eLearning is Dead, Long Live Learning

The eLearning industry has been through very little serious disruption since it emerged as a byproduct of internet technologies. It is now facing its very first existential ripple, and from where I stand, it is already clear that the ripple has reshaped the core of the e-learning industry, though the scale of its impact will probably be apparent only by the third quarter of 2027.

Most service and product companies in the space have spent the last 24 months first riding the wave of post-COVID pipelines, then planning for the new wave of technology opportunity AI must have looked like, followed by denial, then waiting, and now pivoting on how to bolt AI onto what they already had. Over the past two quarters almost all eLearning companies have added AI something-or-other to their marketing literature. The smarter ones have started looking for alternative revenue streams.

Every SaaS company that ships an AI feature now inherits a cost line that didn't exist before: inference. ICONIQ's 2026 benchmark puts average gross margin on AI-native product lines at ~52%, against the 70–80% traditional SaaS built its valuations on. One listed education-technology company, NetClass Technology, reported a negative 6.4% gross margin on its AI computing line specifically — a single example, but a concrete one, priced below cost to hold market share while it worked out the unit economics.

And the budget has already moved: companies cutting L&D spend in 2026 cut it by $126,000 on average, while the same companies increased AI spend by $88,000 on average, in the same cycle, for nearly half of them. That's not a forecast. That's a self-reported reallocation that already happened.

That reallocation is what produces a new hierarchy in the industry: money is moving up-chain, toward whoever controls the AI models, and away from whoever used to sit between the model and the buyer. Here's what the new learning food chain looks like once the money's direction is established.

Layer one: the labs issue the credential themselves

At the top of the new training/learning food chain are the companies that own and create generative AI. The erstwhile dominant players — the LMS, HRMS, ERP, and CRM platforms that spent two decades owning enterprise learning infrastructure — are now on the next rung down.

Anthropic's Claude Partner Network — a confirmed $100 million investment, plus Claude Certified Architect, Claude for Education, and its own course platform. OpenAI Academy, a commitment to certify 10 million Americans by 2030, with Walmart, John Deere, Lowe's, BCG, and Accenture among the launch partners. Google Skills and the Google AI Professional Certificate, with Walmart and Verizon as employer partners. Microsoft's Copilot Academy inside Viva Learning — the one player straddling both the model side and an owned LMS product outright.

Layer two: Talent Transformation & Tech Consultants

A new category of entrant — Talent Transformation — sits just below the labs: specialist outfits with domain expertise in workforce planning, technical training, reskilling, upskilling, and related technology deployment. They are distinct from the erstwhile L&D, HR, and IT consultancies, and they are outcome-focused rather than content-focused.

The big tech consultant space is also being reorganised. Accenture and BCG are both named partners in the OpenAI program above; Accenture is also Anthropic's partner, training roughly 30,000 of its professionals on Claude. These firms aren't being bypassed; they're being pulled closer to the top of the chain, paid to do the enablement and change-management work the labs don't want to run themselves.

Cognizant made the category explicit in April 2026 with Skillspring, a "Talent Transformation Platform" marketed explicitly as moving "beyond static courses and compliance-driven training." That's a service layer too, and it's growing.

Layer three: intermediaries repackaging layer one's authority

DeepLearning.AI Pro says it plainly: instructors drawn from OpenAI, Anthropic, and Google. IBM's certificates, built on GPT and Llama APIs, distributed through Coursera. Coursera itself now hosts OpenAI, Anthropic, and Google's own courses directly — rent on shelf space, not ownership of the goods.

Layer four: real differentiation versus a costume

ElevenLabs trains its own foundation models in voice rather than wrapping someone else's — the clean case for genuine differentiation inside a single modality. DaVinci Resolve, Blackmagic Design's editing and color-grading suite, is the sharper example: decades of proprietary tooling, plus a deliberate integration of Claude and ChatGPT for specific editorial tasks, as reported by Animation World Network, with Blackmagic CEO Grant Petty on the record. Proprietary depth with a frontier model bolted on for one job, not a frontier model wearing a UI.

