In 2025, venture capital “came back” – at least on the surface.

Global startup funding climbed back toward pre‑slowdown levels, with roughly US$100–120 billion deployed in Q3 alone and nearly half of that capital flowing into AI companies. But underneath the headline numbers, capital was highly concentrated: a small set of mega‑rounds, mostly in AI and infrastructure, captured an outsized share of the dollars while overall deal counts stayed muted.

From our vantage point at Celesta Capital, this is not just another cycle. It’s a once‑in‑a‑generation reset of the technology stack.

And 2026 won’t be “just another year of AI.” It will be the year the real economy catches up with the hype – and the year we begin to see real winners and meaningful consolidation.

Since 2013, our funds have focused on deep tech – with an emphasis on semiconductors and intelligent systems, technology-enabled industry transformation, and bio‑convergence – and today the opportunities surrounding these sectors are more attractive than any time in recent memory. 

Below are some perspectives on current areas of opportunity we’re tracking as we look to 2026 and the year to come.

‍

1. Deep tech’s share of venture capital will keep climbing.

Deep tech is likely to command an increasingly large slice of global venture flows as AI continues to drive a full‑scale rearchitecting of the technology landscape—from compute and power delivery to data infrastructure and bio‑convergence.

Deep tech already accounts for roughly 20% of global VC, up from about 10% a decade ago, according to BCG, and some recent estimates suggest it may be closer to one‑third of all venture funding in 2024. Much of this is being driven by the explosion of AI hardware: forecasts see AI chipsets alone reaching $323 – $453 billion USD by 2030, while total semiconductor revenue could hit $1 trillion USD as early as 2030. On top of that, McKinsey estimates roughly $5.2 trillion of data‑center investment will be needed by 2030 just to meet AI demand.

Those numbers demonstrate what we all understand implicitly; that capital cannot remain concentrated in software and model layers. It must flow into fueling innovation within the physical and infrastructure layers which will enable high performance AI applications at scale and unlock the AI-driven industry transformation that has been promised.

This aligns with Celesta’s long‑standing focus on semiconductors, intelligent systems, and next‑generation infrastructure, sectors that sit at the heart of this rebuild. Looking forward, the distinction between “AI investing” and “deep tech investing” will continue to blur. AI will simply be the demand driver for an enormous amount of power electronics, packaging, networking, industrial automation, and bio‑convergence innovation.

‍

2. Hardware and IP drive increasingly strong returns.

While Silicon Valley was built on deep tech with the likes of Fairchild Semiconductor and early computing companies, it fell out of favor for several years as many VC investors were allured by the software market.

This shift wasn’t surprising. For sector-agnostic investors, it’s often easier to comprehend and underwrite software than to evaluate the deeper engineering complexities or scientific feasibility of deep tech businesses. For the past decade, many funds have built their strategies around these capital‑light, software‑only approaches. That made sense when incremental cloud software could grow fast on relatively stable infrastructure. Software companies can require less upfront capital to develop an initial product, even if total lifetime capital is similar once go‑to‑market costs are included, which typically is much larger for software. Software also can lend itself to earlier exits; deep tech demands methodical business building.

But as AI reshapes the technology stack, significant high‑growth, defensible opportunities are increasingly in hardware, chip design, advanced packaging, and systems – areas many sector-agnostic software investors often avoid. Today, hardware businesses are seeing the strongest gross margins they’ve had in years, up more than 5% across the semiconductor industry over the past year. Rising valuations in hardware and deep tech are substantiated by strong revenue potential forecast over the next decade. Meanwhile, many software companies are seeing the erosion of one of their greatest historical advantages as they experience margin compression – down nearly 10 percent year-over-year by one report’s estimate – driven by saturated markets and rising costs to serve as they build atop AI models.

Looking ahead, funds that cannot evaluate process technology, yields, chiplet interconnects, or power‑delivery roadmaps may miss out on highly attractive deals while the market increasingly rewards investors and founders who are comfortable with silicon, tools, and systems design, not just application‑layer software.

