news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author: Shen Siqi

Source: Hard AI

In the summer of 2024, the pace of AI development continues to raise market expectations. In this wave, hundreds of emerging technology companies have emerged in the US tech startup mecca, and the performance of the Magnificent Seven in the capital market has also been rising.

However, in sync with the record-breaking AI capabilities, a tense atmosphere is gradually spreading with the high summer temperatures. AI startups in Silicon Valley and the Bay Area are experiencing a quiet shock.

The incident revolves around a young industry star named David Luan. As the co-founder and CEO of Adept AI, David was passionate about participating in various exchanges and roadshows a few months ago, and he vowed with like-minded practitioners that the disruptive application of AI would challenge the territory of traditional IT giants and open up a whole new world. Then, in recent days, he made an unexpected turn - leading the company's core team to join Amazon collectively.

This explosive news quickly spread in the industry and caused widespread discussion.

The survival dilemma of AI unicorns

To understand David's choice, one may need to put oneself in the shoes of AI unicorns and understand the dilemma they are in. For start-ups engaged in the research and development of large-scale language, image, video and even multimodal models of this generation of AI, the research and development arms race has already become astronomical and is still escalating.

Anthropic's CEO recently stated publicly that the current training cost of AI models has reached the level of US$1 billion, and in the next three years, this figure may soar to US$10 billion or even US$100 billion.

Investments of this magnitude pose an insurmountable hurdle for most startups, which are struggling to balance cash flow as their previous round of funding quickly runs out.

Character.ai is another typical case. Although the company has a large number of active users, its annual revenue is less than 20 million US dollars, and it lacks the ability to expand R&D returns. Industry insiders explained that in the AI ​​industry, having advanced technology does not mean the ability to monetize, and having unique and available closed-source capabilities and a wide range of users does not mean that you can make money. The real challenge is no longer to launch products and find users, but to form a sustainable business model that can cover R&D investment and maintain company operations. "

Product competition is becoming increasingly fierce, user touch costs have soared after rounds of launch competition, and the sharp deterioration of the financing environment has made the situation of AI unicorns even worse.

According to the latest report from Deutsche Bank, in the first quarter of 2024, the total amount of seed and early-stage financing for generative AI startups in the United States was only US$123 million, a plunge of more than two-thirds from the previous quarter. The number of financing transactions also dropped sharply from more than 70 in the previous quarter to 34.

Faced with such high R&D costs, difficult to balance cash flow and an increasingly tight financing environment, David Luan's choice does not seem so surprising.

New model of "alternative acquisition" for large companies

It has always been common for large technology companies to acquire AI startups. Why has the case of Adept AI attracted extraordinary attention? The reason is that Amazon has adopted a new "alternative acquisition" model.

This model, called "reverse acquihire" by The Verge, was first used by Microsoft when it acquired Inflection AI in March this year. Microsoft paid $650 million for technology licensing and recruited the core staff and most of the technical team of Inflection AI to form a new Microsoft AI department.

It is worth noting that in the same month that Inflection AI was "acquired" by Microsoft, Reid Hoffman predicted in a speech that what happened at Inflection would become a "model" for future AI transactions. As an investor in Inflection and a member of Microsoft's board of directors, Hoffman has an insider's perspective on both sides. He observed the entire acquisition process and made this prediction, which now seems quite prescient.

Just a few months later, Amazon followed a similar model in its deal for Adept AI. Rather than acquiring Adept AI outright, they acquired about 66% of the company’s employees, including its core team including co-founder and CEO David Luan, and obtained a non-exclusive license to some of its technology.

Startups can only passively accept such "alternative" acquisitions under layers of risk pressure. However, what is the trend that makes large companies design this option?

Venky Ganesan, managing director of Menlo Ventures, gave a brilliant assessment of this practice: "This is a new way for the 'Tech Big Seven' to make acquisitions, acquiring intellectual property and teams without FTC review or approval."

This model can help large companies quickly acquire top AI talent and technology. Through this "alternative acquisition", large companies can obtain a complete, market-tested AI team at one time, greatly accelerating their own AI research and development process.

In addition, this model can help large companies consolidate their dominant position in the field of AI. Deutsche Bank's report pointed out that Amazon's move is to "reduce its dependence on third-party startups." In fact, Amazon has begun developing its own AI chatbot (code-named Metis) and plans to invest $100 billion in data center construction over the next decade.

