From Access to Accountability: The Year the Channel's Margin Was Repriced

Through 2025 and into 2026 the software category did something it had never done in its recorded history: it fell hard while the broader market held. Roughly $2 trillion in software market capitalization has been erased since early 2026, and for the first time on record software now trades at a discount to the S&P 500. The iShares Expanded Tech-Software ETF (IGV) fell about 24% in the first quarter alone, its steepest quarterly decline since the fourth quarter of 2008, and the median software EV/Revenue multiple compressed from roughly 7x entering 2025 to about 3.4x by March 2026.

Every prior software selloff of comparable size tracked the Nasdaq in lockstep. This one did not, and that divergence is the whole signal. A macro correction repriced everything together; this repriced software specifically, because investors revised down how much per-seat software revenue will exist in 2030 once an agent can do the work of several seats. The market is not forecasting a recession in software. It is forecasting a different software.

The same logic reached the services tier in plain view. In June 2026 Accenture fell roughly 18% in a single session, a record one-day drop, after cutting full-year revenue-growth guidance to 3-4% on consulting growth of about 1%. Cognizant, Wipro, Capgemini, and IBM repriced alongside it. Two forces hit one business model at once: demand softened, and the market decided AI is starting to do the work these firms used to bill by the hour. The second force is structural, and it does not stop at the largest integrators. It runs straight down-market into the SMB-to-Corporate accounts the channel serves.

For decades a channel partner's margin rested on a single fact: intelligence was scarce. Expertise was expensive, slow to scale, and hard to replicate, so configuration, advisory, and the knowledge markup justified the rate card. AI does not eliminate intelligence. It eliminates exclusivity over it, and exclusivity is what set the price.

The price signal is unambiguous. Per-token model prices have fallen on the order of 99.7% since the GPT-3 era, and the average cost of a million tokens across major providers dropped roughly from $10 to $2.50 in a single year. The lesson customers are drawing is not "use the cheap model for everything." It is more dangerous for a channel: route each task to the cheapest model that can handle it and reserve the expensive frontier model, which can still cost 20 to 30 times more per token, for the narrow slice of work that genuinely requires it. When intelligence is a commodity routed by price, the markup that used to sit on top of it has nowhere to hide.

TSIA gives the trap a name. In its 2026 analysis of technology services, the core problem of the AI era is the "Cannibalization Dilemma": under per-seat pricing, every improvement in automation becomes a revenue leak. The better the AI performs, the fewer seats the customer needs and the less predictable the renewal becomes. A partner whose economics are tied to seats is, in effect, short its own product roadmap. It loses revenue precisely as the technology it sells gets better.

This bites hardest precisely in the SMB-to-Corporate band. These customers were never paying for frontier-grade sophistication; they were paying for access to expertise they could not hire. The moment that expertise is available at the price of an API call, the markup a partner placed on it collapses, and if producing the work approaches zero marginal cost, the customer can increasingly reproduce it themselves. None of this is anyone acting against the channel. The platforms are driving the cost of intelligence down because that is exactly what their customers and shareholders want. The casualty, the margin built on scarcity, is incidental. Reading it as an attack leads to the wrong response: defending a disappearing layer instead of moving to the one opening up.

Two movements are redrawing the chain in parallel. First, the platforms are bundling AI into the products customers already buy, at near-zero marginal cost, absorbing the commodity layers of the old channel model. Second, the model labs have stood up their own deployment arms: OpenAI's "Deployment Company" raised more than $4 billion (anchored by TPG, at a reported ~$14 billion valuation), and Anthropic launched an enterprise-services venture valued near $1.5 billion with Blackstone, Hellman & Friedman, and Goldman Sachs. That is roughly $5.5 billion directed at the deployment and consulting layer in a matter of weeks.

It is tempting to read this as the platforms turning on their partners. That framing is wrong and unhelpful. A platform owner that can deliver more of the value chain at scale has an obligation to its shareholders to do exactly that; it is competent platform economics, not betrayal. The more useful observation is what the same scale logic pushes platforms away from: the messy, low-volume, high-liability work. Jurisdiction-specific compliance, the non-obvious process logic of a particular vertical, the last mile of a mid-market deployment where someone has to answer for the result. Platforms build platforms, not professions. Their own AI products create deployment and accountability gaps they have neither the structure nor the appetite to fill. In the chain now forming, they need a channel to occupy that layer, arguably more than before.

Underneath the model debate, the market has reorganized into competing infrastructure alliances, and a partner's economics now depend partly on which side of the compute-cost line its customers sit. Anthropic's ~$200 billion commitment to Google Cloud compute and TPUs, with Google investing up to $40 billion into Anthropic at a ~$380 billion valuation, sits alongside Amazon's ~$100 billion AWS commitment and the original Microsoft and OpenAI axis. Purpose-built silicon such as Google's TPU line is estimated to run materially cheaper per token than comparable GPUs at scale, a structural cost gap, not a marginal one, though the precise figure is a directional estimate rather than a published number.