‍

3. In AI, power is becoming the binding constraint.

The loudest story in AI has been about GPU shortages. By 2026, we expect the true bottleneck to be how much power data centers can safely deliver and cool, not how many accelerators they can buy.

There is a critical need for energy‑efficiency solutions as AI drives a once‑in‑a‑lifetime rebuild of the world’s compute infrastructure. Data centers already consume 1.5% of global power and that is expected to double by 2030. As power per AI chip climbs beyond ~1,200 watts, everything from substation capacity to power‑delivery chiplets and cooling technologies becomes the limiting factor.

That’s why we’re backing companies like PowerLattice, which is building integrated‑voltage‑regulation chiplets specifically for high‑power AI processors, and Auradine and Tensordyne, which are rethinking systems architecture around energy use. For cloud providers, enterprises and governments, the message is clear: the next competitive edge in AI will come from watts and thermal envelopes, not just model size. For investors and founders, attention is shifting towards hardware, where physics sets hard limits and competitive moats can be built and sustained.

‍

4. Packaging and interconnect become the new fabs.

For decades, semiconductor value creation was mostly about shrinking transistors. As Moore’s Law slows, how chips are connected and packaged matters as much as the transistors themselves. Now, pressure from AI is making this problem even worse.

As AI and high‑performance systems outgrow monolithic dies, packaging and interconnect are becoming the new scaling architecture. Equipment makers, hyperscalers, and national semiconductor programs are already recalibrating around chiplets and 2.5D/3D packaging, a market expected to reach $55 billion USD by 2030. For founders, that opens room for new IP providers, EDA tools, and packaging‑centric fabs; for investors and strategics, it’s a reminder that some of the most durable value in AI may sit in the “plumbing” between dies rather than in any individual accelerator.

We’ve long held conviction that interconnects were growing as a significant bottleneck to system performance. Our portfolio company Credo is one example of a company now thriving in this space, with a market cap over $25 billion USD. Eliyan is another good illustration, which enables ultra‑high‑bandwidth, energy‑efficient chiplet interconnect for advanced multi‑die packages and has already taped out on Samsung’s 4nm node.

‍

5. Industrial AI outgrows consumer‑facing AI.

Much attention around AI still focuses on foundation models and chatbots. We believe the next opportunity for value creation is in the industrial and enterprise space: AI that quietly improves yield, uptime, and throughput in factories, plants, and logistics networks. The global industrial AI market is expected to grow at a 23% CAGR to $154 billion USD by 2030.

There is a significant opportunity for application‑specific solutions in sectors such as construction, defense, and customer service. Examples in our portfolio include:

• ‍Slate, which uses data to improve construction productivity‍
• Crescendo, which is transforming call centers with AI‍
• Normal Computing, developing AI software to accelerate complex hardware design and engineering with zero defects

These systems don’t always have the glamorous appeal of consumer apps, but they are a lever for real productivity gains and represent sticky, high‑margin software businesses whose value is tied to high-ROI outcomes.

‍

6. Dual‑use deep tech continues its shift from niche to mainstream.

For many years, defense was a difficult category for venture capital. That’s fast changing fast as governments recognize these firms are essential to sovereignty and resilience – global military expenditures are growing at their highest rates since the Cold War. For private industry these technologies can deliver cutting‑edge sensing, imaging, and networking tech that can be repurposed for industrial and civil markets. We expect dual‑use companies to become some of the most attractive growth stories in 2026.

These companies have a unique to develop competitive advantages around structurally strong demand, long product lifecycles, and diversified revenue between defense and commercial buyers. Companies like Percipient.ai (computer vision intelligence for security missions), ideaForge (aerial robotics), and Tonbo Imaging (electro‑optics for defense and infrastructure) are good examples of companies Celesta has invested in in this space. 

‍