Over the past two years, the most talked-about topic in the AI industry has been large models. OpenAI, Anthropic, Google, and xAI have nearly captured everyone's attention.

However, if you follow the flow of capital, you will find something very interesting:

The smartest money has already started to move away from 'large models' themselves.

According to statistics from Crunchbase and publicly disclosed corporate data, the 25 Series A companies with the largest financing scales in the global AI field in 2026 raised a cumulative total of over $6 billion.

Data Source: Crunchbase and Publicly Disclosed Corporate Data

Among them, robotics and embodied intelligence received over $2 billion in financing; AI infrastructure received nearly $1.5 billion; the majority of the remaining funds flowed into autonomous intelligent systems such as cybersecurity, defense, software development, and scientific research.

While the market is still discussing who will win between Anthropic and OpenAI, investors have already started betting on another question:

Where will AI truly create real value?

The answer is gradually emerging—robotics, computing infrastructure, and autonomous intelligent systems that can directly solve real-world problems.

Next, Silicon-Based Gentleman will take you through the dissection of these 25 AI startups with the highest financing amounts to see where the world's smartest money is betting.

/ 01 / Six Embodied Intelligence Companies Raised Over $2 Billion

Among the top 25 financing list, robotics-related companies account for six, with a total amount reaching $2.049 billion, making it the highest-funded sub-sector in Series A financing in 2026.

(1) Apptronik

Apptronik announced the completion of a $520 million Series A extension financing in February 2026, co-led by B Capital and the Qatar Investment Authority, with participation from Google and Mercedes-Benz. The post-money valuation reached $5 billion.

Apollo Humanoid Robot

Apptronik is a company focused on the development of general-purpose humanoid robots, with its core product being the Apollo humanoid robot—standing 1.75 meters tall and weighing 73 kilograms, it can carry a load of 25 kilograms. Its 32 high-torque density actuators form an advanced force control system, covering the upper limits of material handling needs in most industrial scenarios. The company has partnered with Google DeepMind to integrate the Gemini Robotics AI model into Apollo.

Its business model primarily consists of robot sales and Robotics-as-a-Service (RaaS), targeting manufacturing, logistics companies, and retail giants, offering whole machine sales and hourly leasing services.

In 2026, Apptronik completed the construction of the Apollo mass production line, initiated small-batch production (100 units) in Q2, expanded to 1,000 units in Q4, and aimed to deliver 10,000 units in 2027.

(2) Mind Robotics

Mind Robotics announced the completion of a $500 million Series A financing in March 2026, co-led by Accel and a16z, with a post-money valuation reaching $2 billion. The company is an independent entity spun off from electric vehicle manufacturer Rivian, which internally incubated it in November 2025.

Mind Robotics' core product is a general-purpose industrial robot platform that integrates AI models, dedicated hardware, and deployment infrastructure, providing a software-hardware integrated solution focused on automotive manufacturing, warehousing logistics, and industrial automation scenarios.

Its business model primarily consists of robot sales and RaaS subscriptions, targeting the automotive manufacturing, electronics assembly, and new energy industries, offering standardized industrial robots and customized production line solutions.

It plans to deploy hundreds of intelligent industrial robots at Rivian's Normal plant in Illinois by the end of 2026.

(3) Rhoda AI

Rhoda AI announced the completion of a $450 million Series A financing in March 2026, led by Premji Invest, with participation from Temasek, Khosla Ventures, Mayfield, and others, reaching a post-money valuation of $1.7 billion.

Rhoda AI is a company focused on the development of general-purpose robot foundation models, with its core product being FutureVision, capable of uniformly processing visual, language, motion, and environmental perception information to complete autonomous robot operations in real environments.

Its business model primarily consists of model licensing, targeting robot manufacturers, automation companies, and research institutions, providing foundation model licenses.

According to calculations by its investor Mayfield, at a deployment scale of 1,000 units, Rhoda can generate approximately $100 million in annual recurring revenue, becoming a $1 billion ARR business when reaching 10,000 units.

(4) Spirit AI (Qianxun Intelligence)

Chinese robot foundation model company Spirit AI (Qianxun Intelligence) announced the completion of nearly RMB 2 billion (approximately $290 million) in financing in February 2026, with investors including top-tier institutions such as Yunfeng Capital, Hundun Capital, and Sequoia China, reaching a post-money valuation exceeding RMB 10 billion.

