Modal's $15.75B Marks the Third Tier of a Layered Inference Market — And the Cheapest Stack Now Costs the Most

Modal Labs is on the verge of closing a $750 million round led by Accel at a $15.75 billion post-money valuation, according to a Sept. 28, 2026 TechCrunch report. The deal has not closed, but the price tag more than triples Modal's $4.65 billion mark from its May 2026 Series C and roughly 14x its $1.1 billion Series B in September 2025. Five months ago the same company was discussing a $250 million round; four months ago it printed $355 million at $4.65 billion. Capital is repricing the entire inference specialist tier in real time.

Modal disclosed on its Series C blog post that it had surpassed $300 million in annualized revenue as of May 2026, having grown fivefold since September 2025 when it was at roughly $60 million. Sacra estimates the April 2026 run-rate at the same $300 million mark, up from $119 million at year-end 2025. The implied multiple on the new round is therefore around 52x ARR — the highest among the publicly benchmarked inference specialists. CoreWeave trades at ~6.1x TTM revenue, Nebius at ~41x, and IREN at ~21x — and all three of those include a meaningful hardware-financing burden that Modal avoids.

Three private inference specialists have cleared double-digit-billion-dollar valuations in the last six months. Baseten priced a $1.5 billion Series F at $13 billion in June 2026 after ARR scaled to ~$600 million by March 2026, per company disclosure. Nscale raised $2 billion at $14.6 billion in March 2026 and signed a $45 billion compute deal with Anthropic in August 2026 for 460 megawatts at its Monarch, West Virginia campus. Modal is now the third peer into the $10 billion-plus club. Together, the three names have absorbed more than $4.25 billion of private equity in 2026 alone, on top of Baseten's prior ~$580 million and Nscale's $1.7 billion in 2025.

Modal's per-second billing at $0.001097/sec on H100 SXM5 ($3.95/hour) sits below CoreWeave's listed ~$4.25/hour and roughly at Lambda's $3.99/hour floor. The pricing gap to hyperscalers (3-6x) is intact, but the gap to listed neoclouds has compressed to single-digit percent. The competitive set is now competing on developer experience and instant autoscaling rather than on raw hourly cost. That is the structural shift Modal's $15.75B is pricing.

Modal disclosed in a Dec. 28, 2025 blog post that its worker pool had scaled to over 20,000 concurrent GPUs spanning AWS, GCP, Azure and OCI, with more than 4 million cloud instances launched over the prior two years and over 1 billion sandboxes. Critically, Modal does not own the GPUs — it rents them across hyperscalers and Oracle Cloud Infrastructure, where the company formalized a strategic cloud partnership in September 2024. The model is closer to a serverless orchestration layer than to a CoreWeave-style owner-operator: Modal adds developer tooling (sub-second cold starts, autoscaling from zero to 1,000+ GPUs) on top of capacity it sources from elsewhere.

Modal's customer book is also structurally different. Public case studies name Ramp (whose Ramp Inspect agent now writes over half of merged pull requests, per a Feb. 2026 Modal blog post), Lovable, Applied Compute, Cognition, DoorDash and Anthropic's Claude Science integration, which launched in June 2026. Roughly 60% of revenue comes from enterprise customers, Modal disclosed. There is no single anchor hyperscaler contract equivalent to CoreWeave's $22.4 billion OpenAI book or IREN's $9.7 billion Microsoft deal — Modal sells thousands of small, usage-based relationships rather than a few very large ones.

Three listed neoclouds anchor the comparable set: CoreWeave, Nebius and IREN. Each represents a different slice of the inference value chain, and the multiples diverge sharply. CoreWeave is the operating pure-play — $7.59 billion of TTM revenue, $2.575 billion in Q2 2026 alone (+112% YoY) and a 2026 guide of $12.4-13.2 billion — but it carries $51.6 billion of total debt and trades at just ~6.1x sales because investors price in the leverage and the $626 million Q2 2026 net loss. Nebius sits in the middle at $1.355 billion TTM revenue and ~41x sales; its anchor Microsoft deal ($17.4 billion signed Sept. 2025) gives it a hyperscaler floor. IREN is the Bitcoin-mining-to-AI-cloud pivot story: AI cloud revenue hit $70.5 million in Q4 FY26 (June 2026), its first quarter ahead of the $66.7 million Bitcoin mining line, with a year-end 2026 AI cloud ARR target above $4 billion anchored by a $9.7 billion Microsoft contract.

