In 2025, the keyword for U.S. AI venture capital will no longer be 'spray and pray' but heavy bets.

According to TechCrunch, U.S. AI startups completed 64 financing rounds exceeding $100 million each in 2025. Among them, eight companies secured multiple large-scale investments, with valuations continuing to rise. For instance, Cognition AI reached a valuation of $10.2 billion, while Sierra also joined the $10 billion valuation club.

Swipe down to view all content.

Among these projects with over $100 million in financing, the trend toward concentration in leading firms is particularly evident. In just this year, 35 deals exceeded $200 million each in a single round, covering 29 companies.

From the data, these 'smartest' global funds are primarily flowing into two main areas:

One area is reshaping AI's physical foundations: from chips and compilers to inference systems and computing power scheduling, capital is increasingly concentrating in lower-level, more systematically fortified segments.

The other area directly targets core business operations in high-value industries. AI is no longer stay at (this Chinese phrase means 'remaining at' but doesn't have a perfect English equivalent in this context; consider 'relegated to the role of') an auxiliary tool but is beginning to attempt to replace top programmers, participate in scientific discoveries, enter clinical decision-making, and even be introduced into military and security scenarios, assuming critical judgment roles.

Beyond the well-known star companies, Silicon Valley Insider has also selected 16 AI startups worthy of close observation, dissecting their product logic and business models one by one.

/ 01 /

Computing Power Arms Race: A Comprehensive Reconstruction from Chips to Compilation Layers

Among the 18 financing rounds listed, the AI infrastructure sector accounted for 7, each with substantial amounts, such as Cerebras' $1.1 billion and Unconventional AI's $475 million seed round. Investment directions are highly cutting-edge, including photonic computing (Celestial AI), 'redefining AI computers' (Unconventional AI), and compilation and runtime (Modular).

(1) Unconventional AI

Unconventional AI announced on December 8 a massive $475 million seed round led by Lightspeed Venture Partners and a16z. Valued at nearly $4.5 billion in this round, this year-old startup is rethinking the foundations of computers in the AI era.

Unconventional AI is a bio-inspired analog computing startup founded in 2025. Its founder, Naveen Rao (former head of AI at Databricks), and co-founders include MIT Associate Professor Michael Carbin, Stanford Assistant Professor Sara Achour, and former Google engineer MeeLan Lee.

The company's core product is an analog/mixed-signal AI chip + server system, drawing inspiration from the human brain's energy efficiency principles. Through physical-layer probabilistic computing and neuromorphic dynamics, it theoretically achieves 1,000 times greater energy efficiency than traditional GPUs, designed specifically for large model training and inference.

Its business model primarily revolves around chip sales, system integration, and customized services, targeting cloud providers, supercomputing centers, and AI enterprises to offer high-energy-efficiency computing solutions.

In 2025, it completed its core team formation and technical route validation, initiating chip design and tape-out preparations. Currently, it has no publicly disclosed revenue but expects to achieve tens of millions of dollars in revenue by 2026, focusing on energy efficiency breakthroughs and scalable deployment.

(2) Cerebras

AI infrastructure company Cerebras Systems raised a $1.1 billion Series G round, valuing it at $8.1 billion. Announced on September 30, the round was co-led by Fidelity and Atreides Management.

Cerebras is a hardware company specializing in wafer-scale AI computing. Founded in 2016 by Andrew Feldman (CEO) and Gary Lauterbach (CTO), its core product is the CS-3 supercomputer (equipped with the WSE-3 wafer-scale engine), integrating 4 trillion transistors on a single chip, designed for large model training and low-latency inference. Compared to GPU clusters, it achieves over 10 times faster training and lower latency.

Its business model centers on hardware sales, computing power services, and software subscriptions, targeting cloud providers, research institutions, and AI enterprises to offer whole-machine and computing power leasing services.

In 2025, the CS-3 will see scalable deployment, serving core clients like G42 and contributing over 87% of revenue. Simultaneously, it reached a 750-megawatt computing power cooperation agreement with OpenAI (to be deployed in stages from 2026), reducing client concentration risks.

