news

Author: Xuushan, Editor: Ivan

Wolves will hunt when necessary, and each hunt is carefully planned. They often follow the prey, letting it run, and then wait for the right opportunity to strike and hit it with one strike.

The same is true for "hunting" in Silicon Valley.In just four days, three tech giants simultaneously took action against generative AI startups. On August 3, Google spent $2.5 billion to "buy back" Character.AI's 30-person team. On August 1, Microsoft officially announced that OpenAI is one of its competitors in search and advertising businesses. On July 31, Canva announced that it would incorporate generative AI startup Leonardo.ai.

Figure: Leonardo.ai announced that it was acquired by Canva Source: Leonardo.ai official website

Technology giants such as Microsoft and Google follow AI startups like a pack of wolves. Once their products are proven by the market, they quickly strike back and choose to acquire or compete with the AI ​​startups to further consolidate their dominance in the industry.

Faced with the encirclement and suppression of technology giants, AI startups have their own choices. Some AI startups are fighting hard, while many AI startups are lining up to be acquired. "A year ago, I could hardly imagine this," said an investor who has been paying close attention to the AI ​​track.

Just a year ago, AI startups were one of the most sought-after companies in Silicon Valley. At the peak of their popularity, investors in Silicon Valley even had to line up to meet with executives of star AI startups. According to a survey by market analysis agency Dealroom, the amount of funding raised by generative AI startups worldwide in 2023 alone exceeded$25 billion, generative AI has become a veritable "money-making" track.

Figure 1: Total VC investment in generative AI startups worldwide Source: Dealroom

Just when everyone thought that the generative AI industry would continue to thrive, many generative AI startups turned around and embraced the thighs of major Internet companies, seeking acquisitions.“Not only have star AI startups received multiple acquisition inquiries, but the mailboxes of major Internet companies are also filled with acquisition invitations from generative AI startups,” a person familiar with the matter told Silicon Rabbit.

What happened to generative AI companies in Silicon Valley that made them so eager to seek acquisitions? How did Internet giants acquire major generative AI startups? Can the story of generative AI continue?The giants of the Internet era are now launching a "siege" against start-ups in the "AI native" era. We also seeThe offensive and defensive dynamics between generative AI companies and Internet giants are changing。

Chain tricks, psychological tactics, turning the tables,

Thirty-six strategies used by giants to suppress AI startups

In this generative AI "encirclement and suppression", technology giants have used "Thirty-Six Strategies" to make multi-faceted strategic arrangements against AI startups.

Among them, the most eye-catching is Microsoft's use of OpenAI"Chain of tricks"。

Microsoft cleverly opened the door to OpenAI by investing $1 billion in an early phase, and subsequently added $2 billion to further solidify the partnership and make Microsoft OpenAI's exclusive cloud computing service provider.

In January 2023, when all technology companies were still watching the development of generative AI, Microsoft once again took decisive action and, through a strategic investment of up to tens of billions of dollars, firmly "bound" itself to OpenAI, the most promising AI startup at the moment, in terms of funding, technology, and future profit sharing. This also made Microsoft one of the most watched technology giants in the industry, and its market value also exceeded the 3 trillion US dollar mark.

This billion-dollar collaboration is also talked about a lot. According to Fortune magazine, this investment includes multiple stages: in the first stage, Microsoft is entitled to 75% of OpenAI's profits until its investment is recovered; in the second stage, when OpenAI's profits reach a certain amount, Microsoft's profit share will be reduced to 49%, and other investors and employees will be entitled to the remaining profits of the company; in the third stage, when OpenAI's profits reach a higher amount, the shares of Microsoft and other investors will be returned to OpenAI's non-profit foundation.

It can be seen that Microsoft has gained a lot of benefits on the surface. Not only can it sell access to GPT through its cloud service Azure, but it can also compete directly with GPT through its upcoming models. If OpenAI develops smoothly, Microsoft will not only be able to quickly recover its tens of billions of dollars invested, but also take nearly half of OpenAI's profits. Judging from the existing news, it is not an exaggeration to say that OpenAI is currently working for Microsoft to "pay off its debts."

