Software Ate the World. Now Services Are Eating Software.

For every dollar companies spend on software, they spend roughly six on services. That ratio, repeated across several independent market estimates, explains where venture capital is moving in 2026.

In 2011, the claim that software was eating the world sounded contrarian. It turned out right. A decade ago there were 15 SaaS unicorns; Foundation Capital now counts more than 400. Software won. The question now is what comes next, and the answer is services, though not in the traditional sense.

How much bigger is the services market than software?

When an investor at Sequoia says the next trillion-dollar company will be “a software company masquerading as a services firm,” many people picture consultants, hourly billing, and slide decks. That is the wrong mental model. The reason capital is moving is size.

Source	Estimate
NFX	~$5 trillion in knowledge work vs ~$230 billion in B2B software
a16z	~$6 trillion in white-collar services
Foundation Capital	$4.6 trillion opportunity
J.P. Morgan	$3 to $5 trillion
HFS Research	$1.5 trillion

The estimates differ. The conclusion does not: services dwarf software.

Salesforce illustrates the gap. It generates roughly $35 billion in annual revenue while companies spend around $1.1 trillion every year on sales and marketing salaries. Software captured a fraction of the value surrounding the work. The work itself stayed outside the product.

Until now.

What changed in how the work gets done?

AI increasingly performs tasks instead of assisting with them. Sequoia describes the transition as moving from copilots to autopilots. Greylock frames it as the agent completing most of the work while a human reviews the output. NFX reverses the acronym: software as a service becomes service as software.

The strategic consequence matters more than the labels. A company that sells software watches every improvement in the underlying models compress its differentiation. A company that sells outcomes watches every improvement make its delivery faster, cheaper, and more scalable. The technology stops being the product and becomes the engine behind it.

The model providers have reached the same conclusion. OpenAI launched a dedicated deployment company backed by more than $4 billion and acquired Tomoro to build a large forward-deployed engineering organization. Anthropic has made similar moves with partners including Blackstone, Hellman & Friedman, and Goldman Sachs, on the argument that enterprise demand exceeds any single delivery model. Both resemble the approach Palantir established years earlier: embed engineers with customers, solve the problem, then turn bespoke deployments into reusable products. Gravel roads become paved highways, as Palantir put it.

Investors are rewarding that model with software valuations. Sierra, Harvey, Abridge, EvenUp, and Decagon each own outcomes in a single vertical rather than selling seats, and most operate in regulated industries: law, healthcare, accounting, insurance.

What separates a deployment layer from services with AI on top?

The objections deserve a hearing. Better Tomorrow Ventures argues that AI makes services firms more efficient without turning them into software companies, because clients still pay for trust, credentials, and liability. Valere makes the same point in margin terms: a business that depends entirely on rented foundation models keeps its margin pinned to the spread between what customers pay and what the models cost. Software businesses run gross margins of 70 to 85 percent. Traditional professional services often run closer to 30 to 40.

The distinction is the cost curve, and one test captures it. Does serving the second customer cost materially less than serving the first? If yes, the business is accumulating proprietary advantage and starting to behave like software. If not, it is a services business with AI layered on top.

Three elements have to be present for the test to come out yes.

1. A technology stack that performs the work, rather than people assisted by tools.
2. Reusable product modules that make every deployment easier than the last. By the third or fourth engagement, software carries most of the delivery and humans supervise exceptions. That is when a services cost curve starts behaving like a software cost curve.
3. Value capture across the funnel: a free layer that identifies the problem, a paid layer that executes the work, and a product layer that compounds knowledge into reusable IP after the engagement ends.

We call that combination the deployment layer. A deployment layer sells outcomes instead of logins.

Why is quantum the clearest case?

The trend is global. One market makes it unusually clear.

AI is converging around a handful of dominant architectures. Quantum is diverging: multiple hardware modalities, sensing platforms, software stacks, post-quantum cryptography schemes, and industry-specific applications, with no standard toolchain connecting them. McKinsey estimates the sector attracted $12.6 billion in 2025. Very little of that capital has gone toward making the technology deployable inside enterprises.

For buyers, fragmentation is the bottleneck, not hardware. Building qubits is one problem. Integrating systems, managing migrations, producing evidence, and satisfying regulators is another, and that second problem exists today regardless of future hardware progress. DORA is already in force across EU financial institutions, and post-quantum cryptography migration is already on regulatory timetables. That second problem is deployment work, the kind our build engagements are scoped to deliver.

The more fragmented a market becomes, the more valuable its deployment layer becomes.

Software ate the world by replacing tools. Services are eating software by owning outcomes. The defining companies of the next decade will look like services businesses to their customers and operate like software companies underneath. The rest will be consulting firms with better marketing. In a market as fragmented as quantum, that difference is worth the most.