Spirit AI's core product is the Spirit v1 VLA (Vision-Language-Action) embodied large model, enabling robots to 'see' scenes, 'understand' instructions, and autonomously complete operations, already capable of autonomously performing tasks in complex industrial environments. The model has been applied to CATL's battery production line, where one robot can replace 2-3 shifts of manual workstations, completing quality inspection processes such as high and low voltage testing.

Its business model primarily consists of model licensing and joint development, providing localized intelligent solutions for Chinese robot manufacturers and manufacturing companies. The company has established strategic partnerships with leading companies such as Bosch and JD.com, with the collaboration with Bosch covering the entire chain of data collection, model training, and global channels.

Currently, commercialization is still in its early stages, with the model's comprehensive success rate for new tasks around 40-50%. Management believes that large-scale commercialization will only be feasible when the success rate reaches 60-70%, estimated to mature around 2028.

(5) X Square (Autonomous Robotics)

Chinese industrial and service robot company X Square announced the completion of a RMB 1 billion (approximately $140 million) Series A++ financing in January 2026, with participation from ByteDance, Sequoia China, and Shenzhen Capital Group.

X Square's core product is the self-developed general-purpose embodied intelligence large model WALL series and matching robot hardware (such as the Quantum series wheeled dual-arm robot and five-finger dexterous hand), aiming to provide robots with an intelligent 'brain' to understand and operate the physical world, achieving end-to-end perception, decision-making, and operation.

Its business model primarily consists of robot sales, model licensing, and industry solutions, providing end-to-end intelligent solutions for advanced manufacturing, autonomous logistics, elderly care, and home service scenarios.

In March 2026, X Square launched the world's first robot home cleaning service in Shenzhen in collaboration with the 58 Daojia platform, achieving large-scale deployment of robots in complex C-end environments.

(6) Lightwheel (Guanglun Intelligence)

Chinese physical AI simulation company Lightwheel announced the completion of a total of RMB 1 billion (approximately $140 million) in Series A++ and A+++ financings in March 2026, led by Ant Group.

Lightwheel's core product is the Physical AI simulation and synthetic data platform, providing high-fidelity SimReady asset libraries, physically accurate simulation environments, and evaluation tools to generate large-scale synthetic data required for robot and autonomous driving training, addressing the data scarcity problem in physical AI.

Its business model primarily consists of data subscriptions and project-based services, targeting robot, autonomous driving, and AI companies, providing simulation and synthetic data solutions. Clients include global leading companies such as NVIDIA, Google, Figure AI, ByteDance, and Zhiyuan Robotics.

In the first quarter of 2026, the company secured RMB 550 million in new orders, with single-quarter revenue surpassing that of the entire year of 2025.

/ 02 / AI Infrastructure Accounts for 32%, Becoming the Hottest Early-Stage Sector

The AI infrastructure sector accounted for 8 out of the 25 financings, with a total amount reaching $1.485 billion, second only to the robotics sector. Directions such as optical computing, quantum acceleration, network switching, chip interconnects, and AI chip design simultaneously received capital pursuit.

(1) Ricursive Intelligence (USA)

US AI chip design company Ricursive Intelligence announced the completion of a $300 million Series A financing in January 2026, led by Lightspeed Venture Partners, with participation from DST Global and NVIDIA's NVentures, reaching a post-money valuation of $4 billion.

Ricursive's core product is an AI-driven chip design platform that shortens the chip design cycle from years to days, with its technology already applied to the design of four generations of Google's TPU. The company has built a recursive feedback loop between AI and chips, using AI to design better chips and then using better chips to train stronger AI.

Its business model primarily consists of an AI chip design automation platform, targeting chip manufacturers, cloud service providers, and AI companies, providing AI-driven full-process chip design solutions.

(2) OLIX

British optical computing company OLIX announced the completion of a $220 million Series A financing in February 2026, led by Hummingbird Ventures, reaching a post-money valuation of $1 billion.