Compared with Microsoft, Google's acquisition strategy for Character.AI is more like a carefully planned "psychological attack".

Internally, Google successfully persuaded its former employees, Character.AI founders Noam Shazeer and Daniel De Freitas, to give up their newly founded companies and return key products and technologies to Google by offering generous salary packages and playing the emotional card of their former employer.

Externally, Google not only eliminated a strong generative AI competitor ahead of schedule, but also hit the morale and loyalty of generative AI startup teams. It is worth mentioning the timing of the acquisition - the acquisition price of Character.AI was much lower than the market estimate.

Figure: Character.AI official website

In March this year, Microsoft acquired Inflection Al for US$3 billion, twice its highest valuation of US$1.5 billion. However, five months later, the acquisition price of Character.AI was only US$2.5 billion, half of its highest valuation of US$5 billion.The shrinking valuation of generative AI startups also vaguely reveals that people's views on the trend of generative AI technology are returning to rationality.

This type of CEO acquisition is also known asSynthetic acquisitions。"The main reason for these acquisitions is thatLarge technology companies face regulatory pressure on mergers and acquisitions。”

Investors don't lose a penny in these acquisitions.According to The Information, Character.AI investors have received returns of at least 50% of their original investment.2.5 times。Inflection’s AI investors will still receive 1.1 to 1.5 times their initial investment, in part because the company may not spend all the money it raised. And if Inflection successfully develops new AI products, its original investors may receive additional returns.

"Companies that can do it make it themselves, and companies that can't do it buy it." Compared with the first two technology giants, Canva is more direct, using only one trick“Turning the tables”，The entire team of generative AI startup Leonardo.ai was directly incorporated into the group. Canva, a graphic design giant, has 150 million monthly active users from more than 190 countries around the world, and an average of 200 designs are created every second. The addition of Leonardo.ai has undoubtedly enhanced the competitiveness of Canva's innovative products.

This move not only allows Canva to make up for its shortcomings in the field of generative AI design in a relatively short period of time, but also leverages Leonardo.ai's innovative capabilities and existing market foundation to accelerate the development and iteration of its own products, allowing Canva to stand out in the increasingly competitive AI design tool market.

Not only that, tech giants are also very good at"Alliance"——Through a certain amount of capital injection, generative AI startups will be recruited into their own ecosystem.

At present, the generative AI industry chain mainly covers upstream infrastructure such as data centers and computing power research and development. The midstream is mainly responsible for AI model development and data processing and development tool processing, and then to the downstream diversified AI application market.

Figure: Artificial Intelligence Industry Map

Whether it is NVIDIA + AI big model startups or Microsoft + AI application startups, by building a generative AI ecosystem, technology giants can provide more comprehensive services and solutions through the capabilities of AI startups, thereby enhancing their competitive advantages in the market.

Nvidia is good at using this strategy. It has not only acquired AI management startups Run:ai and deep learning AI startup Deci, but has also invested in 14 generative AI startups including Mistral AI, Cohere, and Together AI.

Figure 1: A summary of generative AI startups deployed by North American tech giants (by Silicon Rabbit)

Microsoft has reached agreements with large language model companies such as Cohere and Mistral AI to introduce them to the Azure platform. Mistral AI will sell its models directly in the Microsoft cloud, becoming the second company after OpenAI to provide commercial AI models on the Azure cloud platform. Subsequently, Cohere will also put the enterprise-level AI models Cohere Command R and Command R+ models into Azure AI as managed services.

In this way, Microsoft can expand the influence of its Azure platform and provide more diverse AI models to meet the needs of different users. On the other hand, Microsoft can also add more new customers to its Azure cloud service, driving the growth of its cloud service business.

Figure 1: Microsoft Azure AI platform

"For startups like Character and Inflection ... (being acquired) is their best option," said Gayatri Sarkar, founder of Advaita Capital, an investment firm focused on the AI ​​track. "Some startups need a lot of money,It would be better for them if they were brought under a larger umbrella。

Guru Chahal, another partner at Lightspeed Venture Partners, said he was not surprised that the AI ​​industry was experiencing significant consolidation. "Not every company is reaching their full potential," said Chahal, whose firm has invested in AI model developer Mistral.“Our thesis has always been that the leading-edge model will be oligopolistic.”

