Top 10 US-Based Tech Startups to Watch in 2026

Every year the startup market produces companies that are well-funded. Fewer produce companies that might actually matter. This list is about the second category.

Choosing which young companies to pay attention to in a moment when venture capital is flowing at record levels requires a different lens than the one most funding announcements invite. A large round tells you that sophisticated investors made a bet. It does not tell you whether the underlying technology works, whether customers will pay for it at sustainable margins, or whether the founding team can execute across the decade that building a durable business actually requires.

The companies on this list were chosen because each one is doing something that has not been done before, in a market that will need what they are building, and in most cases because there is early evidence that the thing they are building actually works.

1. Corgi Insurance

Founded in 2024 by Nico Laqua (CEO) and Emily Yuan (COO, formerly of OpenAI), Corgi Insurance is headquartered in San Francisco and operates as a full-stack insurance carrier built for startups. In May 2026, Corgi raised $160 million in a Series B led by TCV, bringing total funding to $268 million. The valuation reached $1.3 billion in May and climbed again to $4 billion in a July 2026 extension.

The distinction that separates Corgi from other insurtech companies is the word “carrier.” Most insurance technology companies are brokers or managing general agents: they sell policies underwritten by someone else and are therefore limited in how much they can change the fundamental product.

Corgi acquired a decades-old licensed insurance company for approximately $35 million, received full regulatory approval in July 2025, and now writes, underwrites, and processes claims on its own paper. That means it can build AI systems that operate across the entire insurance stack because it owns the entire insurance stack. Quotation, underwriting, policy issuance, and claims processing all run on AI systems the company built internally.

The early commercial evidence is the reason this company belongs on a watch list rather than a wish list. By January 2026, Corgi had crossed $40 million in annual recurring revenue since receiving its carrier license only six months earlier. Customers include Deel and Bland.ai. The product covers general liability, cyber, tech errors and omissions, directors and officers, and employment practices liability coverage, the bundle of insurance lines that most startup-focused policies bundle awkwardly. Corgi quotes in minutes rather than the days or weeks that traditional brokers typically require.

What remains genuinely unknown is whether Corgi’s AI-driven underwriting models perform well across market cycles. An insurer is ultimately judged on claims performance over years, and Corgi’s carrier license is barely a year old. The next stress test will be a claim of meaningful size, handled at the speed the platform promises.

2. Starcloud

Founded in January 2024 by Philip Johnston (CEO, ex-McKinsey), Ezra Feilden (CTO, ex-Airbus Defence and Space), and Adi Oltean (Chief Engineer, ex-SpaceX Starlink and Microsoft Azure), Starcloud is headquartered in Redmond, Washington. In March 2026, the company raised $170 million in a Series A, bringing total funding to $200 million and reaching a $1.1 billion valuation.

Starcloud’s thesis is that terrestrial data centers are running into hard physical constraints: energy availability, water for cooling, permitting timelines measured in years, and geographic concentration risk. Space offers unlimited solar energy, radiative cooling that costs nothing, and the ability to add compute capacity at the speed of a rocket launch rather than the speed of a construction permit.

In November 2025, Starcloud became the first company to train a large language model in space, running nanoGPT on its Starcloud-1 satellite. It also ran Google’s Gemma model on the same orbital GPU, demonstrating that the concept works in practice rather than only in simulation.

The honest framing is that Starcloud has proven a concept and not a business. The gap between training a small model on a single NVIDIA H100 in low Earth orbit and operating a gigawatt-scale orbital data center is a gap of many years, many rocket launches, and engineering challenges that have never been solved at commercial scale. The team’s backgrounds in space infrastructure (Feilden), cloud computing (Oltean), and enterprise go-to-market (Johnston) are well-suited for the journey. The question is whether the launch cost reductions that make the economics plausible continue at the pace the model requires.

3. Positron AI

Founded in 2024 by Thomas Sohmers, Barrett Woodside, and Edward Kmett, with Mitesh Agrawal as co-founder and CEO, Positron AI is an AI chip company designing inference-optimized hardware as an alternative to NVIDIA GPUs for running, rather than training, large models. In January 2026, the company closed a $234 million Series B, bringing total funding to $309 million and a valuation of $1.06 billion.

The business case rests on a real gap in the current AI infrastructure market. Training requires general-purpose parallel compute that NVIDIA’s H100 and H200 GPUs do well. Running large models in production, particularly at consumer scale, involves continuous inference workloads where the energy consumption and cost-per-token of GPU inference creates meaningful economic friction. Positron’s Atlas FPGA-based systems are currently shipping, and the company’s roadmap targets a custom ASIC that would improve inference efficiency further. The company claims its systems can run models with up to 16 trillion parameters, a scale that matters as frontier models continue to grow.