OLIX's core product is the OTPU (Optical Tensor Processing Unit), an optical digital processor optimized for AI inference, performing bit-level precise calculations. The OTPU integrates SRAM architecture and photonic interconnect technology, completely bypassing HBM memory and advanced packaging, aiming to solve the 'memory wall' bottleneck restricting large-scale AI deployment.

Its business model primarily consists of photonics AI inference chip sales, targeting cloud service providers and AI infrastructure operators in data centers, providing software-hardware integrated inference acceleration solutions.

Olix expects its first commercial product to launch in 2027 and plans to expand its engineering teams in the UK and North America to scale up R&D and prepare for mass production.

(3) Neurophos

US optical inference chip company Neurophos announced the completion of a $110 million Series A financing in January 2026, led by Gates Frontier, with participation from Microsoft M12 and Bosch Ventures, reaching a total financing of $118 million.

Neurophos' core product is the Optical Processing Unit (OPU), which can be directly deployed in data centers as an alternative to GPUs. The company claims that its OPU chip performance has surpassed NVIDIA's B200 AI GPU. Its chips can run at frequencies up to 56GHz, with peak computational speeds reaching 235 POPS per second and power consumption of 675 watts, while the B200's computational speed is only 9 POPS per second with power consumption of 1,000 watts.

Its business model primarily consists of OPU chip sales and software-hardware integrated inference solutions, targeting cloud service providers and edge computing nodes.

Neurophos expects to initiate actual deployment testing in 2027, launch commercial systems in 2028, initially targeting the AI inference and data center module markets, and collaborating with Norwegian data center operator Terakraft for real-world testing.

(4) Sygaldry

US quantum acceleration AI server company Sygaldry announced the completion of a $105 million Series A financing in March 2026, led by Breakthrough Energy Ventures, with participation from multiple institutions including Y Combinator, the University of Michigan, and 468 Capital.

Sygaldry's core product is the quantum acceleration AI server, which accelerates AI model training and inference by co-deploying with existing GPU clusters in data centers, utilizing the parallelism of quantum computing to reduce operational costs and energy consumption in data centers.

Its business model primarily consists of quantum acceleration server sales and computational infrastructure services, targeting cloud service providers and AI data center operators.

The company is currently advancing the R&D and testing of quantum acceleration AI servers, aiming to achieve commercial production before 2030.

(5) Kandou AI

Swiss chip interconnect company Kandou AI announced the completion of a $225 million Series A financing in March 2026, led by Maverick Silicon, with participation from SoftBank, Synopsys, and Cadence, reaching a post-money valuation of $4 billion.

Kandou AI's core product is copper-based high-speed connectivity chips, aiming to solve data transmission bottlenecks in AI systems. As AI cluster scales expand, fiber optic solutions become costly, while copper connections suffer from severe crosstalk and power consumption spikes at extremely high frequencies—Kandou's solution converts crosstalk between cables into usable signal energy, transmitting 5 bits of data simultaneously over 6 lines with power consumption only half that of similar technologies.

Its business model primarily consists of IP licensing and chip sales, targeting hyperscale data center operators and AI infrastructure clients, providing high-speed interconnect IP and chips.

The company has already shipped over 20 million units, with its technology applied in data centers, AI, and consumer electronics.

(6) Upscale AI

US AI computing network company Upscale AI announced the completion of a $200 million Series A financing in January 2026, co-led by Tiger Global, Premji Invest, and Xora Innovation, reaching a post-money valuation exceeding $1 billion, with total financing reaching $300 million.

Upscale AI's core product is the SkyHammer™ AI in-rack network chip and system, solving communication bottlenecks between chips in AI clusters, directly competing with NVIDIA's core interconnect chip NVSwitch used in its top-tier AI cabinets (such as the NVL72 integrating 72 GPUs).

Its business model primarily consists of chip sales, system integration, and software subscriptions, targeting cloud providers, supercomputing centers, and AI companies, providing AI-specific network solutions.

The company currently plans to start shipping its first commercial product to the market in 2026.

(7) Eridu

US AI cluster network company Eridu announced the completion of a $200 million Series A financing in March 2026, co-led by Socratic Partners, John Doerr, Hudson River Trading (HRT), and Capricorn, reaching a total financing of $230 million.

