25 top inference chip startups… | PitchBook News Summary

1. Key Themes

Theme: AI compute is shifting from training to inference, pulling venture capital toward purpose-built chips

Inference is expected to dominate AI compute demand

The article frames inference as the center of gravity for future AI workloads, which is driving specialized-silicon investment.

"By 2030, an estimated 75% of AI compute demand will come from inference, the work of running models rather than training them."

Purpose-built chips are targeting GPU weaknesses on cost, energy and memory

Startups are attacking the specific bottlenecks of general-purpose GPUs.

"That shift is redirecting capital toward purpose-built chips designed to beat general-purpose GPUs on cost, energy use and memory bottlenecks."

Inference chip funding is at record pace

VC dollars into inference are already high and accelerating.

"VC-backed inference deals hit $5.2 billion in 2025, and 2026 is on pace to top it, with $4.2 billion raised through late August."

Capital efficiency (valuation built per VC dollar) is the ranking lens

PitchBook ranks the field by how much valuation each company has created per dollar raised, rather than by raw funding.

"Our list ranks the top VC-backed inference AI chip companies by capital efficiency: How much post-money valuation has each built for every dollar of venture capital raised?"

Theme: AI neocloud demand is outstripping available capital, and debt is a major funding source

Record neocloud VC investment is still seen as insufficient

Venture funding into neoclouds is up sharply, yet Nvidia's CEO says the need is greater.

"Global VC investment in AI neoclouds—companies that sell remote access to GPUs for AI workloads—has reached $6.6 billion across 58 deals so far this year. That's more money and deals than any prior year and 51% above last year's total of $4.4 billion." "the demand for neoclouds is off the chart," but that "the demand of the capital is much greater" than availability.

Neocloud financing blends VC mega-rounds with debt

Equity alone is not funding the buildout.

"Much of the AI neocloud sector is being funded by VC mega-rounds and by borrowing to bankroll chips." "Working capital could be debt," said David Yakobovitch... "It doesn't necessarily mean equity."

Nvidia is both supplier and investor

Nvidia's financial interest in neocloud growth is circular, which is worth weighing when assessing its demand commentary.

"Nvidia itself has been investing heavily in the AI neocloud space. It is among the investors in Nscale and Together AI. It also sells the GPUs that neoclouds are raising money to buy." "Jensen has a lot of skin in the game," Yakobovitch said.

Theme: Large rounds in defense, biotech and AI-adjacent categories

Big checks are flowing to AI-driven drug development and defense-adjacent biotech

"Biotech startup Pathos AI, which uses an AI platform in oncology drug development, secured a $237 million Series E led by Bristol Myers Squibb at a $3 billion valuation." "Warhound, a startup genetically engineering dogs for military use... emerged from stealth with $200 million from investors including Sequoia, 8VC and Corner Ventures at a $1.2 billion valuation."

Liquidity is returning through IPOs and strategic M&A

"Zetwerk... raised $518 million in its IPO and was valued at $2.7 billion after its market debut Friday." "Biotech Lycia Therapeutics... raised $317.5 million in its IPO." "Rivian acquired UK-based world model startup Third Dimension."

2. Contrarian Perspectives

Neocloud capital demand may be partly a function of Nvidia's own incentives The article's headline sources of optimism come from Nvidia's CEO, who benefits when neoclouds buy more GPUs. An investor should discount "off the chart" demand claims accordingly, especially since the funding relies heavily on debt.

"Nvidia itself has been investing heavily in the AI neocloud space... It also sells the GPUs that neoclouds are raising money to buy." "Much of the AI neocloud sector is being funded by VC mega-rounds and by borrowing to bankroll chips."

Capital efficiency, not total funding, may be the better signal for inference chip bets The conventional view tracks who raised the most. PitchBook argues for a valuation-per-dollar lens, under which a company's headline raise matters less than the value created from it.

"But raising money is only half the story." "Etched leads the field at 12.3x, turning $1.7 billion raised into a $21 billion valuation."

Note that this metric relies on post-money valuations, which are set by investors rather than realized in the market, so high efficiency scores reflect pricing sentiment as much as operating performance.

3. Companies Identified

Etched

Description: VC-backed inference AI chip startup
Why mentioned: Ranked #1 on capital efficiency
Quote: "Etched leads the field at 12.3x, turning $1.7 billion raised into a $21 billion valuation."

Element Labs

Description: Israeli inference chip company
Why mentioned: Ranked second on capital efficiency
Quote: "Israel's Element Labs follows at 8.5x"

Luminous Computing

Description: Photonic-compute chip company
Why mentioned: Ranked third
Quote: "ahead of photonic-compute player Luminous Computing at 6.1x."

d-Matrix

Description: Inference chip startup using in-memory compute
Why mentioned: Named as a company PitchBook analysts are watching
Quote: "d-Matrix, which performs calculations directly in memory"

Tensordyne

Description: Inference chip startup with an addition-based architecture
Why mentioned: Watched by PitchBook analysts
Quote: "Tensordyne, whose chip swaps power-hungry multiplications for simpler additions."

Nvidia

Description: GPU maker and investor in neoclouds
Why mentioned: CEO says neocloud capital needs exceed supply; Nvidia invests in and sells to neoclouds
Quote: "the demand of the capital is much greater" than availability.