Layer five: data labelling, annotation, and the automated AI training value chain

This is a completely new layer as well, and one closely tied to the AI labs since it provides services the labs directly need. It's already a huge space: Surge AI, Scale AI, TELUS Digital, Appen, and Cogito Tech — and it runs on domain expertise more than raw labour: these companies recruit doctors, lawyers, engineers, and other subject-matter specialists specifically to rank and critique AI outputs in their field, then onboard and train them on how to apply that expertise inside an AI-training workflow rather than a classroom or a client engagement. For a displaced instructional designer or SME, this is one of the few places the underlying skill is still wanted — just repurposed, and paid for differently.

Layer six: eLearning, authoring tools, and LMS — what actually collapses

The eLearning and corporate-training companies were never built around learning or training as a discipline. They were built around the fragmentation of production — an obtuse toolchain that required an instructional designer, a developer, a graphic designer, a voice artist, and a QA/LMS admin to ship one module. The company existed to coordinate five specialists a client couldn't coordinate themselves. That's the thing collapsing, because AI increasingly lets one person do what took five — the coordination problem the company was solving no longer requires a company to solve it.

Some eLearning companies are adding "AI" teams and trying to pivot into RLHF, data-annotation, and labelling work, since that's where the bigger revenue projects currently are. But that's already a mature and crowded space with dedicated players, and arriving late to it is a harder position than building the layer from scratch.

The bypass that may be coming

Learning may finally have a real B2C play — small, and for now unmeasured. There's no granular data on this yet, but common people and small and medium companies can increasingly bypass tools and build or learn directly with Claude, ChatGPT, or Gemini, skipping the production-fragmentation layer entirely.

The learning industry spent two decades selling the idea that learning requires scaffolding, a semester structure, a credentialed intermediary — that people can't reliably teach themselves. Some of that was true. Some of it was the industry's own incentive talking, since a model where mastery took days instead of months would have collapsed the business that sold the months. AI is already delivering — not promising — learning at one's fingertips.

We need to watch the CAGR and growth numbers

The learning industry's own market reports will keep telling a growth story, and the reports won't be wrong. Technavio has corporate training adding $47.84 billion at a 7.9% CAGR through 2030. Mordor Intelligence has the global e-learning market climbing to $461.92 billion by 2031. Custom Market Insights projects e-learning past $2 trillion by 2035. These numbers are plausible, probably even conservative. What they don't show is who inside that number is actually growing.

Anthropic's annualized revenue grew from roughly $1 billion at the start of 2025 to $65 billion by the end of July 2026, the latter figure confirmed by CNBC — a 65x increase in nineteen months. OpenAI's revenue grew from roughly $20 billion at the end of 2025 to more than $40 billion by August 2026, as reported by Bloomberg, citing people familiar with the matter.

Neither figure is learning-specific revenue; both are total company revenue across coding, enterprise deployment, and consumer subscriptions, and neither company breaks out what share comes from training or certification products specifically. But set against the one figure that is learning-specific — the $126,000 average cut to L&D budgets against an $88,000 average rise in AI spend, in the same companies, in the same cycle — the direction is unambiguous even without a clean apples-to-apples number. Capital is moving up-chain, at a pace no CAGR forecast in the eLearning industry's own reporting comes within an order of magnitude of matching.

Part of why the traditional players can't close that gap is a pace they were never built to run at. OpenAI's 2025 financials, leaked and independently verified by the Financial Times, as reported by Fortune, show a $20.9 billion operating loss against $13.07 billion in revenue — burning capital faster than it earned it, on purpose, to fund a model-improvement cycle that ships new capability roughly every month. I can't think of an eLearning company, public or private, that has a balance sheet built to run a loss at that scale for that long, and the model labs aren't competing against them on content quality — they're competing on a capital and iteration speed the category was never designed to survive.

And the labs have a structural advantage: they don't need to hire a domain expert to enter a vertical. The same Claude, GPT, or Gemini model is already inside retail, agriculture, home improvement, telecom, and consulting simultaneously — one product, every industry, no vertical-specific SME hired and no content team built for any of them. An eLearning company had to recruit a compliance expert to sell into pharma and a different one to sell into finance. A frontier model doesn't need to own that expertise structurally to operate in either market — the compliance review itself doesn't disappear, but the vendor's need to build and staff a permanent expert bench per vertical does. The model and the end user are most of the production line now, where they used to be the last step in a much longer one.

And no, you can't slow it down — it's got a life of its own, and it's growing every day.