Eridu's core product is a high-radix AI network switch, with one switch capable of replacing 30 traditional low-radix switches, significantly reducing network hierarchy, lowering network power consumption by up to 70%, and saving up to 40% in capital expenditures, solving the 'network wall' bottleneck in AI clusters.

The business model primarily focuses on the sales of switches and software subscriptions, targeting cloud providers and supercomputing centers to offer AI-specific data center network solutions.

The company is collaborating with TSMC to advance the development of switches based on custom chips. It has completed product definition validation with several leading hyperscale cloud service providers, with technical details and partnership plans set to be announced later in 2026.

(8) Aria Networks

U.S.-based data center networking company Aria Networks announced in X month of 2026 that it had completed a $125 million Series A funding round, co-led by Sutter Hill Ventures, Atreides Management, Valor Equity Partners, and Eclipse Ventures.

Aria Networks' core product is the Deep Networking platform, an AI data center network solution based on 800GbE/1.6T Ethernet switches designed to address communication bottlenecks in large-scale GPU clusters. The platform embeds microsecond-level telemetry and intelligent agents into switch ASICs, enabling automatic fault rerouting and improving Model FLOPS Utilization (MFU) by over 3%.

The business model centers on the sale of integrated hardware-software solutions for Ethernet switches, servers, and cloud components, targeting hyperscale cloud service providers and large-scale AI data center operators.

The company's products feature a chip-agnostic design, allowing operators to freely switch between hardware from different vendors such as NVIDIA and Google. The CEO revealed that the company has secured multiple customer orders and is currently deploying products in early adopter networks.

/03/ From Defense Security to Human Behavior Simulation: AI Enters the 'Deployment Era'

If robots execute tasks in the physical world and infrastructure supports intelligent operations, then these companies are attempting to directly build AI systems capable of autonomously completing complex work.

In terms of AI implementation, a total of 11 companies that secured over $50 million in Series A funding have raised $2.74 billion cumulatively, covering areas such as autonomous software development, scientific reasoning, cybersecurity, defense decision-making, and human behavior simulation.

(1) Hark

U.S.-based personal AI company Hark announced in May 2026 that it had completed a $700 million Series A funding round, led by Parkway Venture Capital with participation from mainstream chip giants including NVIDIA, AMD, Intel, and Qualcomm, reaching a valuation of $6 billion. This marks the largest global AI Series A deal in 2026 to date.

Founded in late 2025 by Brett Adcock (a serial entrepreneur and founder of Figure AI and Archer Aviation), Hark's core product is a personal AI assistant that deeply integrates a self-developed multimodal model with custom hardware devices, functioning as an intelligent assistant that operates 24/7 and proactively perceives user needs.

The business model primarily focuses on hardware sales + subscription services, targeting C-end consumers with hardware products and monthly AI service subscriptions.

The company currently employs 70 people and has built a dedicated data center equipped with NVIDIA B200 GPUs. Abidur Chowdhury, a former Apple product executive, serves as the design director. The first batch of multimodal models is set to launch in late summer 2026, with hardware devices to follow after the model release.

(2) Blitzy

U.S.-based generative software development company Blitzy announced in May 2026 that it had completed a $200 million Series A funding round, led by Northzone with participation from PSG, Battery Ventures, and others, reaching a valuation of $1.4 billion.

Blitzy's core product is a multi-agent autonomous software development platform capable of reading and understanding an entire enterprise's historical codebase (supporting over 100 million lines of code) in one go, automatically clarifying the system's business logic and module relationships. It then coordinates thousands of specialized AI agents to work in parallel, autonomously completing coding, testing, and validation. The company claims it can accelerate enterprise engineering speed by over 5x, with over 80% of development work completed autonomously.

The business model adopts a two-stage pricing structure: Enterprises pay an assessment fee of up to $250,000, followed by pricing of $0.20 per line of code generated or annual project-based pricing ($500,000 to over $10 million), directly tied to usage and delivered value.

Target customers include Global 2000 companies across 10 industries, with dozens of enterprises already adopting the platform, including QAD and State Street.

The company achieved a record-breaking score of 66.5% on the SWE-Bench Pro benchmark test, surpassing other major competitors. The funding will be used to expand the R&D team to 300 people and deepen customer collaborations in regulated industries such as government, financial services, and insurance.