The risk here is the structural challenge facing any semiconductor startup: the cost of tapeout, the time required to validate a new chip architecture in customer production environments, and the competitive pressure of an NVIDIA that is not standing still. Positron has raised serious capital from serious investors including Valor Equity Partners and Atreides Management. Whether the hardware performance advantage is wide enough and durable enough to sustain a business is the question that the next two to three years will answer.

4. Chai Discovery

Founded in 2024 by Joshua Meier, Jack Dent, Matthew McPartlon, and Jacques Boitreaud, Chai Discovery is a San Francisco-based AI biotech company building foundation models for molecular structure prediction and drug design. In July 2026, the company raised a $400 million Series C led by Index Ventures, Kleiner Perkins, and Sequoia Capital, bringing total funding to approximately $630 million.

Chai Discovery is on this list not because of the funding, but because of what the funding is responding to. In 2025 and 2026, Chai signed commercial partnerships with Eli Lilly (mid-eight figures annually), Pfizer, and Novartis, three of the largest pharmaceutical companies in the world. The partners cited by Index, Kleiner, and Sequoia as the rationale for their Series C participation are active users of the technology, not early adopters who agreed to a pilot. That is a meaningful commercial signal in an industry where AI partnerships in drug discovery have often been announced without producing licensed compounds.

The technical foundation is legitimate. Chai-1, released in September 2024, matched or exceeded AlphaFold3 on the PoseBusters benchmark with 77% success versus 76%, while operating without the multiple sequence alignments that AlphaFold3 requires. Chai-2 introduced de novo antibody design at a scale that prior approaches could not achieve, hitting a roughly 16% success rate across 52 novel antigen targets. In 2025, Chai-1 was cited in 304 published research papers. Academics citing a commercial tool that extensively in peer-reviewed work reflects genuine technical credibility. Chai-3 launched in 2026, and the pipeline of model generations shows a team that is iterating with speed rather than coasting on an initial release.

5. Bedrock Robotics

Founded in 2024 and headquartered in San Francisco, Bedrock Robotics is led by CEO Boris Sofman, who previously built and sold Anki, the consumer robotics company, to Amazon. In July 2025, Bedrock emerged from stealth with $80 million across a Seed round led by Eclipse and a Series A led by 8VC. By mid-2026, total funding had reached $350 million.

The problem Bedrock is addressing is concrete and immediate. The United States is in the middle of a manufacturing reindustrialization push, with $238 billion invested in domestic manufacturing facilities in 2024 alone, and a persistent shortage of construction workers that is currently estimated at 500,000 people. Data centers, semiconductor fabs, housing, roads, and ports all require heavy machinery operators that do not exist in sufficient numbers to meet the build rate the reindustrialization agenda requires.

Bedrock’s approach is not to build new machines, which would require customers to replace billion-dollar fleets. It retrofits existing excavators with a hardware and software system it calls the Bedrock Operator, which can be installed in a single day and enables the machine to operate autonomously.

The company is running on active construction sites in Arizona, Texas, and Arkansas. That is not a controlled demo environment. It is production. The excavators are moving real dirt on real projects with real project schedules and real financial consequences for errors. Bedrock’s 2026 target is operator-less deployment, meaning a machine running autonomously without a human in the cab. The distance between autonomous-with-oversight and genuinely-autonomous is always larger than early demonstrations suggest, but the industrial context and the founding team’s execution track record give Bedrock more credibility than most hardware automation startups at a comparable stage.

6. XBOW

Founded in 2024 by Oege de Moor, the creator of GitHub Copilot, XBOW is headquartered in Seattle, though much of the team works from Malta, where de Moor is based. In March 2026, XBOW raised a $120 million Series C led by DFJ Growth and Northzone at a valuation exceeding $1 billion, with a $35 million strategic extension in May from NVIDIA, Accenture, Samsung Ventures, and SentinelOne’s S Ventures.

XBOW runs thousands of autonomous AI agents that probe software for vulnerabilities the way a skilled human penetration tester would, but continuously and at machine speed rather than through the periodic engagements that have been the security industry standard. When attackers use AI to find and exploit vulnerabilities faster than human security teams can detect them, the asymmetry of periodic pen testing becomes untenable. XBOW addresses that asymmetry by making the defense continuous.

The commercial signal that distinguishes XBOW from theoretical security products is investor participation from NVIDIA, SentinelOne, Samsung, and Accenture, several of which are also XBOW customers. An enterprise customer making a financial investment in a security vendor is a different signal than a strategic partnership announcement.

