The $380B orchestration bet: Understanding the ‘coding wedge’ as AI labs move beyond the model layer - SiliconANGLE

Wall Street blinked — and $285 billion vanished.

Not because revenue collapsed. Not because guidance cratered. But because a handful of artificial intelligence labs signaled, unmistakably, that they’re no longer content supplying models to software-as-a-service companies. They want the whole stack.

What has been called the “SaaSpocalypse” isn’t about software dying. It’s about control — specifically, who controls the orchestration layer of enterprise AI. And right now, that battle is accelerating faster than most incumbents — and investors — are prepared to process.

Almost two weeks later, the AI world is still talking about the SaaSpocalypse triggered on Feb. 5 by the announcement of vertical plugins, such as legal and finance, for Anthropic PBC’s Claude Cowork. The fact that such a product launch could wipe out $285 billion in market value is a testament to Anthropic’s momentum, especially considering that no matter how powerful Cowork may be, it’s still just a niche product at this point.

Last week, TheCUBE Research partner Raphaëlle d’Ornano at Decoding Discontinuity published a deep-dive analysis offering a more sophisticated perspective on the challenges facing SaaS incumbents. Today, she published the second part of that analysis, breaking down the issue from the perspective of AI labs as they try to bullrush their way into this territory.

If you’re trying to understand Anthropic’s $380 billion valuation — and the existential risks hiding underneath the headline number — there are some important insights here.

Those start with her concept of the “coding wedge,” which I think is crucial for interpreting the current dynamic driving many of the headlines we see and the product launches from the big model providers.

Here is how she defines and summarizes it:

The “coding wedge” has become the strategic beachhead where winning developers in coding workflows represents the point of leverage to win enterprise AI adoption and broader orchestration control…. Coding emerged as the first domain where autonomous agents were reliable enough for real production use, becoming the practical proving ground for capability and cost…. Whoever controls how developers build with AI gains control of the orchestration layer, which then extends from developers to all knowledge workers and enterprise workflows.

Viewed through that lens, the events of the past few months form a distinct narrative.

In the latest research analysis, this is the core thesis:

AI labs are no longer primarily model companies. They are building enterprise operating systems. Every significant product OpenAI Group PBC and Anthropic has shipped in the past six months is orchestration infrastructure — not model improvements. Systems designed to coordinate agents across enterprise workflows, accumulate institutional context and create switching costs that justify platform multiples.

And almost all of them are aimed at developers and coding. Anthropic built a huge advantage here, which is why it captured such a larger enterprise share. OpenAI is trying to regain lost ground. This is the frontline for winning the orchestration layer.

This matters because of that $380 billion number. At Anthropic’s $14 billion annualized revenue run rate, that’s a 27x multiple. If Anthropic is selling intelligence in a market where intelligence is converging, that multiple is aggressive to the point of recklessness. If Anthropic is selling orchestration — meaning workflow coordination, enterprise context, accumulated institutional memory — then 27x may be the entry price for a generational franchise.

Two architectures, two different bets on the future

Raphaëlle unpacks how OpenAI and Anthropic have arrived at the orchestration layer with fundamentally different visions. The divergence tells you everything about each company’s beliefs about model commoditization.

OpenAI’s Frontier is a horizontal control plane. It manages agents from any provider. Agents get employee-like identities, scoped permissions and audit trails. Think Workday for AI labor. The strategic calculus is precise: Concede agent-layer competition in exchange for control-plane dominance. If enterprises standardize on Frontier for agent governance, OpenAI wins regardless of which model powers any individual agent.

The agents become interchangeable. The control plane does not.

Anthropic’s Cowork is the opposite — a bet on vertical integration: Build outward from the agent itself. Claude Code proved the model could handle complex enterprise tasks. MCP standardized how agents connect to systems. Cowork extends to knowledge workers. Vertical plugins go into legal, finance and sales. The logic is somewhat Apple-like: Control the end-to-end experience. Make execution quality the moat.

If intelligence commoditizes fully, Frontier wins. If intelligence retains meaningful differentiation when embedded in orchestration, Cowork wins.

The clock nobody talks about

Meanwhile, the part of this analysis that should keep investors up at night is the compute trap. Consider what Anthropic Chief Executive Dario Amodei said on the Dwarkesh Podcast. If Anthropic’s revenue comes in at $800 billion instead of $1 trillion, “there’s no force on Earth, there’s no hedge on Earth, that could stop me from going bankrupt.”

These companies are building the future while running along a knife’s edge. No company in enterprise technology history has operated at this level of capital intensity with this degree of revenue uncertainty over this compressed timeline.

And yet, to win the orchestration layer, you need frontier-quality models. You can’t orchestrate enterprise workflows with a mediocre model. But maintaining frontier models requires massive and growing compute. To fund the compute, the revenue growth must materialize on schedule.

As Raphaëlle points out, if enterprise adoption moves at enterprise speed instead of startup speed, the financial model breaks.

The bottom line

AI labs and SaaS incumbents are racing toward the same prize from opposite ends of the stack. The potential initial public offerings of OpenAI and Anthropic this year will offer a cornucopia of metrics. At the same time, it will test the ability of markets and investors to read and interpret them — to understand which ones matter and what they really reveal.

Image: SiliconANGLE/Reve