Revenue exceeded $1.4 billion in 2025, with a target to surpass $2.5 billion in 2026.

(3) Celestial AI

AI infrastructure company Celestial AI completed a $250 million Series C round, valuing it at $2.5 billion.

Celestial AI is a photonic computing and AI acceleration company founded in 2020 by Alex Wright-Gladstein (CEO) and David Wright (CTO). Its core product is the Photonic AI Accelerator, which uses optical signals instead of electrical signals for matrix operations, enabling low-power, high-bandwidth AI inference. Designed for large models, edge computing, and data centers, it achieves over 100 times greater energy efficiency and 90% lower latency than traditional GPUs.

Its business model primarily involves chip sales, IP licensing, and customized integration, targeting cloud providers, edge device manufacturers, and autonomous driving enterprises to offer photonic acceleration chips and solutions. With chip sales and IP licensing as its core revenue streams, it generated approximately $80 million in revenue in 2025, with a target of $200 million in 2026.

In 2025, it completed chip mass production and delivered to its first batch of clients, deploying in data center inference and edge AI scenarios, with a client renewal rate exceeding 90%. In 2026, it plans to expand into autonomous driving and industrial automation, launching a second-generation photonic chip to enhance multi-model support and security compliance.

(4) Modular

Modular announced a $250 million financing round on September 24.

Modular is a company focused on unified AI computing infrastructure. Founded in 2022 by Chris Lattner (creator of LLVM/Clang and former head of Swift at Apple) and Tim Davis (former core member of Google's TensorFlow).

Its core products include the Mojo programming language (Python syntax, C/CUDA-level performance), the MAX inference engine (cross-hardware optimization, supporting NVIDIA/AMD/Apple Silicon), and the Mammoth scheduler (K8s-native, boosting GPU utilization to over 90%), forming a full stack of 'development-compilation-deployment-scheduling' to enable 'write once, run on multiple hardware'.

Its business model involves charging based on computing power/inference volume + revenue sharing with cloud providers, targeting enterprise developers, cloud service providers, and chip manufacturers to offer unified AI runtimes and optimization services.

With an Annual Recurring Revenue (ARR) of approximately $80 million in 2025, it targets $150 million in 2026, focusing on scalable enterprise deployment.

(5) Fireworks AI

Fireworks AI announced a $250 million Series C round on October 28, valuing the company at $4 billion.

Fireworks AI is an enterprise-grade open-source large model cloud platform founded in 2022 by Lin Qiao (former head of AI platform architecture at Meta and PyTorch core member), James Reed (former Meta engineer), and others.

Its core product is the AICloud platform, offering one-click deployment of hundreds of open-source models (Llama 3.1, DeepSeek, Stable Diffusion, etc.), LoRA fine-tuning, multi-LoRA parallelism, and private model uploads (up to 405 billion parameters). Its self-developed FireAttention and speculative decoding achieve 30-40 times faster inference and 80% lower costs.

Its business model involves charging based on tokens/computing power/fine-tuning tasks, targeting technology enterprises (Uber, DoorDash, Notion) and AI startups to provide production-grade AI infrastructure.

In 2025, it served over 10,000 enterprises and hundreds of thousands of developers, processing over 10 trillion tokens daily, with clients including Samsung and Shopify. With an ARR exceeding $280 million in 2025, it targets $450 million in 2026, solidifying its position as a leading open-source model cloud service provider.

/ 02 /

Penetrating Core Business Operations: AI's 'Replacement' and 'Enhancement' in Vertical Sectors

AI application financing is concentrated in vertical sectors with clear payment capabilities and high barriers, such as healthcare, enterprise services, defense, and programming.

(1) Cognition AI

Cognition AI announced a $400 million Series C round on September 8. Led by Founders Fund, the round valued the company at $10.2 billion.

Cognition AI is a company focused on AI autonomous programming and software development. Founded by Scott Wu, a former competitive programming champion, its core product is Devin—the world's first AI engineer capable of independently completing full software development tasks. Devin can autonomously plan projects, write code, debug and fix issues, deploy applications, and collaborate with developers for iterative improvements, significantly boosting R&D efficiency.