The company has also appointed a CISO with real credentials in Nico Waisman, formerly of Lyft, who assembled the human hacker team that trained XBOW’s autonomous systems. The product was embedded in the Microsoft Security Ecosystem in March 2026. The question for XBOW is whether autonomous pen testing can remain ahead of the autonomous attack capabilities it is designed to counter as both sides of that equation accelerate.

7. Goodfire

Founded in June 2024 by Eric Ho (CEO), Dan Balsam (CTO), and Tom McGrath (Chief Scientist), Goodfire is a San Francisco-based applied research lab working on AI interpretability. In February 2026, the company raised $150 million in a Series B, bringing total funding to $207 million.

Goodfire occupies a category that the mainstream AI industry has consistently undervalued: understanding what is actually happening inside the neural networks that are being deployed everywhere. Most AI development treats models as black boxes. They are trained, evaluated on benchmarks, and deployed with limited understanding of the internal mechanisms that produce particular outputs or failures. Goodfire’s research targets this gap using techniques like sparse autoencoders, which identify interpretable features within neural networks, making it possible to explain model behavior and, crucially, to edit models in targeted ways rather than retraining them wholesale.

The team has published three of the most cited papers in the mechanistic interpretability field, which is an unusual foundation for a company as young as Goodfire. The Ember platform translates interpretability research into practical tools for organizations that need to understand, debug, and steer AI systems. As AI regulation matures globally and organizations face increasing pressure to explain their AI systems’ outputs to regulators, auditors, and customers, the ability to inspect and modify model behavior without retraining becomes commercially valuable rather than academically interesting.

Goodfire’s research focus is the correct long-term bet. The near-term question is whether the enterprise market for interpretability tools is developing fast enough to build the revenue base a $207 million funding commitment implies.

8. Also

Also began in 2022 as a skunkworks project inside Rivian Automotive under CEO RJ Scaringe, who sits on the company’s board. In March 2025, Rivian spun Also out as an independent company with $105 million in backing from Eclipse Ventures. By July 2025, a follow-on investment from Greenoaks Capital valued the company at $1 billion. Total funding has reached approximately $300 million.

The TM-B, Also’s first product, is a Class 3 electric bike priced at $4,500 at the performance tier. The description does not capture what makes it interesting. The TM-B uses pedal-by-wire technology that eliminates the mechanical connection between pedaling and movement, replaces it with software-controlled torque, and makes the riding feel programmable in the way that software is programmable rather than fixed in the way that mechanical drivetrains are fixed.

The top frame is modular and swappable, converting the same bike between solo commuter, cargo carrier with a bench seat, and utility configurations without requiring a different bike for each use case. The battery doubles as a power bank. The bike has LTE, Bluetooth, GPS, and a security system that remotely locks the frame, wheels, and battery and notifies the owner in real time.

The Verge called the Chromatic the best Game Boy ever made. The Washington Post called the TM-B “fun to ride and surprisingly difficult to steal.” Wired described it as solving the age-old “N+1” problem, the cycling community shorthand for always needing one more bike. Also partnered with DoorDash in March 2026 to accelerate autonomous delivery on the platform. A four-wheeled cargo variant is in development for Amazon.

Also began deliveries in summer 2026 after supply chain delays pushed back the original spring timeline, a setback that frustrated some early customers but did not change the product’s technical ambition. The real test is whether a $4,500 e-bike from a startup without a retail distribution network can build the customer base that justifies the underlying platform investment.

9. Genspark

Founded in 2023 by Eric Jing (CEO, founding member of Microsoft Bing), Kay Zhu (CTO, formerly of Google’s deep neural ranking team), and Wen Sang (COO, Y Combinator and Khosla-backed enterprise SaaS background), Genspark is headquartered in Palo Alto. In November 2025, the company raised $275 million in a Series B at a $1.25 billion valuation. As of June 2026, the valuation reached $2.6 billion following an additional $100 million Series B extension, with total funding exceeding $645 million.

Genspark is on this list because of its revenue trajectory, which is the metric that most AI workspace companies prefer not to discuss in detail. Sacra estimates Genspark reached $250 million in annualized revenue by March 2026, up from $85 million at the end of 2025. The company has approximately 2 million monthly active users and an estimated 100,000 paying customers at $30 per user per month. For a company with fewer than 200 employees, the revenue-per-employee ratio is remarkably high, which reflects an AI-native product architecture that does not require large professional services or customer success teams to deliver value.

