Look across this year's largest growth-stage checks and a clear pecking order emerges: physical infrastructure and scarce capacity plays -- AI compute (Crusoe's reported ~$3 billion raise), energy (National Grid's $1.75 billion Joulent investment), even crypto infrastructure (a16z crypto's $2.2 billion fifth fund) -- are commanding capital at a scale that dwarfs typical application-layer software rounds by an order of magnitude or more.
The pattern isn't confined to AI. It holds across categories that on the surface look unrelated: AI data centers, physical power generation, and crypto/blockchain infrastructure are all pulling growth-stage mega-checks, while pure application-layer software -- even well-regarded vertical SaaS and AI-application companies -- is raising at meaningfully smaller scale unless it has a direct tie to underlying infrastructure scarcity or genuinely defensible proprietary data.
This compounds a concentration dynamic already visible at the very top of the market: H1 2026's headline $510 billion global VC funding total included OpenAI and Anthropic capturing 43% of all dollars raised globally, meaning the other 57% is being split across every other startup, in every other sector, worldwide -- and within that remaining pool, physical infrastructure plays are again capturing disproportionate share relative to application-layer software.
The logic tracks scarcity economics directly: gigawatts of power, advanced-node chip manufacturing capacity, and (per a16z crypto and Haun Ventures' back-to-back billion-dollar-plus fund closes) even institutional-grade crypto infrastructure are all genuinely rate-limited resources that can't be built or acquired quickly regardless of capital available -- while software, even excellent software, can typically be built and scaled without the same multi-year physical-capacity constraints, making it a comparatively less scarcity-driven investment thesis right now.
Compared to the 2010s software-eating-the-world era, when SaaS and application-layer companies commanded premium growth-stage valuations almost by default, this represents a meaningful thesis inversion: physical infrastructure, once considered a slower, more capital-intensive and less venture-attractive category, is now where growth investors are willing to write their largest, highest-conviction checks.
For growth-stage software founders, the implication is direct: differentiation on product alone is increasingly insufficient to command top-tier growth-stage attention and pricing; founders need either a genuinely rare data or distribution moat, or a structural tie to one of the scarce physical resources (compute, power, chips) currently commanding premium capital.
For infrastructure-focused GPs and LPs, this is validation that the physical-infrastructure thesis has real staying power beyond a single hot quarter -- it's now spanned AI compute, energy and crypto infrastructure simultaneously, suggesting a structural capital reallocation rather than a narrow, category-specific trend.
The bear case: physical infrastructure investments carry meaningfully different risk profiles than software -- longer payback periods, higher capital intensity, and direct exposure to the AI-bubble risk the Bank for International Settlements just flagged regarding Oracle's debt-financed Stargate commitments; a broad correction in AI infrastructure demand would hit this cohort much harder than a typical software valuation reset.
What to watch: whether application-layer software valuations continue compressing relative to infrastructure plays through the second half of 2026, and whether any of this year's infrastructure megadeals show early signs of the demand assumptions embedded in their valuations proving too aggressive.