Eclipse Won $2.5 Billion Betting on What Nobody Wanted to Touch
When Lior Susan founded Eclipse Ventures in 2015, the prevailing logic in Silicon Valley was simple: software scales without factories, without inventory, and without workers. Software-as-a-service companies were capturing the attention of the best funds and the best engineers. Betting on semiconductors, industrial robotics, or physical computational infrastructure was, at best, an eccentricity. At worst, a thesis error.
Susan described it without embellishment at a recent StrictlyVC event in San Francisco: the early years were "quite lonely." Eclipse arrived in 2016 with an investment of $6.5 million in Cerebras Systems, a chip startup designed for artificial intelligence workloads that at the time did not exist at commercial scale. Ten years later, that bet grew to $147 million in total invested capital and generated $2.5 billion in returns when Cerebras went public in May 2026 at a price of $185 per share, in an IPO that raised an additional $5.5 billion. The multiple was 17 times the invested capital.
The number is extraordinary. But the question that matters for any value systems analyst is not how much Eclipse made. The question is what incentive structure made that return possible, and whether that structure holds or whether we are already observing the peak moment before the readjustment.
Why Hardware Returned to Being the Scarce Asset
The fall from grace of software as a purely defensible asset was not gradual. It was abrupt and has a precise mechanic. The emergence of code generation tools such as Anthropic's Claude Code or the most recent models from OpenAI put on the table a possibility that markets took a couple of years to digest: if any company can generate custom software at a fraction of the previous cost, the value of software packaged in standard licenses compresses. Not all at once, but in a sustained manner.
Susan synthesized it with a phrase that circulated widely after the event: "The competitive advantage in software disappeared. You can generate code for practically anything." And then he added the part that matters: "What you cannot do with generated code is manufacture silicon wafers. For that you need machines, clean rooms, and supply chains that take decades to build."
That gap is the asset. Not in the abstract sense of "barrier to entry," but in the concrete operational sense: the human capital, regulatory permits, supply agreements, and physical infrastructure required to produce advanced hardware cannot be replicated in an eighteen-month funding cycle. Physical scarcity generates a value premium that software, today, cannot claim with the same credibility.
Public markets are already reflecting this reasoning. Susan pointed out that TSMC and Micron shares reached all-time highs in the months leading up to Cerebras's IPO. At the same time, a considerable portion of enterprise software values suffered declines during the first quarter of 2026, precisely because investors began to discount the possibility that large companies would reduce their SaaS licenses and build proprietary tools with language models. It is a redistribution of value between layers of the technology stack: hardware rises, generic software applications fall.
The Portfolio Mechanics and Where the Risk Actually Accumulates
The Cerebras return is the headline, but the pattern Eclipse has built in parallel deserves attention on its own merits. According to data Susan presented, companies in Eclipse's portfolio raised approximately $15 billion from external investors during 2025, and in the first quarter of 2026 alone had already raised an additional $4.5 billion. To calibrate that figure: in Eclipse's first eight years as a firm, the cumulative total raised in external rounds by its portfolio companies did not exceed $4 billion.
That scale jump is not just a signal of success; it is also a signal of structural tension. When capital arrives in quantities of that magnitude and at that speed in sectors with long development cycles, pressure on maturation timelines intensifies. The four portfolio companies that raised massive recent rounds have very distinct profiles from one another: Wayve (autonomous driving, $1.2 billion with Nvidia, Uber, and three car manufacturers), True Anomaly (defense and space, $650 million), Bedrock Robotics ($270 million), and Oxide Computer ($200 million). Eclipse entered all of them as a Series A investor.
Eclipse's model has a clear internal coherence: enter early, sustain long capital cycles, and capture the return when the late market arrives with large checks. Within that scheme, dilution risk is real but manageable if conviction in the thesis is sustained. The risk that is harder to manage is that of timing: if the late market's enthusiasm for physical hardware arrives in a short window and cools before these companies reach revenue generation at the scale that justifies their valuations, the model distributes losses toward later-stage investors, not toward Eclipse.
This is not an accusation. It is the standard mechanic of early-risk investing. But the pattern deserves to be made visible: Eclipse wins first, later private capital investors and eventually public shareholders assume the majority of the commercial execution risk.
The Five Factors Susan Aligned and What Is Missing from That Equation
The thesis Susan articulated in San Francisco goes beyond the sectoral bet. He described five conditions that, according to him, are converging for the first time in American industrial history: technology (AI as an enabler of physical hardware), capital (record flows toward infrastructure sectors), customer demand (industry and government as sustained buyers), talent (engineers migrating from software toward robotics, semiconductors, and space), and policy (subsidies and favorable regulation from the federal government).
The comparison to Henry Ford and Andrew Carnegie was deliberate. Both operated in moments where physical infrastructure was the bottleneck of the economy and where capital, technology, and policy aligned to accelerate a generational transformation. The analogy has historical force, but it also carries an implicit warning that Susan did not develop: both Ford and Carnegie built empires on a highly concentrated distribution of value. The question of what percentage of that value ended up in the hands of workers, suppliers, and the communities where they operated remains one of the most documented debates in economic history.
Susan's analysis of the five factors is strategically sound and empirically backed by capital flows. Where the equation has an unanswered variable is in the final distribution among the actors of the system: first-tier hardware manufacturers, their supply chains in emerging markets, the infrastructure operators that depend on those chips, and the displaced workers in sectors that automation absorbs. The fact that the five factors are aligned for founders and early investors does not imply that they are aligned for all participants in the system.
The Moment of Validation Is Not the End of the Cycle
The Cerebras IPO and Eclipse's returns are not the closing of a thesis; they are its most visible market test so far. The distinction matters because the capital now entering robotics, space defense, infrastructure computing, and semiconductors will reach its economic maturity in a cycle that has not yet completed its first full turn.
What Eclipse has built over a decade is a privileged entry position in sectors where physical scarcity remains structural. That position is not lost with Cerebras's success; on the contrary, it strengthens it for the fundraising rounds and co-investors who will now compete to be in the same portfolio. The risk that does increase with success is the pressure on earlier-stage companies to justify valuations inflated by late-market enthusiasm, before their business models have demonstrated the cash flow generation necessary to sustain them.
The value distribution Eclipse has achieved with Cerebras is, within the parameters of venture capital, a legitimate and well-executed result of a conviction maintained for a decade. The sustainability of that model going forward depends on whether the companies that are now raising hundreds of millions in late rounds manage to translate that injection into operational returns before the cycle of enthusiasm for hardware stabilizes. If they succeed, the Eclipse thesis will be validated in its entire architecture. If they do not, Cerebras's returns will be the peak of a curve, not the beginning of a sustained upward slope.