The multiple gap is the central read. CoreWeave's 6x is a debt-loaded, contract-heavy story that the public market has already marked down — the stock is at $85.07 against a 52-week range of $60.55-$153.20 and trades below most analyst targets despite the $104 billion backlog. IREN's 21x reflects the AI-cloud pivot still in front of execution: the $4 billion year-end ARR target versus today's $282 million annualized AI cloud rate (4x Q4 run-rate) is a step-change that has not yet landed. Nebius's 41x presumes the Microsoft capacity translates into recurring revenue through 2026. Modal's 52x sits above all three because private markets are paying for the developer-experience layer and the AI-coding tailwind rather than for hardware financing. If CoreWeave is the bond market's verdict on GPU rental, Modal is the venture market's verdict on the inference wrapper.

Every neocloud in the set depends on NVIDIA for supply. The question is whether NVIDIA's strategic capital has effectively arbitraged the market. NVIDIA disclosed an 11.5% passive stake in CoreWeave in January 2026 — 47.2 million shares acquired at $87.20, now worth ~$3.66 billion at the $85 stock price — and previously licensed Groq's inference IP for ~$20 billion, halving Groq's valuation to $3.5 billion in August 2026. NVIDIA's TTM revenue is $303 billion with $192.9 billion of net income; it both sells chips to every neocloud and owns a meaningful equity slice of the largest one.

For Modal, the lack of a hardware-financing burden is the moat. Modal rents capacity across Oracle OCI and the hyperscalers rather than buying $35 billion of GPU inventory like CoreWeave. That removes the 10%+ interest-rate exposure on chip-backed loans (CoreWeave's March 2026 $8.5 billion chip-secured facility priced with Moody's at A3 but the August 2026 $2.6 billion leveraged loan required a 5.5 percentage-point yield spread to clear). Modal's economics look more like a software margin profile — gross margin on consumption compute is high because Modal does not own the silicon. The trade-off is that Modal pays hyperscaler rates and cannot lock in capacity the way CoreWeave and Nebius can with their long-dated contracts. The $750 million in new capital extends the runway but does not change the structural dependency on the same GPU supplier.

Over days and quarters, the marginal investor question is whether Modal's $15.75B pulls private peers higher or resets the listed neoclouds lower. CoreWeave at 6x sales is the cheapest major inference infrastructure listing globally despite the $104 billion backlog; if Modal's premium holds, the listed discount becomes harder to defend and the stock should narrow the gap. IREN's $4 billion year-end ARR target versus today's run-rate implies ~14x sequential growth from AI cloud alone in two quarters — the most binary catalyst in the set.

Over 1-3 years, the question is not whether neocloud revenue keeps growing — Mordor Intelligence projects the neocloud market from $35.2 billion in 2026 to $236.5 billion by 2031, a 46.4% CAGR, and ABI Research expects inference to represent 80% of neocloud revenue by 2030 versus a training-dominated mix today. The question is which layer captures the margin as training-and-inference converges. Anthropic's August 2026 $45 billion deal with Nscale, IREN's $9.7 billion Microsoft contract, Nebius's $17.4 billion Microsoft deal and CoreWeave's $22.4 billion OpenAI book together represent ~$94 billion of contracted AI cloud revenue — most of it locked at hyperscaler-leveraged margins, not at Modal's software-style economics.

Modal's $15.75B is a bet that the developer-experience layer keeps extracting economics as inference workloads proliferate, even as the underlying GPU rental layer consolidates around two or three anchor hyperscaler contracts per neocloud. The bear case is straightforward: NVIDIA controls both the supply and a strategic stake in the largest customer of that supply; serverless wrappers compete on tooling alone, which is a feature, not a moat; and the $1.5 billion Nscale IPO at $30 billion would print a public-market valuation that benchmarks Modal's $15.75B from below. The bull case is that inference volume grows 40x in 2026 alone (Baseten's disclosed metric), and the layer that owns the developer relationship captures recurring revenue at 50%+ gross margin regardless of who owns the silicon.