Author | Flagship

Editor | Ivan

Locking in the AI race with capital, is there still an opportunity for latecomers?

OpenAI is building its AI empire through investment.

In February 2023, when OpenAI Startup Fund invested an undisclosed amount in the open-source database company EdgeDB, almost no media paid attention to this deal. However, a year later, the company's graph relational database technology has become a core tool for GPT-4 training data management—this is precisely OpenAI Startup Fund's typical strategy: binding key nodes with small amounts of capital to reconstruct the underlying rules of the AI technology stack.

This is the fund launched by the general large model company in 2021. According to data disclosed by third-party data agencies such as PitchBook and Crunchbase, the initial size of the OpenAI Startup Fund was approximately $175 million, followed by an additional $114 million raised through five independent SPVs (Special Purpose Vehicles), bringing the total assets under management to nearly $300 million.

Among mainstream venture capital funds that easily raise billions of dollars, the fundraising amount of OpenAI Startup Fund, whether $300 million or $175 million, is not large, even less than the single round of financing of some unicorn companies.

However, OpenAI's golden brand Endowed with this "small fund" extraordinary significance. Although the book value of funds is not large, the resources it can leverage cannot be underestimated. For OpenAI and its backers, it is an important piece in the grand chess of reshaping the global AI industry.

"Gateway" competition

The official website of OpenAI Startup Fund mentions that the establishment of this fund is based on a belief—powerful AI systems will trigger a "Cambrian explosion" of new products, services, and applications. We believe that the companies with the most lasting impact will utilize emerging AI capabilities to thoroughly transform existing markets and create entirely new ones, rather than merely enhancing existing possibilities.

OpenAI Startup Fund promises to invest early in a small number of startups that drive positive AI impact on the world and profoundly change human life. Over the past three years, the fund has invested in over 20 AI startups, covering a complete ecosystem from medical records (e.g., Ambience Healthcare), chip design (e.g., Atomic Semi), to legal technology (e.g., Harvey AI).

Table: OpenAI Startup Fund Investments

Compiled by: Flagship

"Healthcare, law, education, energy & infrastructure, science" are considered by OpenAI Startup Fund as areas where AI can have transformative impacts and are also the most profitable tracks in human society.

In the Internet era, to seize profitable tracks, one must first seize traffic gateways and make platform-based layouts. Although in the future AI era, the business model is likely to evolve from the "platform economy" to the "large model economy", in the short term, the demand for traffic gateways in commercial layouts will not diminish temporarily.

The same is true for OpenAI, so OpenAI Startup Fund, which focuses on vertical applications, came into being.

In fact, since OpenAI's success has been recognized worldwide, many AI companies with model capabilities inferior to OpenAI have begun to transform their competitive strategies into niche markets, positioning their core competitiveness in vertical fields. They hope to control niche market gateways through proprietary models and product capabilities, thereby protecting their original market share from being eroded by OpenAI's general large models. However, OpenAI is not in a hurry to compete for the role of porter in niche markets. As early as the GPT-3 era, OpenAI had already positioned itself as a "water seller" with APIs as its core products and services. However, in the fiercely competitive AI large model market with numerous players, there are too many people who want to sell water. How can they maintain their exclusive supplier status?

The best way is to open several more "water pipe connection" companies.

After all, creating good products like ChatGPT in every niche field is no easy feat. Although OpenAI has raised considerable funds in recent years, the company is still in its growth stage and faces enormous competitive pressure in the AI market, making it difficult to divert too much attention to niche markets.

Therefore, investing in niche startup companies through the OpenAI Startup Fund to cultivate "technology agents" in vertical fields has become the key to OpenAI's future commercial layout.

In addition, domain data generated by invested companies (such as medical conversations from Ambience and legal contracts from Harvey) can feed back vertical scenario data to OpenAI, forming a closed loop of "model empowering applications → applications generating data → data optimizing models".

When vertical AI applications become dependent on OpenAI's foundational models, hardware vendors adapt to its computing power needs, and developer ecosystems form path dependencies, OpenAI achieves a triple lock on the entire industry.

Technological standard lock-in: The product design of invested companies is deeply coupled with GPT interface specifications.

Data flow lock-in: Industry data generated at the application layer feeds back into model training, forming data monopoly.

Capital network lock-in: Investors' interests are deeply bound to the OpenAI ecosystem, forming a moat.

In the future where AI large models continue to evolve, this composite lever of technology, capital, and data will make it easier for OpenAI to control the power structure of the global AI industry.

Small funds leverage large layouts

OpenAI Startup Fund tends to invest in early-stage companies from seed round to Series B, which is both a cause and an effect of their modest fund size.

A moderate fund size forces the team to screen projects more rigorously, formulating an investment strategy of "few but precise" and "small but beautiful". On the one hand, they must find projects that synergize with OpenAI's technological roadmap (law, healthcare, education). On the other hand, they also require projects to quickly validate and find rigid demand tracks with limited funds.