Nscale

Description: AI neocloud
Why mentioned: Nvidia and DataPower Capital are investors
Quote: "It is among the investors in Nscale and Together AI."

Together AI

Description: AI neocloud
Why mentioned: Nvidia and DataPower Capital are investors
Quote: "It is among the investors in Nscale and Together AI."

Pathos AI

Description: Biotech using AI in oncology drug development
Why mentioned: $237M Series E at $3B valuation
Quote: "secured a $237 million Series E led by Bristol Myers Squibb at a $3 billion valuation."

Warhound

Description: Startup genetically engineering dogs for military use
Why mentioned: Emerged from stealth with $200M at $1.2B valuation
Quote: "emerged from stealth with $200 million from investors including Sequoia, 8VC and Corner Ventures at a $1.2 billion valuation."

Star Therapeutics

Description: Biotech developing drugs for unmet-need diseases
Why mentioned: $200M round
Quote: "raised a $200 million round led by OrbiMed and Janus Henderson Investors."

Mave

Description: Tel Aviv-based AI security operations startup
Why mentioned: Emerged from stealth with $34M
Quote: "emerged from stealth with $34 million in funding, including a $28 million Series A led by Red Dot Capital Partners"

Maxinsights

Description: Data infrastructure platform for physical AI
Why mentioned: $30M seed
Quote: "raised a $30 million seed round led by Peak XV Partners."

QuantrolOx

Description: Measurement and automation platform for quantum technologies
Why mentioned: €12M Series A
Quote: "secured a €12 million Series A led by IW Capital."

Chromalloy

Description: Aviation aftermarket services provider (Veritas Capital)
Why mentioned: Potential sale valued above $10B
Quote: "could be valued at over $10 billion in a deal, Bloomberg reported."

Goodles

Description: Boxed mac-and-cheese company (L Catterton-backed)
Why mentioned: Acquired by Barilla
Quote: "Barilla paid $750 million, Bloomberg reported."

Zetwerk

Description: Manufacturing business backed by Peak XV and Accel India
Why mentioned: IPO
Quote: "raised $518 million in its IPO and was valued at $2.7 billion after its market debut Friday."

Lycia Therapeutics

Description: Biotech for allergies and autoimmune diseases (Versant-backed)
Why mentioned: IPO
Quote: "raised $317.5 million in its IPO."

Third Dimension

Description: UK-based world model startup
Why mentioned: Acquired by Rivian
Quote: "Rivian acquired UK-based world model startup Third Dimension"

Partners Group

Description: Alternative asset manager
Why mentioned: Private credit evergreen fundraising
Quote: "closed its second rated feeder vehicle of 2026 for its private credit evergreen strategy, raising $510 million from US insurers"

4. People Identified

Jensen Huang

Description: Nvidia CEO
Why mentioned: Says neocloud capital demand far exceeds supply
Quote: "the demand for neoclouds is off the chart," but that "the demand of the capital is much greater" than availability.

David Yakobovitch

Description: General partner at DataPower Capital, investor in Nscale and Together AI
Why mentioned: Explains neocloud financing and Nvidia's incentives
Quote: "Working capital could be debt... It doesn't necessarily mean equity."

Paul Condra

Description: Senior Director, Global Head of Private Markets Research at PitchBook
Why mentioned: Author of the inference chip analysis
Quote: "Our list ranks the top VC-backed inference AI chip companies by capital efficiency"

Jacob Robbins

Description: Technology Reporter at PitchBook
Why mentioned: Author of the neocloud piece
Quote: "Venture investors have never invested more in AI neocloud companies, according to PitchBook data."

Ben Lamm and Palmer Luckey

Description: Colossal Biosciences' Lamm and Anduril's Luckey
Why mentioned: Backers of Warhound
Quote: "supported by Colossal Biosciences' Ben Lamm and Anduril's Palmer Luckey"

5. Operating Insights

Benchmark capital efficiency, not just capital raised Founders and investors can use valuation-per-dollar-raised as a quick comparative metric for hardware startups.

"How much post-money valuation has each built for every dollar of venture capital raised?"

Plan for debt financing when your business is GPU-intensive Capital-heavy infrastructure businesses should treat debt as part of the stack rather than relying solely on equity.

"Much of the AI neocloud sector is being funded by VC mega-rounds and by borrowing to bankroll chips."

Differentiate on a specific architectural bottleneck Inference chip startups win attention by attacking a concrete weakness of GPUs (memory movement, power-hungry multiplication, photonics).

"d-Matrix, which performs calculations directly in memory, and Tensordyne, whose chip swaps power-hungry multiplications for simpler additions."

6. Overlooked Insights

Evergreen fund launches hit a record in 2025 but have slowed sharply in 2026 This is a notable shift in private-markets product formation that could affect fundraising channels for managers.

"New evergreen fund launches reached a record high in 2025, led by credit strategies, though 2026 activity has slowed considerably."

Governments are becoming direct capital allocators to venture Canada's program targets emerging managers and life sciences specifically.

"The Government of Canada announced the Venture and Growth Capital Catalyst Initiative, in which it will provide C$1 billion ($714 million) in total funding through Funds-of-Funds, Life Sciences and Emerging Manager streams toward private investments."