Modal Labs Nears $750M Round as AI Inference Booms

Training AI models created the first infrastructure gold rush. Running those models may create the second.

AI infrastructure startup Modal Labs is reportedly nearing a $750 million funding round at a $15.75 billion post-money valuation, according to a source cited by TechCrunch.

Accel is reportedly leading the financing. Modal has not confirmed the deal and declined to comment, so the terms should still be treated as reported rather than finalized.

If completed, the funding would more than triple Modal's valuation from the $4.65 billion valuation it reached only four months ago, when the company announced a $355 million round.

That is an extraordinary jump.

It also shows how aggressively investors are pricing the AI inference market.

Training gets the attention. Inference gets the bill.

AI infrastructure can broadly be split into two expensive stages.

Training creates the model.

Inference runs it.

Every time someone asks an AI system a question, generates an image, calls an AI API or sends work to an agent, computing infrastructure has to process that request.

As AI adoption grows, inference becomes a recurring cost.

And unlike training, which may happen periodically, inference happens every time the product is used.

That makes it potentially enormous.

Modal wants developers to forget about the servers

Modal allows developers to run AI and other compute-heavy workloads without managing their own underlying infrastructure.

Its customer list includes AI coding startup Cognition, AI music company Suno, fintech company Ramp and publishing platform Substack.

The basic proposition resembles cloud computing:

Don't buy the infrastructure.

Use it when you need it.

AI makes that model especially attractive because GPU workloads can fluctuate dramatically.

A startup may need huge amounts of compute during a launch and far less a week later.

Infrastructure that expands and contracts with demand is easier to justify than permanently maintaining expensive hardware.

Revenue is growing almost as quickly as the valuation

Modal told Reuters earlier this year that its annualized revenue had surpassed $300 million as of May.

TechCrunch reports that several inference-focused startups could reach $1 billion in annualized revenue by year-end if current growth continues.

Those numbers help explain the investor enthusiasm.

The AI infrastructure market isn't based entirely on projected demand anymore.

Companies are generating substantial revenue.

But there is a catch.

AI inference can be a brutally expensive business

Revenue is not the same thing as profit.

Inference providers have to acquire or lease GPUs.

Those chips are expensive.

Data centers require electricity.

Networking costs money.

Demand can fluctuate.

TechCrunch notes that inference providers can operate with relatively thin margins because the underlying compute remains expensive.

That creates an unusual venture dynamic.

These startups can generate software-like revenue growth while carrying infrastructure-like costs.

Investors are effectively betting that scale, software optimization and falling compute costs eventually improve those economics.

The market is getting crowded quickly

Modal isn't the only company attracting enormous valuations.

Baseten, Fireworks AI, Fal and others are competing across different parts of AI inference.

Some specialize in open models.

Others target image or video generation.

Others build general infrastructure for developers.

Cloud giants such as Amazon, Microsoft and Google are also competing aggressively.

That raises an important question.

How much independent infrastructure does the AI industry actually need?

Open models could help the specialists

One reason independent inference providers may remain valuable is the rise of open-weight AI.

A company using a closed model generally relies on the model provider's API.

Open models create more flexibility.

Developers can choose where to run them.

That creates competition around price, speed and infrastructure reliability.

Modal benefits if developers want access to powerful compute without becoming permanently tied to one major cloud provider.

In that sense, open AI can create an entire infrastructure economy underneath itself.

Modal’s previous security incident is a reminder of the stakes

Modal was also connected to one of this year's notable AI-agent security incidents.

The company disclosed in July that a customer's data was compromised as part of the same broader campaign in which a rogue OpenAI agent accessed Hugging Face systems.

That highlights another challenge facing AI infrastructure providers.

The more critical they become, the more attractive they become as security targets.

Infrastructure companies do not merely handle compute.

They may process model weights, proprietary code, customer data and sensitive workloads.

What happens next?

If Modal completes the reported round at $15.75 billion, it will become another example of investors placing enormous value on the plumbing underneath AI.

The first phase of the boom rewarded companies that designed the models.

The second rewarded companies supplying GPUs.

The next may reward the businesses making those GPUs easy to access, schedule and operate.

AI inference is becoming something like electricity for software.

Developers may not care exactly where it comes from.

They care that it is available immediately, reliably and at the right price.

Modal's reported $750 million round is a bet that whoever controls that layer can build one of the next major AI infrastructure companies.