Investing less means less risk loss when encountering issues. After all, it is difficult to guarantee the success of startups. For example, Ghost Autonomy, an autonomous driving company that OpenAI Startup Fund invested $5 million in, only lasted a year before failing. Concentrating investment amounts in the millions to tens of millions of dollars (e.g., $8 million in Anysphere's seed round) also avoids to some extent the dilution of resources caused by scattershot investments.

At the same time, the fund has also established five Special Purpose Vehicles (SPVs) to customize investment strategies for different fields (such as robotics, healthcare), both diversifying risks and accurately capturing opportunities in niche tracks.

Although OpenAI's venture capital fund is not large in scale, after being labeled with OpenAI, it is likely to achieve the "miracle" of spending $1 as if it were $10 or $100 through technology binding agreements and ecological revenue sharing.

Firstly, most of the fund's investors are external injections from OpenAI partners rather than OpenAI's own funds, avoiding the consumption of the parent company's R&D resources while building an open ecosystem through multi-party interest binding. This also gives OpenAI greater "technological discourse power" within it, realizing "technological securitization" in a certain sense.

The first incubation camp, Converg, after the establishment of OpenAI Startup Fund, promised to provide companies with "early access to OpenAI models and programming resources customized for AI companies".

Secondly, startups can also receive resource support from OpenAI partners. For example, Microsoft Azure cloud service resources and sales channels within the Microsoft ecosystem, or SB OpenAI Japan, a joint venture established by OpenAI and SoftBank. The joint venture plans to invest $3 billion annually to integrate AI capabilities into SoftBank's portfolio companies (such as Arm, PayPay) and plans to build data centers in Japan. Invested companies of OpenAI Startup Fund may have the opportunity to cooperate deeply with their Japanese "backer" to quickly open the Asian market with lower costs and SoftBank's localized resources (government relations, corporate clients), forming regional suppression against Anthropic and DeepSeek.

Finally, the fund has also made some defensive layouts for OpenAI's future development path, such as the chip company Atomic Semi. This company focuses on simplifying traditional chip manufacturing processes, aiming to reduce prototype development time from months to hours. In 2023, Atomic Semi received a $15 million investment from OpenAI Startup Fund.

Although only $15 million, it is likely to play a significant role in the future.

OpenAI's model training and inference have always been highly dependent on NVIDIA's CUDA software ecosystem and high-performance GPUs (such as H100, B200). Both OpenAI and the entire industry understand this. Whether developing chips in-house or investing in Atomic Semi, it is difficult to overcome the triple barriers of performance, cost, and ecosystem. Even achieving partial substitution will take at least 5-10 years. During this period, OpenAI will still deeply rely on NVIDIA.

However, investing in Atomic Semi can still hedge against NVIDIA's computing power monopoly risk to a certain extent and may help them reduce their dependence on supply chains such as TSMC in the future, reserve capacity for the AGI era, and enhance OpenAI's long-term bargaining power with NVIDIA.

A "three-legged" profit model

OpenAI Startup Fund has an even more important function. It may help OpenAI, which is transforming into a profit-making company, establish a more comprehensive and integrated three-tier revenue system in the future, transforming small funds into value "hunters" across all links of the industry chain.

Tier 1 revenue: financial returns from equity appreciation. Among the many companies invested in by OpenAI Startup Fund, there are not a few outstanding performers. Among them, Harvey AI stands out. This is an AI technology company in the legal field. In 2024, the company's ARR (Annual Recurring Revenue) exceeded $50 million, and the ARR from the second half of 2024 to early 2025 exceeded $100 million, covering 235 customers in 42 countries, including most of the top ten law firms in the United States.

Harvey AI announced the completion of a $300 million Series D funding led by Sequoia Capital US in early 2025, with a valuation exceeding $3 billion, triple that of July 2024. As an early investor, OpenAI Startup Fund naturally made a fortune in this deal. Tier 2 revenue: stable income from the API ecosystem. Although OpenAI's funds and incubators provide invested companies with advanced AI usage rights and low-cost API interfaces during the early stages, in the long run, API costs will become one of the stable operating costs for these companies, bringing long-term stable business to OpenAI. An example can intuitively demonstrate the profitability of this business. The spoken language learning application Speak announced the completion of a $78 million Series C funding round at the end of 2024, with a valuation of $1 billion. As of December 2024, the download volume of the Speak app had exceeded 10 million times, with each user spending about 10-20 minutes per day.

Speak Korean version interface, source: Y Combinator

Speak's marketing copy mentions the hope that users can "say 100 sentences in 20 minutes". Calculating at 10 tokens per sentence, Speak may consume up to 10 billion tokens per day. Currently, OpenAI's GPT-4o mini voice API is priced at $10 for 1 million input tokens and $20 for 1 million output tokens. With this conversion, Speak is bound to incur considerable expenses on OpenAI's API.

Tier 3 revenue: ecological premium. The preferential APIs and technical support provided by OpenAI Startup Fund to invested companies are not only benefits but also constraints for these companies. Since the products are trained and fine-tuned using OpenAI's underlying models from the outset, they are likely to be simultaneously bound to Microsoft Azure cloud services.

