Deep tech has entered a paradoxical moment. Capital is pouring into the sector at unprecedented volume, yet much of it is clustering within a narrow slice of the market—primarily AI infrastructure—creating structural imbalances. While investors chase capacity-intensive compute, entire domains of fundamental innovation remain undervalued and undercapitalized. This misallocation is not a temporary distortion but a defining feature of the current cycle.
Alongside this concentration, a second major shift is reshaping the venture ecosystem: the rise of secondary markets as a viable, scaled liquidity infrastructure. These markets are no longer tactical tools for insiders; they have become an essential layer in the capital stack, rivaling traditional exits. Their growth is tightly linked to extended private-company lifecycles and the growing mismatch between available capital and investable opportunities.
Celesta Capital’s track record—110 investments and 43 exits—offers a window into what succeeds when innovation requires heavy engineering, long development timelines, and high precision in market selection. The firm’s results demonstrate that disciplined, fundamentals-first investing can outperform even in saturated markets.
The central challenge now is allocation. With AI infrastructure absorbing disproportionate attention, investors must decide how to balance momentum-driven bets against overlooked opportunities in sectors where the economics are stronger, competition is thinner, and value creation is grounded in fundamentals rather than hype.
The rise of secondary markets is the product of several converging forces. Private companies are staying private longer, fueled by massive capital availability that reduces pressure to exit. At the same time, the supply of quality deals has not kept pace with capital inflows, forcing investors to seek liquidity through nontraditional channels. Secondary markets emerged as the system’s relief valve, enabling investors to redistribute risk and recycle capital without waiting for IPOs or acquisitions.
Historically, venture investors accepted a binary exit expectation: returns would come from strategic buyers or public markets. That model is breaking down. As companies accumulate large private valuations and extend their timelines, LPs demand earlier liquidity, and GPs require ways to demonstrate tangible performance. Secondary transactions now occupy a structural role, providing interim liquidity that reshapes fund design, pacing, and risk appetite.
Deep tech adds further complexity. Traditional secondary buyers focused on predictable cash flow, prioritizing businesses with stable revenue and operational maturity. The new wave of AI-era secondaries is different. Buyers today are often betting on strategic position—market share, technology diffusion, and ecosystem advantages—even when revenues are nascent. The expectation is not near-term profitability but potential dominance in a winner-take-all market structure.
This shift changes portfolio construction for investors in capital-intensive innovation. Earlier liquidity opportunities reduce the duration risk typically associated with deep tech, allowing funds to pursue larger, more ambitious bets while keeping DPI within acceptable ranges. But this also intensifies the tension between IRR and distribution timing. LPs increasingly prioritize capital returned over notional valuation, pushing funds to identify secondary paths earlier in a company’s lifecycle.
For deep tech investors, understanding the secondary market is no longer optional. It is a critical component of value realization, risk management, and long-term portfolio strategy.
The rapid acceleration of AI infrastructure investment has created an environment in which supply is growing faster than demand. Capital is flowing into GPU clusters, data centers, and semiconductor designs at a pace reminiscent of previous technology booms. The cost of innovation has collapsed—enabled by cloud resources, open-source tooling, and a new generation of developer productivity—but this masks underlying economic instability.
While startups can build powerful AI systems with minimal headcount, the infrastructure supporting those applications remains massively capital-intensive. This mismatch leads to a familiar cycle: capacity is built on optimistic projections of future usage, and when demand grows more slowly than expected, the burden falls hardest on mid-tier players without deep balance sheets.
Hyperscalers and cash-rich neocloud platforms can sustain overcapacity as a strategic investment. Their timelines are measured in decades, and their capital allocation models tolerate periods of inefficiency. Smaller infrastructure companies lack that buffer. They face volatility driven by hyperscaler purchasing cycles, rapid technological obsolescence, and shifting customer expectations. The result is a whiplash effect that can wipe out promising players despite strong underlying technology.
Historical parallels are instructive. Telecom overbuilds in the early 2000s followed a similar pattern: exuberant investment outpaced real-world demand, and the subsequent correction reshaped the industry for years. Cleantech 1.0 saw a comparable shift when capital intensity collided with slower-than-expected adoption. In both cases, mean reversion eventually restored equilibrium—but not before significant losses for investors caught in the wrong layers of the value chain.