Its business model primarily involves enterprise subscriptions and customized services, targeting technology companies, financial institutions, and software outsourcing firms, charging based on seats and functional modules.

With an ARR exceeding $50 million in 2025, it targets $120 million in 2026, focusing on scalable deployment and industry customization.

(2) Sierra

Bret Taylor's customer service AI agent platform, Sierra, raised $350 million in a new financing round. Announced on September 4, the round valued the company at over $10 billion.

Sierra is an AI-driven enterprise-grade conversational interaction platform founded by former core members of Google's Assistant team. Its core product is SierraAI, a multi-modal intelligent interaction system capable of building highly human-like customer service, sales, and employee assistants for omnichannel automated interactions.

The product features intent understanding, contextual memory, and complex task processing capabilities, seamlessly integrating with enterprise systems like CRM and ERP.

Its business model involves SaaS subscriptions and charging based on interaction volume, targeting industries like retail, finance, healthcare, and telecommunications to offer standardized platforms and customized solutions.

Generating $80 million in revenue in 2025, it targets $150 million in 2026.

(3) Ambience Healthcare

Ambience Healthcare, a five-year-old company building an AI medical operating system, secured a $243 million Series C round.

Ambience Healthcare is an AI-driven clinical documentation automation company founded by a team from the medical AI and NLP fields. Its core product is the Ambience AI platform, which uses ambient sensing voice technology to capture doctor-patient conversations in real-time, automatically generating structured clinical notes, ICD-10/CPT codes, and supporting over 200 specialties with customization. It seamlessly integrates with mainstream EHR systems like Epic and Cerner.

The platform reduces doctors' documentation time by 45% and improves coding accuracy and compliance.

Its business model involves SaaS subscriptions and charging per provider, targeting hospitals, clinics, and medical groups based on the number of clinical users and functional modules. In 2025, it was deployed in multiple large healthcare systems (such as John Muir Health), expanding specialty coverage from emergency medicine and oncology to psychiatry and anesthesiology, with a client renewal rate exceeding 90%.

With an ARR exceeding $80 million in 2025, it targets $150 million in 2026.

(4) OpenEvidence

OpenEvidence, which specializes in building AI chatbots for the medical field, secured a second round of financing in 2025. On October 20, the company announced the completion of a $200 million Series C round, valuing it at $6 billion.

OpenEvidence is an AI clinical decision support company founded in 2021 by Daniel Nadler (founder of Kensho) and Zachary Ziegler. Its core product is the OpenEvidence platform, an AI question-answering system for doctors based on authoritative medical literature like NEJM and JAMA. It provides real-time clinical question answers with literature citations, supporting USMLE-level accuracy (reaching 100% in 2025) across diagnosis, treatment, medication, and coding scenarios.

The platform is freely available to certified doctors, integrating with EHRs and clinical calculators.

Its business model involves pharmaceutical/device advertising and enterprise subscriptions, monetizing through high-value medical advertising while offering customized deployment and data insight services to hospitals.

Covering over 10,000 hospitals/clinics across the U.S. in 2025, it supports an average of 18 million clinical consultations monthly, serving over 100 million patients. Generating $50 million in revenue in 2025 (with a 30% month-over-month growth rate), it targets $120 million in 2026.

(5) EliseAI

EliseAI, a medical and housing automation platform, raised $250 million in a Series E round, valuing it at $2.2 billion. Announced on August 20, the round was...

EliseAI is a vertical sector AI automation platform founded in 2017 by Minna Song and Tony Stoyanov. Its core product is a multi-modal AI assistant focusing on two major scenarios: property management and medical outpatient services. In property management, it automates rental inquiries, appointment scheduling for property viewings, maintenance responses, and rent collection. In medical outpatient services, it automates appointment scheduling, insurance verification, pre-operative notifications, and billing inquiries, supporting text interactions in 51 languages and voice interactions in 7 languages.

