How much capital went to quantum hardware vs. software in 2025?

This reads like a lookup question. It is not. The record totals for 2025 are published; a hardware/software split of that capital is not — no major count releases one under a shared taxonomy. What can be answered with sources is: how much money moved, where the biggest checks went, and what the observable proxies say about the split. Status as of: August 2026.

How much money moved in 2025?

Three counts, three scopes, three numbers:

Inside McKinsey's $12.6B: 90% went to quantum computing startups (the rest to sensing and communication), 97% was private capital (2024: 67%), and roughly 60% of the total sat in the top ten deals. MIT's count adds the calibration the headlines skip: even at its 2024 high-water mark, all of quantum was under 1% of global venture funding, and QIR labels its own data "indicative rather than complete."

These are not contradictions to resolve. They are scopes to read — all-quantum-tech vs. private-VC-only vs. publicly-announced-only — and the honest headline is common to all three: record capital, heavily concentrated.

Where did the biggest checks go?

Concentration makes the top of the table the story. Sixty percent of $12.6B in ten deals means the character of those deals is the character of the year:

Every entry above the last line is hardware or hardware-anchored full-stack. The one software entry is in the table because it is the largest quantum-software round in the field's history — and at $110M it is about a ninth of PsiQuantum's single check, smaller than the gap between Quantinuum's round and IQM's.

So what is the hardware/software split?

The honest answer, at three levels of evidence:

By round composition (2025). The disclosed top of the market is hardware. Three hardware rounds of ≥$300M closed within eight days of September 2025 alone; no software round in history has reached $300M.

By the one taxonomy that publishes both categories (2024). MIT QIR logs $621M for quantum software against $1.59B for quantum computing rounds — software at roughly 28% of that pair. That is the most software-favorable number any count yields, it is a 2024 figure, and it depends entirely on taxonomy: full-stack companies (which raise the largest rounds) sit on the "computing" side.

By record sizes. Largest software round ever: $110M. Largest hardware round of the same year: $1,000M. Ratio: ~9×.

Under every count, most of the capital is building machines. How small the software share is — a few percent by top-round composition, ~28% under QIR's 2024 taxonomy — depends on who counts and how. Nobody publishes a clean 2025 split. That absence is itself the finding: the field measures its machines' funding better than it measures the funding of the layer that would prove the machines useful.

Why the split matters

The software layer is where the deciding questions live: is this machine worth using, on which problem, against which classical baseline, at what cost per answer. Capital allocation currently prices the existence of the machines far above the evidence layer that would price their usefulness. That is an observation about incentives, not about any vendor — every company named above discloses its rounds honestly; it is the aggregate that skews. For the multi-year trajectory of the money itself, see the VC map.

Rosetta holds no proprietary funding data. Every figure above comes from the cited public counts, with their stated scopes.

What we don't know

• A clean 2025 hardware/software split under a shared taxonomy. No source publishes one; we will not invent it.
• Whether Classiq's record round marks a re-rating of the software layer or an outlier. One round is one data point.
• How much of the $56.7B in announced public commitments (QED-C) converts to deployed capital, and on what schedule.
• China's true totals. Every Western count flags the gap; McKinsey's 6.3× multiple and QED-C's 2× do not resolve it.
• Full-year 2026. The year is running; this map gets re-dated when the counts land.

Rosetta Q publishes verdicts with reproducible raw data. This is educational content, not a product claim. Figures as reported by the cited sources; methodologies differ and are flagged where they do. Dated 2026-08-02.