But the most important thing, of course, is that it circumvents the antitrust review of the regulatory authorities. After all, "poaching" plus "technology licensing" is significantly different from traditional "acquisitions". Within the scope of authorization of the current federal trade law, it is difficult for the FTC to directly intervene in such business activities.

A helpless choice for AI unicorns?

For David Luan and his Adept AI team, giving up R&D and especially financial independence to join Amazon was not an easy decision, but it was a necessary one in the current market environment.

Adept AI admitted in a company blog: "Continuing with our original plan to both build useful general intelligence and develop multi-agent AI products would require us to spend a lot of attention raising funds for the basic models rather than (spending on) turning our product vision into reality."

At the same time, an Adept AI employee revealed to the media: "In the current environment, independent development is becoming increasingly difficult. Joining Amazon can at least ensure that the technology can continue to develop rather than being shelved due to funding issues."

In fact, the case of Inflection AI has already sounded the alarm for the industry. The Information reported that Inflection AI had been struggling to find an effective business model before being "acquired" by Microsoft.

As far as the US AI industry is concerned, an obvious fact has emerged: the door is closing. Most of the current financial or human investments in generative AI are unlikely to enjoy the same "rocket-like" returns as OpenAI.

You know, according to the data revealed by Sam Altman to employees, OpenAI's annual revenue this year is only US$3.4 billion, far from the tens of billions.

Are there more startups waiting to be sold “alternatively”?

The story of Inflection AI is no longer an isolated case today, and there will be more Adept AIs in the future. In fact, these two isolated acquisitions with the same pattern indicate that the AI ​​industry is facing a round of large-scale consolidation.

Deutsche Bank's report pointed out that in addition to Inflection AI and Adept AI, which have been "acquired", there are many emerging AI teams that have abundant AI talents and are in urgent need of funding, computing resources and customer resources support, such as Cohere, AI21 and Stability AI.

However, potential investors will only be more cautious. Therefore, it is not ruled out that the above-mentioned companies will tend to seek "alternative" cooperation with large technology companies in the future.

This trend of integration has already attracted the attention of regulators. The European Commission previously stated in its review of Microsoft's investment in OpenAI that they want to ensure that similar transactions do not "escape our merger control rules." Although the EU ultimately decided not to launch a formal investigation into the Microsoft-OpenAI transaction, they are still reviewing the exclusivity clauses in the partnership.

Meanwhile, the US Federal Trade Commission (FTC) has reportedly begun an antitrust investigation into Microsoft's "reverse acquisition recruitment" Inflection AI.

Regulators are closely watching the integration trend in the AI ​​industry to prevent various attempts to circumvent antitrust regulation. Under this supervisory pressure, Microsoft and Apple recently announced that they would cancel their original observer seats on the Open AI board to avoid suspicion of control and manipulation by related large companies.

Will this new type of integration stifle innovation? To what extent will the dominance of large technology companies affect the diverse development of AI technology? How should regulators find a balance between protecting the rights of innovators and preventing monopolies? These questions still have no complete answers.

So, the Chinese market...

What impact will this trend have on China's AI industry?

In fact, China's AI financing market is still hot. Baidu, Alibaba, Tencent and other large technology companies are investing heavily in AI technology, and many AI startups have also received considerable financing. However, as the US market indicates, the timeline of hot money flow will test the business model in the next stage.

Whether in the United States or China, emerging AI companies or any startups always have to face the question of whether operating cash flow can cover expenses. Regarding the accounts, there may be miracles that work temporarily, but there is no magic that can be cast for a long time.

The problems currently faced by the US AI industry will also be challenges that the Chinese market must deal with in the future. Just like the hot innovative drug bubble market five or six years ago. Companies that cannot find a sustainable business model will eventually have no choice but to be acquired by large companies, "alternative acquisitions" or "traditional liquidation."

Conclusion

The AI ​​industry is essentially not much different from other industries. New technologies make new routes possible, and startups still need to complete high-quality milestones on the new routes, find a good business model to sell development results, expand their scale in the process of recovering initial investment, and return profits to the market in various forms.

At present, industry consolidation seems to be an inevitable trend. Reid Hoffman's prediction a few months ago is coming true. This "alternative acquisition" model is likely to become the main way of AI industry consolidation in the future. Of course, there is still uncertainty about the regulatory response to this.

For AI startups, pursuing technological innovation is a noble means, and high-quality products and sustainable models are the core of the business. From the perspective of investors and large technology companies, how to obtain truly scarce AI capabilities in an environment of fierce competition and stricter supervision is an important issue in the next stage.