnight.
Tactical Advice:
Deeply research and understand the needs of these large buyers.
Customization and performance optimization are key differentiators.
Establishing relationships with these clients often requires ecosystem credibility and integration readiness.
2. Edge Segment: Many Smaller Customers
In edge computing, startups must engage with a broader and more fragmented customer base. Success here often depends on identifying a beachhead market—a small but receptive niche that can rapidly adopt and validate your solution.
Tactical Advice:
Use small-scale trials to mature your product.
Build early reference customers to gain credibility.
Prioritize speed and ease of deployment.
An Example of How Strategic Collaboration Can Unlock Adoption from a Hyperscaler
One of the most effective strategies for early-stage startups—especially in deep tech—is to leverage ecosystem partnerships. In many cases, startups struggle not because of technical shortcomings, but because they can’t deliver a complete, integrated solution.
Here’s one example: two startups co-developed a solution and pitched it jointly to a major hyperscaler. Rather than approaching the opportunity independently, they assembled a consortium of ecosystem partners to present a unified, end-to-end offering. This collaborative approach significantly boosted their credibility and ultimately led to a high-value contract.
Investor Perspective: How Early-Stage Startups Can Signal Credibility
As mentioned, one of the most critical inflection points in a deep tech or semiconductor startup’s journey is when it secures a strategic engagement with a hyperscaler. While such partnerships can significantly enhance company valuation, understanding what types of contracts signal real traction—and how to communicate that value to investors—is essential for startup CEOs.
However, at the early stage, securing a fully binding commercial contract is rare, especially in capital-intensive verticals like AI accelerator chips or advanced interconnects, which often require tens to hundreds of millions of dollars just to reach production tape-out.
So, what do VCs look for in the early stages of companies?
Investors understand that full contracts are unlikely at the seed or Series A stage. Instead, they look for strong signals of customer intent and engagement.
Two key indicators stand out:
1. Depth of Technical Engagement
Investors evaluate whether the startup has entered deep, engineering-level discussions with hyperscaler customers. Such interactions provide tangible evidence that a large customer is seriously considering adoption, even in the absence of a signed contract.
2. Strategic Investors on the Cap Table
If the investment arm of a major hyperscaler or enterprise customer has participated in the startup’s funding round, it’s a powerful indicator of strategic alignment. Even a modest equity stake demonstrates buy-in and significantly de-risks the investment from a VC’s perspective.
Key Evaluation Criteria at Pre-Seed Stage
At the pre-seed (and seed) stages, it is rare for startups in this space to have complete products, formal customer contracts—even a Letter of Intent (LOI)—or production-level readiness. The capital intensity and technical complexity of the sector mean that hardware prototypes are often still under development.
Despite limited traction metrics, investors rely on a set of qualitative and strategic indicators to assess early-stage potential:
1. Team Credibility and Technical Track Record
Deep technical expertise is a non-negotiable requirement.
Founders must demonstrate past experience with tape-out processes, hardware commercialization, or fabrication environments.
Prior success in bringing technologies to market weighs more heavily than early prototypes.
2. Customer Signals and Reference Interactions
While formal contracts are uncommon, early customer conversations or reference checks can provide valuable insight.
Evidence that the startup is engaging with relevant industry players, even informally, helps validate the market need and potential alignment.
The Silent Gold Rush of AI-Powered Lab Automation
Five startups, one rumor: the lab is going fully autonomous.