The product, Genspark AI Workspace, handles research, communication synthesis, content creation, and task automation through a single interface. The Claw AI assistant, launched in March 2026, provides a private alternative to open-agent platforms for enterprises with sensitive data requirements. The founding team’s background in search and ranking systems, specifically in building systems that synthesize information rather than simply index it, is directly relevant to what Genspark is building. The question, as with any high-ARR AI workspace company in 2026, is how durable that revenue is once larger competitors embed comparable functionality inside tools people already pay for and use every day.

10. ModRetro

Founded in 2023 by Palmer Luckey (creator of Oculus VR, co-founder of Anduril Industries) and Torin Herndon (CEO), ModRetro is headquartered in Irvine, California. The company has raised approximately $19 million to date and, as of March 2026, was in early talks to raise a new round at a $1 billion valuation.

ModRetro is the outlier on this list. It is building FPGA-based retro gaming hardware with an obsessive commitment to authenticity that the team describes as “hundreds of irrational decisions.” The Chromatic, released in 2024 at $199, is a handheld console built on the same FPGA chip technology used in the open-source MiSTer project. It plays original Game Boy and Game Boy Color cartridges and newly developed Chromatic-native titles with zero input latency, through a display that was manufactured specifically and solely to render 160 by 144 pixels at the exact color temperature of the original hardware. It sold out on launch. It sold out again when restocked in July 2025 with new colors and configurations.

The M64, ModRetro’s second product and its first home console, is an FPGA-based Nintendo 64 priced at $199. It runs original N64 cartridges. Demos appeared at GameStop locations in spring 2026. The company operates a small independent game publishing studio producing both original titles and physical re-releases for the Chromatic platform. A limited-edition Anduril-branded Chromatic was released in December 2025, a collaboration that provoked predictable commentary given Anduril’s defense technology focus and did not appear to meaningfully affect sales.

ModRetro belongs on a watch list because it represents something equally rare: a hardware product company with genuine product-market fit in a market many observers had declared dead, run by a founder with one of the more unusual track records in American technology.

What These Tech Startups Tell Us About Where Technology Is Going After 2026?

Taken together, these ten companies map a more specific picture of 2026 technology than “AI is everywhere” covers. The picture has three structural features that are worth naming.

The first is that the constraint is no longer software. Several of these companies, Starcloud, Positron AI, Bedrock Robotics, and Also, are building hardware because software alone cannot solve the underlying problem. AI requires compute that consume electricity at a scale that terrestrial infrastructure cannot sustainably supply: Starcloud’s orbital data centers are a response to that hard ceiling. AI inference at consumer scale is economically constrained by GPU energy costs: Positron’s inference chips are a response to that economic friction.

Physical construction cannot be automated with software that does not also control physical machines: Bedrock’s hardware retrofit approach is a response to that gap. E-bike platforms that commoditize frames and standardize battery interfaces require building a proprietary architecture from the beginning, not bolting one on later. When software people start building hardware, and when hardware people with backgrounds in automotive engineering build bicycles, it usually means the software-only version of the solution did not work.

The second feature is that the most interesting AI applications in 2026 are not the frontier model providers. They are the companies finding novel ways to use or constrain the models that already exist. Goodfire is not training a larger language model. It is building the tools to understand, debug, and edit the ones that already exist.

Chai Discovery is not competing with GPT-5 or Claude. It is applying AI to a scientific domain where the combination of falling compute costs, falling synthesis costs, and accumulating biological data has created a window for AI models to compress drug discovery timelines in ways that human researchers working on traditional timelines cannot match. XBOW is not replacing human security professionals. It is running autonomous penetration testing at a frequency and scale that human teams cannot sustain, in response to attackers who are using the same automation advantage.

The third feature is that some of the most durable businesses being built in 2026 involve replacing the regulatory and institutional infrastructure of industries that technology avoided for decades. Corgi is not building a software layer on top of insurance. It acquired a license, hired actuaries, and became an insurer. That is a different bet than the insurtech bets that came before it: slower, harder to copy, and structurally advantaged if the AI-driven underwriting and claims processing performs as claimed over a full market cycle. The companies that choose to become regulated rather than to work around regulation tend to build slower and last longer.

What these ten companies collectively do not tell us is which of them will still exist and be relevant in 2033. That is always the honest limit of any watch list compiled in real time. What they do tell us is where people with serious technical backgrounds, real money, and demonstrated ability to execute are choosing to spend the next decade of their working lives.

In 2026, those choices are concentrated in AI infrastructure that runs on different physics than GPU clusters, in biological science where AI has genuinely changed what is computationally possible, in physical automation for labor-constrained industries, and in the interpretability and safety tooling that honest observers of AI development know is necessary even when the market has been slow to fund it. That combination of priorities tells you something true about where the next decade of technology is actually headed.