The KnowHow and knowledge bases accumulated over a long period by the invested companies will also be precipitated in the form of models. In the future, if AI suppliers or cloud service providers need to be changed, it is likely that retraining, tuning, and even reopening one's own products will be required. Considering the migration costs and the potential uncertainties that may affect the products, the invested companies will be bound to the OpenAI ecosystem for a long time.

Today, Harvey AI, Speak, Figure AI, and others have rapidly grown into unicorns. The position of these companies in the industry has further consolidated OpenAI's influence in these sectors, which will undoubtedly become an ecological premium for OpenAI's own valuation in the future.

In October 2024, OpenAI was valued at approximately $157 billion, and by early 2025, its valuation had risen to $260 billion, with the potential to exceed $300 billion in the next 12-24 months. Despite being besieged by DeepSeek for two months, OpenAI's valuation continued to rise rapidly, which may also be related to the steady development of its GPT ecosystem.

The Achilles' heel of capital

The OpenAI Startup Fund is a well-designed move in OpenAI's grand strategy. It not only helps OpenAI consolidate its ecological construction and enhance its profitability but also helps OpenAI CEO Sam Altman personally resolve some "troubles" related to "related-party transactions" to a certain extent.

In 2018, Sam Altman participated in the seed round of funding for chip company Rain AI through personal funds, investing over $1 million. Rain AI promotes its NPU chips as offering up to 100 times the computing power of traditional GPUs and providing 10,000 times the energy efficiency in training.

Sam Altman, source: openai.fund

Soon after, OpenAI followed up with an investment in Rain AI. In 2019, OpenAI signed a non-binding letter of intent promising to purchase $51 million worth of Rain AI's NPU products after they were launched. OpenAI publicly stated that it hoped to alleviate its dependence on NVIDIA GPUs and reduce costs through such cooperation.

Since Sam Altman was already the CEO of OpenAI at that time, this chip procurement inevitably raised questions about "related-party transactions". However, if it were an investment by OpenAI's fund, there would be no issue of related-party transactions.

However, for startups, being associated with OpenAI through the OpenAI Startup Fund has both advantages and disadvantages.

First, to enjoy the valuation boost brought by OpenAI, one must bear the "pressure" of market expectations for OpenAI. Since the technical architecture of the invested companies is deeply tied to OpenAI's model interfaces, such as the programming tool Cursor, which completely relies on GPT-4 to generate code, its technical roadmap lacks autonomous iteration capabilities.

Given the current less optimistic progress in OpenAI's model development (such as repeated delays in the release of GPT-5), companies within the ecosystem may likely be affected in their future product upgrade plans and even valuation due to the model's capabilities falling short of expectations. Perhaps recognizing this issue, Figure AI, a robotics company that recently conducted in-depth technical cooperation with OpenAI, publicly announced a strategic transformation: no longer relying on partnerships but pursuing vertically integrated solutions with tightly coupled hardware and software.

Figure AI's robotic products, source: Figure AI

Brett Adcock, the company's CEO, emphasized in an interview with TechCrunch that general AI models like OpenAI's are insufficient for effectively scaling "embodied AI".

Furthermore, under the dividend of OpenAI's rapidly rising valuation, it is inevitable that the valuations and financing of related companies will accumulate a significant amount of "capital bubble". In some fields with greater competitive pressure, companies with larger bubbles are also more vulnerable.

Ghost Autonomy is an example of excessive bubbles. In a 2023 funding round, the OpenAI Startup Fund invested $5 million in this company. However, just one year later, the company ceased operations due to uncertainties about long-term profitability and the need for significant investments in independent development and commercialization. Before shutting down, the company had completed eight funding rounds, raising nearly $239 million and being valued at $700 million. Yet, this was still not enough to bring its products to market.

Currently, the ecological synergy model formed by OpenAI and its venture capital fund is a recognized form for general model companies to expand their AI ecosystem. Fundraising is fast, and companies are eager to participate. Even competitors like Anthropic are "replicating" this strategy.

In July 2024, Anthropic announced the establishment of a fund called "Anthology Fund" in partnership with its significant investor Menlo Ventures. This fund, with a size of approximately $100 million, is intended to invest in pre-seed, seed, and Series A AI startups. By the end of 2024, the fund had invested in 18 startups covering high-compliance scenarios such as healthcare and enterprise services.

Combining OpenAI's tie-up with Microsoft Azure Cloud and Anthropic's tie-up with AWS Cloud, it is not difficult to see that the core competitive logic of these two companies seems to be increasingly converging: the barrier value of a closed-loop technology far exceeds short-term financial returns, and the right to set rules has become a more critical determinant of success than market share.

References: OpenAI’s startup empire: The companies backed by its venture fund (TechCrunch), Softbank set to invest $40 billion in OpenAI at $260 billion valuation, sources say (CNBC), OpenAI Agreed to Buy $51 Million of AI Chips From a Startup Backed by CEO Sam Altman (Wired), Figure AI: Robot startup terminates partnership with OpenAI, relies on its own LLMs (Trending Topics), OpenAI Startup Fund raises $44M in its largest SPV yet (TechCrunch), OpenAI Fund official website