(3) Axiom

U.S.-based mathematical reasoning AI company Axiom announced in March 2026 that it had completed a $200 million Series A funding round, led by Menlo Ventures, reaching a valuation of $1.6 billion.

Founded in 2025 by Carina Hong (with MIT/Oxford/Stanford backgrounds), Axiom's core product is the AxiomProver system—based on the Lean formal verification language, it ensures 100% correctness of code and mathematical reasoning, fundamentally addressing AI hallucination issues.

The system has autonomously proven multiple mathematical conjectures that remained unresolved for decades: It scored a perfect mark in the Putnam Mathematical Competition; autonomously proved Problem 124 from the Erdős problem set, which had been unresolved for about 30 years, and Problem 481, unresolved for 45 years.

The business model centers on verifiable AI technology, targeting financial institutions, research institutions, and critical infrastructure enterprises with services ranging from mathematical proofs to code verification.

In 2026, it has attracted multiple intention clients (potential clients). The company employs over 20 people, with a team including former Meta AI research director Shubho Sengupta (CTO) and mathematical luminary Ken Ono, who resigned from his tenure to join the company.

(4) Armadin

U.S.-based cybersecurity company Armadin announced in March 2026 that it had completed a $189.9 million Seed + Series A funding round, led by Accel with participation from GV, Kleiner Perkins, In-Q-Tel, and others. This marks the largest early-stage funding in cybersecurity history.

Armadin's core product is an autonomous AI attack agent platform that deploys thousands of AI agents to simulate real hacker attacks 24/7, automatically discovering security vulnerabilities in enterprise networks, verifying whether these vulnerabilities can truly be exploited, and providing specific remediation plans.

The business model primarily focuses on enterprise subscriptions and usage-based pricing, targeting critical industries such as finance, government, and technology to provide continuous AI-driven security validation services. The company formed a strategic partnership with Unit 42 under Palo Alto Networks in April 2026, integrating its AI attack engine into the Frontier AI Defense service to enable automated external attack surface assessment.

The company has initiated closed testing with multiple Fortune 500 companies, with its products validated in multiple financial institutions and large enterprise environments.

(5) Kai

U.S.-based Agentic security company Kai announced in March 2026 that it had completed a $125 million Seed and Series A funding round, led by Evolution Equity Partners with participation from N47 and multiple strategic investors.

Kai's core product is a unified Agentic AI security platform that deploys numerous autonomous AI agents to continuously monitor enterprise networks, cloud servers, endpoint devices, and various security tools. When abnormal behavior or security vulnerabilities are detected, the AI automatically analyzes, filters risks, and executes defensive actions based on priority.

The business model primarily focuses on enterprise subscriptions, targeting critical infrastructure, finance, energy, pharmaceuticals, and other industries to provide integrated AI security solutions.

During its stealth phase, the company collaborated with Chevron's Technology Venture Catalyst Program and secured customer orders across multiple industries, including energy, pharmaceuticals, and automotive.

(6) HavocAI

U.S.-based maritime defense autonomy company HavocAI announced in May 2026 that it had completed a $100 million Series A funding round, co-led by Cobalt Capital and Boardman Bay Capital with participation from Lockheed Martin, bringing total funding to $200 million.

HavocAI's core product is an all-domain collaborative autonomy software stack that supports unmanned platforms at sea, in the air, and on land to execute missions collaboratively in GPS/communication-denied environments, with a single operator commanding thousands of unmanned assets. The software stack consists of three parts: Havoc OS (real-time decision-making and hull control), Havoc Cloud (cloud interconnection), and Havoc Control (command interface), and has successfully integrated with the Link 16 military data link.

The business model centers on defense contracts, software licensing, and hardware integration, targeting the U.S. Department of Defense and allied forces to provide unmanned system autonomy solutions covering sea, land, and air domains.

The company has delivered dozens of autonomous surface vessels to the U.S. Department of Defense, participated in the U.S. Navy's "Silent Swarm" exercise and secured procurement, and plans to conduct large-scale collaborative combat demonstrations on the West Coast and Gulf of Mexico in Q4 2026.