Its business model involves SaaS subscriptions and charging based on units/outpatient volume, targeting large U.S. apartment operators and medical chain institutions based on scale and functionality.

Covering over 10% of the U.S. apartment market in 2025, with rapid growth in medical services, it achieved an ARR exceeding $100 million in 2025, with a year-over-year growth rate exceeding 100%. In 2026, it plans to expand into European properties and more medical specialties in North America, strengthening private deployment and compliance capabilities, with a target of $200 million.

(6) Uniphore

Enterprise AI startup Uniphore, after announcing a $260 million Series F round on October 22, is valued at $2.5 billion. This round includes investments from Snowflake Ventures, NVIDIA, Databricks Ventures, and AMD.

Uniphore is an enterprise-grade conversational AI and automation platform founded in 2008 by Umesh Sachdev and Ravi Saraogi.

Its core product is the Business AI Cloud, which integrates voice/text analysis, AI agents, and process automation, covering scenarios like customer service, sales, marketing, and HR. It supports multi-language interactions and real-time insights, featuring zero-code deployment, data security compliance, and full-link automation capabilities, enabling deep integration with existing enterprise systems.

The business model combines SaaS subscriptions with seat/interaction-based billing, targeting industries such as finance, retail, telecommunications, and government. It provides standardized platforms and vertical solutions.

By 2025, the company will complete the integration of multiple products, launch industry-specific AI suites, and serve over 2,000 enterprises (including multiple Fortune 500 companies), with a customer retention rate exceeding 85%. Revenue is projected to reach approximately $230 million in 2025 (FY24: approximately INR 19,420 crore), with a 2026 target of $350 million.

(7) Shield AI

Shield AI, an AI-powered defense technology startup, raised $240 million in its Series F funding round, which closed on March 6. The round valued the company at $5.3 billion.

Shield AI is a defense AI autonomous systems company founded in 2015 by Ryan Tseng and Brandon Tseng (a former SEAL team member).

Its core product is the Hivemind Autonomous AI Stack, which enables drones and aircraft to autonomously execute missions in GPS/communication-denied environments. This includes the Nova quadcopter (for close reconnaissance), the V-BAT vertical takeoff and landing drone (for long-duration surveillance), and the X-BAT autonomous fighter. The products have been deployed in combat by the U.S. military and its allies, supporting swarm coordination and navigation in complex environments.

The business model primarily relies on defense contracts, hardware sales, and software licensing, targeting the U.S. Department of Defense, allied forces, and aviation OEMs. Charges are based on system delivery and software licensing. In 2025, the company secured a major U.S. military contract, deploying the V-BAT in the Indo-Pacific and Europe, while licensing Hivemind to manufacturers such as Airbus and Kratos.

Revenue reached $267 million in 2025, up 64% year-over-year, with a 2026 target of $400 million.

(8) Sesame

Sesame, a voice AI company, completed a $250 million Series B funding round co-led by Sequoia and Spark Capital. The round was announced on October 21, with participation from SignalRank.

Sesame specializes in AI voice interaction and personalized assistants. Founded by core members of Google's former voice team, its flagship product is Sesame—an intelligent assistant supporting multi-scenario voice interactions across smart homes, vehicles, offices, and mobile devices. It enables natural conversations, task execution, contextual memory, and cross-device collaboration.

The product features low-latency responses, dialect/accent adaptation, and privacy-focused local computing, seamlessly integrating with mainstream hardware and systems.

The business model primarily focuses on B2B licensing and subscriptions, targeting hardware manufacturers, automakers, and software platforms. Charges are based on device licensing and functional modules. Additionally, a consumer subscription version offers personalized assistant services.

By 2025, Sesame achieved large-scale deployment in automotive and smart home scenarios, partnering with leading automakers and IoT vendors. Voice interaction accuracy improved to 98%. Revenue reached $60 million in 2025, with a 2026 target of $120 million.

/ 03 /

From 'Prediction' to 'Discovery': AI-Driven Scientific Paradigm Revolution

Lila Sciences (two funding rounds), Periodic Labs (AI scientists), and SandboxAQ (AI+quantum) all point in the same direction: AI for Science.