The practical implication for a partner is not to pick a winner. It is to stay model-sovereign: decouple model allegiance from infrastructure allegiance, build genuine multi-model competency across Claude, GPT, and Gemini, and treat the cloud beneath as a deployment choice rather than an identity. Building a practice around which ecosystem wins is a bet on something the partner does not control.

The repricing assumes AI is already doing the work. The operating reality is more uneven, and that gap is the channel's opening. About 72% of enterprises now run at least one AI workload in production, yet only around 6% qualify as high performers attributing 5% or more of EBIT to AI, and only about 39% report EBIT impact at the enterprise level at all. Adoption is near-universal; realized financial impact is rare. That distance between deploying AI and getting paid outcomes from it is not a transition problem to wait out. It is the commercial space a channel partner is uniquely positioned to fill, because closing it requires exactly the proximity, domain depth, and willingness to be accountable for a result that platforms and labs are structured to avoid.

The direction of travel is not in doubt. IDC projects AI-related IT spending to grow at roughly a 32% compound rate over five years, with agentic systems reaching about half of all AI spending by 2029; it expects some 40% of Global 2000 roles to involve working with AI agents as soon as 2026, and around 45% of organizations to orchestrate agents at scale by 2030. Tellingly, roughly 70% of large-company CEOs say they are pursuing AI's return as revenue growth without adding headcount, which is the per-seat compression thesis stated from the buyer's side. The agents are coming at scale; the unmet need is someone to make them produce a dependable outcome inside a specific business.

The forecast would be incomplete without the capital backdrop, because it shapes who buys and sells partner businesses in 2026. Private equity is holding the largest stock of unsold software assets on record: roughly 32,000 portfolio companies representing about $3.8 trillion in unrealized value, average holding periods stretched to around seven years, and distributions below 15% of NAV for four consecutive years. Many of those companies were underwritten against a pre-generative-AI world, growth, leverage, and competitive assumptions that no longer hold. The AI repricing arrived inside an exit window that was already closed.

For the channel this cuts two ways. Sponsors under pressure to deploy elevated dry powder will keep consolidating partner and ISV businesses, producing fewer, larger, better-capitalized partners with broader offering sets. But the assets that command strong valuations in this environment are precisely the ones that made the transition this forecast describes: workflow ownership, vertical depth, outcome-linked economics. "Acquired five years ago against a seat-based growth story" is now a discount, not a premium. A partner positioning to sell into this market is really being asked one question by buyers: do you own a customer outcome, or do you resell access to one?

PitchBook's read on the same reckoning is instructive, because software is so central to private-equity portfolios, roughly 18% of US PE deal value in 2025 and about a quarter over the past five years. Deal-making has not stopped; it has become selective. Take-private value rose to $64.3 billion in the first quarter of 2026, up more than 50% year-over-year, and median buyout multiples held around 10x EV/EBITDA, but the capital is concentrating on a specific profile. PitchBook is explicit about the sort: the winners are defined by proprietary data large models cannot replicate and mission-critical workflows with demonstrable automation value; the losers are seat-based models and point solutions with replicable functionality. That is the same dividing line this forecast draws, written in the language of the people deciding what a partner business is worth.

Our partner model, carried forward from prior Channel Forecasts, still frames the magnitude well: a conservative three-year scenario in which AI-driven agents reduce a partner's legacy services billings by roughly 30% and human-delivered support revenue by as much as 60% implies workforce reductions on the order of half of services-delivery headcount simply to hold gross margin, unless the revenue is replaced with outcome-priced, IP-wrapped offerings. The early evidence that replacement is possible is real: roughly 41% of MSP revenue growth is now attributed to AI-related services, versus about 33% from traditional seat growth, and building proprietary IP and vertical solutions has become the number one stated partner priority for 2026.

The clearest signal that this is the intended direction comes from the platforms' own incentive design. Microsoft's 2026 partner-program changes now require an accredited software designation to access Azure IP co-sell, raised AI incentives by roughly 50%, and increased Azure outcome-based incentives by about 70% year-over-year, while framing its marketplace as a ~$300 billion opportunity by 2030. AWS has added agentic-AI partner categories with extra marketing funds, enabled variable "as-delivered" marketplace payments for professional services, and is explicitly shifting partner standing "from validation to value realization." The platforms are not merely tolerating an outcome-based channel; they are funding the move toward it. This is the "platform as ally" point made concrete: a partner that delivers outcomes drives adoption of the platform's now-inexpensive capability into accounts it cannot service intimately, and the incentive structures are being rewritten to pay for exactly that. TSIA puts the same conclusion plainly: AI does not eliminate services, it makes them mandatory.

Across every partner type the question is the same: who owns the customer workflow consequence? If a solution fails and the customer absorbs the downside, the vendor is optional. If the vendor absorbs the downside, the relationship resembles infrastructure. Commoditization is coming for the markup; it is not coming for accountability. When the intelligence in an offer is racing toward free, accountability for the result is what the customer is still paying for, and it is the layer platforms are content to leave to a channel.