We can see that in this battle for dominance in generative AI, technology giants have strategically encircled AI startups. After such a series of moves, most AI startups have almost become inextricably linked with technology giants, making it even more difficult for them to embark on an independent path.

Funding, talent, and competitiveness: Generative AI startups face three challenges

The "encirclement and suppression" of generative AI startups by technology giants has opened the prelude to the technology giants' ruthless harvesting of AI companies.

Under the "encirclement and suppression", generative AI startups still need to overcome three major challenges.

The first is the financial test.For generative AI under the violent aesthetics of the Scaling law (i.e., as the model, data set size, and the number of floating-point calculations used for training increase, the performance of the model will improve), funding is the most difficult hurdle for every company, and each generative AI company faces different funding challenges.

Star AI startups like OpenAI, which focus on the development of large models, spend money quickly and earn little because the development of large models often requires data collection and processing, model design, training, evaluation and optimization, among which the most expensive part is usually the model training part.

The 2023 Artificial Intelligence Index Report released by Stanford University mentioned that the cost of training GPT-3 with 175 billion parameters in 2020 was about $1.8 million. GPT-3 has 117 times more parameters than GPT-2, and its training cost is alsoIncreased 36 timesIn recent years, the high-frequency and fast-paced release of GPT-4, GPT-4o, and GPT-4o mini also means that OpenAI's R&D costs are constantly rising.

According to a report by The Infomation, OpenAI’s operating costs are estimated to be as high as $8.5 billion, including inference costs, training costs, and labor costs, while its revenue is approximately between $3.8 billion and $4.5 billion, indicating that the company may faceUp to $5 billionhuge losses.

With 3.5 billion in revenue, 5 billion in net loss, and 10 billion in investment "arrears", OpenAI, the most promising startup in the generative AI field, is still burdened with a huge financial burden.

How to calculate the economic accounts is the basis for whether OpenAI can survive after the craze fades.If start-ups are unable to survive financially, it will give Internet giants an opportunity to take action. When starting a business in the generative AI track, "cost-cutting" is basically hopeless, while "open source" often fails on the road to commercialization.

For example, Character.AI focuses on virtual social interaction. Although its users are very active, it faces problems such as users’ unwillingness to pay for its products and difficulty in forming a closed business loop. Currently, it has less than 100,000 subscribers, accounting for less than one thousandth of the total number of users.

Today, brands like Anthropic,Stability AIGenerative AI application companies such as Inflection AI, which are relatively well-known but not as popular as OpenAI, have all encountered similar problems as Character.AI. For example, Anthropic has an annual revenue and expenditure gap of up to $1.8 billion, and Stability AI operates without the support of a tech giant and faces great financial pressure.

Figure 1: Stability AI official website

This has also led to some AI startups currently hoping to be incorporated into large companies and obtain more financial and resource support.

Talent is another challenge facing generative AI companies besides funding and revenue.

The development of generative AI is mainly based on deep learning and NLP. But in fact, it has only been more than seven years since Google proposed the Transformer architecture in 2017. There are not many R&D personnel who are truly rooted in this field, and even fewer of them can earn a million-dollar annual salary. Therefore, each of them is the object of competition among major companies.

Headhunting firm Rora once revealed that OpenAI offers some employees an annual salary of up to $865,000, including a base salary of $665,000 and stock options of $200,000. Not to mention, when OpenAI poached senior researchers from Google, it promised employees an annual salary of tens of millions.

In addition to capital flow and talent competition, startups are also facing a more intense product competition environment.

The speed of releasing large model parameters is getting faster and faster, and there are more and more products of the same type, but the product prices are getting lower and lower. According to incomplete statistics from Silicon Rabbit, in the first seven months of 2024 alone, technology giants have launched11 modelsAI models are spread in all directions.