AI infrastructure is heading toward a similar reckoning. The timing is impossible to predict, but the dynamics are clear. Investors should prepare for multiple scenarios, from a gentle rebalancing to a faster and sharper correction. The key is to distinguish between segments with sustainable demand signals and those propped up by momentum-driven capital flows.
To navigate this landscape, investors need a sector framework that cuts through hype and highlights where real value is created. Deep tech can be organized into three investment sandboxes, each defined by its own economics, maturity, and competitive intensity.
Sandbox 1: Physical Infrastructure Layer. This includes hardware systems, data center design, and semiconductor technologies. These areas are foundational but uneven. Some segments, such as high-performance compute and networking fabrics, face significant oversupply risk. Others—specialized accelerators, energy-efficient architectures, and next-generation interconnects—may remain structurally constrained. Investors must differentiate between commodity capacity and scarce capability.
Sandbox 2: Intelligence Applications Layer. The applications layer includes vertical AI tools, automation systems, and domain-specific solutions across healthcare, logistics, financial services, and industrial ecosystems. Adoption cycles are accelerating, lowering time-to-revenue and compressing the path to validation. Competition is intense, but business models are clearer and capital efficiency is higher than infrastructure plays. For many investors, this sandbox provides the most balanced risk-reward profile.
Sandbox 3: Bio-Convergence. At the intersection of biology, robotics, and precision hardware, this sandbox encompasses surgical systems, diagnostics, miniaturized devices, and biologically integrated instrumentation. It remains underfunded relative to its potential, partly because it requires domain expertise and regulatory familiarity. Yet the technical moats are strong, the competition for capital is lower, and the pathways to defensibility are more robust.
Capital requirements vary sharply across these sandboxes. Hardware infrastructure demands patient capital and long cycles. AI applications require less upfront investment but heavier emphasis on customer acquisition and product iteration. Bio-convergence companies often need early scientific validation but can scale efficiently once product-market fit is established.
Celesta Capital’s exits reflect this distribution: Credo in semiconductor connectivity, Habana Labs in AI processing, Innovium in high-performance networking, IdeaForge in drone systems, and Berkeley Lights in digital biology. The pattern reveals that opportunities span all three sandboxes when investors apply disciplined selection and focus on the intersection of technical differentiation and market pull.
The valuation reset in SaaS offers a preview of what lies ahead for AI and deep tech. Multiples that once reached 20x ARR have normalized to the 5–7x range as markets recalibrate around profitability, capital efficiency, and predictable growth. AI valuations, inflated by infrastructure-driven excitement, are likely to undergo similar compression.
A return to fundamentals means investors must re-anchor assessments around revenue visibility, gross margin structure, and a clear path to profitability. Breakthrough technology remains essential, but commercial traction will determine which companies scale and which stall as science projects.
Faster product cycles and shorter time-to-revenue present a dual dynamic. They enable earlier validation and earlier course correction, reducing long-term risk. But they also expose weak business models quickly, requiring investors to tighten diligence and stress-test assumptions more rigorously.
Discipline thus becomes a competitive advantage. In an overheated market, avoiding overpriced deals is as important as discovering undervalued ones. Investors who emphasize fundamentals will not only preserve capital but also position themselves for high-quality opportunities as valuations normalize.
The coming cycle will reward investors who align strategy with structural shifts across liquidity, sector diversification, and valuation discipline. Each stakeholder group faces distinct strategic adjustments.
For LPs: Assess whether fund managers demonstrate restraint in pricing, thoughtful sector allocation, and clear access to secondary liquidity. GPs should show evidence of navigating both momentum markets and complex deep tech arenas.
For GPs: Balance exposure between AI infrastructure momentum plays and contrarian bets in bio-convergence or applied AI. Maintain consistent diligence standards across both. Build portfolios that can tolerate extended timelines but also capture early liquidity opportunities through secondaries.
For founders: Shorter time-to-revenue is now expected. Capital efficiency and monetization clarity differentiate companies that secure funding from those that struggle. Founders should view liquidity planning as strategic, integrating secondary options into broader growth planning.
Across all groups, scenario planning is essential. A soft landing might see gradual normalization of capacity and valuations. A hard correction could reset the infrastructure layer rapidly, creating opportunities for well-capitalized investors and challenges for those overexposed to overbuilt segments.
Discipline, diversification, and fundamentals-driven decision-making will define the next era of deep tech investing. In a market shaped by both excess and opportunity, investors who ground their strategies in structural understanding will be best positioned to capture long-term value.