(1) Lila Sciences

On October 14, Lila Sciences announced the second round of its Series A funding in 2025, raising $350 million.

Lila Sciences is an AI-driven scientific superintelligence platform incubated by Flagship Pioneering in 2023 and founded by Geoffrey von Maltzahn (CEO).

Its core product is the AISF (AI Science Factory), which integrates generative AI, autonomous scientific units, and automated laboratories to enable end-to-end autonomous research across life sciences, materials science, and chemistry. This covers hypothesis generation, experimental design, automated execution, results analysis, and iterative optimization, accelerating drug discovery, new material development, and carbon capture technology breakthroughs.

The business model primarily relies on B2B R&D collaborations, technology licensing, and revenue sharing from outcomes. It partners with pharmaceutical companies, material enterprises, and research institutions, charging based on project milestones and intellectual property royalties.

By 2025, the platform completed validation across multiple fields, successfully discovering new antibodies, catalysts, and carbon capture materials. It partnered with several biotechnology firms, generating approximately $50 million in revenue in 2025, with a 2026 target of $150 million.

(2) Periodic Labs

On September 30, Periodic Labs announced a $300 million seed funding round led by Felicis and a16z, with participation from NVIDIA and others, valuing the company at over $1 billion.

Periodic Labs is an AI-driven materials science startup founded in 2025 and headquartered in San Francisco. Its founders are Liam Fedus, former VP of Research at OpenAI and a key contributor to ChatGPT, and Ekin Dogus Cubuk, head of materials science at Google DeepMind.

Its core product is a trinity scientific stack: AI large models (research assistants) + high-fidelity physical simulations + automated robotic laboratories, forming a closed loop of 'virtual inference-physical validation-data feedback' to accelerate the discovery of critical materials such as superconductors, semiconductors, and batteries.

The business model focuses on B2B R&D collaborations and technology licensing, partnering with semiconductor, energy, and aerospace companies to develop new materials. Charges are based on project milestones and outcome-based revenue sharing.

By late 2025, the company launched its first enterprise collaboration projects, focusing on high-temperature superconductors and next-generation semiconductor materials. No public revenue data is available yet, with income expected from upfront payments on collaborative projects and technology licensing. Revenue is projected to reach tens of millions of dollars in 2026.

(3) SandboxAQ

On April 4, SandboxAQ completed a $450 million Series E funding round, valuing the company at $5.7 billion.

SandboxAQ is a technology company specializing in quantum+AI. Spinning off from Alphabet in 2021, its core team consists of quantum physicists, cryptographers, and AI experts. Its founder and CEO, Jack Hidary, is a former Google executive and a veteran in the quantum computing field.

Its core products include Post-Quantum Cryptography solutions (PQC, resistant to quantum attacks), quantum sensing and imaging systems (for healthcare, defense, and industrial inspection), and an AI-driven quantum simulation platform (accelerating materials and drug development).

The business model primarily relies on B2B subscriptions and project-based services, targeting industries such as finance, government, healthcare, and semiconductors. It provides security compliance, research acceleration, and industrial inspection services.

From 2025 to 2026, SandboxAQ partnered with multiple financial institutions and government agencies on PQC deployment. Its quantum sensing products entered medical and industrial mass production.

Revenue is projected to exceed $150 million in 2025, with a 2026 target of $250–300 million.

/ 04 /

Conclusion

The 2025 funding landscape reveals that AI's value is shifting from 'ubiquitous potential' to 'concrete productivity.' Capital is betting on who can use AI to genuinely reduce R&D costs, replace core human labor, push physical limits, or monopolize vertical industry data loops.

Money flows to the most fundamental and expensive transformations in physical computing, as this underpins all AI ambitions. It also goes to vertical domains' core segments with immediate economic value, where clear payment willingness and replacement costs exist.

Notably, capital is systematically supporting the emerging paradigm of 'AI for Science,' aiming to secure a ticket to the next industrial revolution—the intelligence revolution.

By Lang Lang