Google has launched five AI models, including Gemini 1.5 Flash and Pro, the open source model Gemma, the video generation model Veo, and the text image model Imagen. Meta has released the large language model Llama 3.1, the video processing model Segment Anything Model 2 (SAM 2), the hybrid model Meta Chameleon, and the AI ​​music generation model JASCO. Microsoft and Apple have released the Phi-3 series of AI small language models and the Apple Intelligence AI large model, respectively.

Meta releases Llama 3.1

In addition to the number of volumes, big manufacturers are alsoVolume Price. Google Gemini 1.5 Flash model offers a lower price of $0.13 per 1M tokens (0.013/K). AWS offers Titan Text Express and Titan Text Lite models at $0.20 per 8K tokens (0.025/K) and $0.15 per 4K tokens (0.0375/K), respectively. When the industry is competing on price, startups can't stay out of it. OpenAI's GPT-4o model is priced at $2.50 per 128K tokens (approximately 0.0195/K) based on its version on August 6, 2024.

In addition, competition among startups is as fierce as ever, said Lightspeed’s Chahal.“We’re seeing shorter and shorter intervals between funding rounds, and more companies are starting businesses.”

Large investment, difficulty in making profits, and fierce competition have become common problems facing all generative AI startups. Many AI startups have already collapsed in this turbulent market change.

According to Titanium Media,ChatGPTFrom the date of publication (November 30, 2022) to July 29 this year, the number of newly registered domestic artificial intelligence (AI)-related companies that are now in an abnormal state of cancellation, revocation or suspension of business reached 78,612.Nearly 80,000 businesses have disappearedFrom this corner of China, we can also get a glimpse of the difficult living environment of global generative AI startups.

The wolf pack's "encirclement and suppression" began from this moment. However, this "encirclement and suppression" is not a bad thing for the generative AI industry and AI startups. Industry integration and acquisitions are extremely common in the business world, and for giants and startups, it is also a behavior that each party needs.

Layoffs, price increases, cooperation,

AI startups strive to break through

To live or to die, in the business world sometimes there is not enough time to make a choice.

Layoffs are the first step for many AI startups to break through.AI programming unicorn Replit announced a 20% layoff of 30 employees. AI speech recognition software startup Deepgram also announced that it had laid off about 20% of its employees. The CEO of AI chatbot Jasper also announced on LinkedIn that the company is laying off some employees in order to focus on adjusting the direction of the business.

Other AI startups are seizing core market demand to generate more revenue.For example, OpenAI has impacted the traditional search business through SearchGPT and entered a new business track. Midjourney has achieved an annual revenue of $100 million through a paid subscription model, and its AI drawing tool has accumulated nearly 15 million users on Discord. Anthropic has also told some investors that the company's annualized revenue is $100 million and is expected to reach $500 million by the end of this year. In addition, it is reported that Harvey AI's total revenue has exceeded $25 million. Judging from the existing information, some generative AI startups have established a relatively stable way of making profits and have reached the "basic line" to maintain survival.

For example, AI search engine company Perplexity has launched a revenue sharing plan for publishers. Publishers will be paid when Perplexity makes money from instant replies based on their articles. It is reported that OpenAI is also secretly connecting with some Microsoft customers such as some luxury goods companies to further expand its customer base.

Figure 1: Perplexity official website

certainly,More generative AI startups are seeking collaboration with big companiesMicrosoft, Nvidia, and Google have also successively reached certain cooperation platforms with AI startups. As mentioned earlier, Microsoft's Azure AI is a typical representative of them.

Just like the competition, cooperation and win-win in the animal world, enterprises in the business ecosystem are also performing the drama of "survival of the fittest" in such a state.

Judging from the actions of the three giants Microsoft, Google, and Canva at the same time, the generative AI track is undoubtedly at a time of reshuffle. The giants are trying to dominate the market with their resources and influence, while startups are constantly breaking through and challenging the status quo with their flexibility and innovation capabilities.

The two sides seek a balance between cooperation and competition, and jointly promote the development and application of AI technology. This is not only a competition about technology, but also a game about wisdom and foresight.

Interaction at the end of the article:

What do you think of the Silicon Valley AI war? Feel free to share your thoughts in the comments section.