(7) Scout AI

U.S.-based defense robotics company Scout AI announced in April 2026 that it had completed a $100 million Series A funding round, co-led by Align Ventures and Draper Associates, bringing total funding to $115 million.

Scout AI's core product is the Fury unmanned combat foundation model, designed specifically for tactical edge environments, supporting multiple platforms such as unmanned ground vehicles and drones, and capable of autonomous navigation and mission execution even when GPS/communication is disrupted.

The business model primarily focuses on defense contracts and software licensing, targeting the U.S. Department of Defense and allied forces to provide AI brains for unmanned systems.

The company has secured defense technology R&D contracts from DARPA, the Army Application Laboratory, and other Department of Defense clients, totaling $11 million. It currently holds four Department of Defense contracts and is supported by DARPA and the U.S. Army.

(8) Arena

AI model evaluation platform Arena (formerly LMArena) announced in January 2026 that it had completed a $150 million Series A funding round, co-led by Felicis and UC Investments with participation from Andreessen Horowitz, Kleiner Perkins, Lightspeed, and The House Fund, reaching a post-money valuation of $1.7 billion.

Arena's core product is a multimodal AI model evaluation platform that generates Elo rankings through anonymous blind testing and voting, covering all mainstream modalities such as text, code, web development, images, and videos.

The business model centers on B-end evaluation services, targeting AI labs, cloud providers, and enterprise clients.

The platform has over 5 million monthly active users across 150 countries, generating approximately 60 million model comparisons per month. Leading AI companies such as OpenAI, Google, and xAI use the platform for evaluation before publicly releasing their models. Just four months after launching its commercialized product in September 2025, its ARR surpassed $30 million.

(9) Fundamental

U.S.-based structured data AI company Fundamental announced in February 2026 that it had completed a $225 million Series A funding round, led by Oak HC/FT with participation from Salesforce Ventures, bringing total funding to $255 million and reaching a valuation of $1.2 billion.

Fundamental's core product is the Nexus large table model, which adopts a non-Transformer architecture optimized for structured data (tables) and can process billions of rows of data. It aims to replace traditional machine learning and data science workflows, with the company claiming superior accuracy over existing solutions in multiple use cases.

The business model primarily focuses on enterprise subscriptions + pricing based on data volume, targeting industries such as finance, retail, and healthcare to provide enterprise-grade data analysis and prediction services.

It has signed seven-figure-dollar contracts with multiple Fortune 100 companies and formed a strategic partnership with AWS, allowing AWS customers to directly deploy the Nexus model in their cloud environments.

(10) Simile

U.S.-based human behavior simulation company Simile announced in February 2026 that it had completed a $100 million Series A funding round, led by Index Ventures with participation from AI luminaries such as Li Feifei, Andrej Karpathy, and Adam D'Angelo. The valuation was not disclosed.

Simile's core product is a human behavior simulation platform that generates personal digital twins through interviews and data, constructing tens of thousands to hundreds of thousands of AI agents with realistic personalities, preferences, and decision-making logic to predict consumer, employee, or group behavioral responses in given scenarios and demonstrate the reasoning behind their actions. This technology can be applied to commercial scenarios such as product testing, market research, policy analysis, and advertising optimization.

The business model primarily focuses on enterprise annual subscription fees, targeting consumer goods, finance, healthcare, government agencies, and other sectors to provide human behavior prediction services.

It is currently collaborating with clients such as CVS Health and Telstra during the testing phase, with product implementation progressing steadily. More clients and partnerships are expected to be announced one after another (successively) after the product's commercial rollout.

(11) Recursive (UK)

UK-based company Recursive announced in May 2026 that it had completed a $650 million Series A funding round, co-led by GV (Google Ventures) and Greycroft with participation from chip giants such as NVIDIA and AMD, reaching a valuation of $4.65 billion.

Recursive expects to launch its first-generation 'Level 1' autonomous training system in mid-2026, with a core focus on recursive self-improvement superintelligence—building AI systems capable of autonomously optimizing their own code, training processes, and architectures.

The business model primarily focuses on technology licensing + cloud services + joint R&D, targeting tech giants, research institutions, and government departments to provide AGI-related technologies and solutions. Recursive expects to announce its technology roadmap and initial research progress within the next year.

Text/Yuanyuan