Executive summary: The post argues that big tech firms like Google, Facebook, Microsoft, and Amazon are acquiring many AI startups, which reduces competition and innovation in the long run.
Key points:
Since 2010, big tech firms have acquired dozens of AI startups, creating "kill zones" that make it unprofitable to invest in new AI companies.
This startup acquisition trend peaked around 2018 and has declined since then, indicating dampened innovation and investment.
Though mergers and acquisitions have short-term benefits, long-term they reduce competition and profits for new entrants.
Big tech's large P/E ratios show markets expect huge future profits from AI, despite little revenue so far.
Regulators should scrutinize these acquisitions more and make big tech prove they won't harm competition.
Big tech partnerships and partial ownership of top AI firms like OpenAI and Anthropic also reduce competition.
Allowing AI startups to grow independently could have created more competition and billionaire founders.
Common ownership by institutional investors in big tech firms mitigates pressures to reduce "moonshot" R&D spending.
However, it also reduces competition and innovation compared to independent AI firms.
Given national security issues, regulators should investigate implications of foreign ownership for firms like Arm.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, andcontact us if you have feedback.
This is the third in a sequence of posts taken from my recent report: Why Did Environmentalism Become Partisan?
Summary
Rising partisanship did not make environmentalism more popular or politically effective. Instead, it saw flat or falling overall public opinion, fewer major legislative achievements, and fluctuating executive actions.
Public Opinion...
I think right now EAs might be making a significant mistake by paying insufficient attention to the political realm. As EAs we tend to figure out what’s most impactful for us to work on and focus hard. That’s great! But there are various actions that are ‘non-delegatable’ - the extent to which an individual can do the action is limited (like voting, going to a protest, making hard money contributions to particular campaigns). It might be useful if we were all more in the habit of doing variou...
New Video from AI in Context: The Fall and Rise of Sam Altman
If you want to skip straight to the video, here it is!
AI in Context is excited to be back with our fourth video! For those just hearing from us, we make videos for 80,000 Hours, telling stories about transformative AI...
'It is hard to favor unspecified changes in the rules. I don't think Google should be able to buy Waze, nor should Facebook buy Instagram.' Richard Thaler, Chicago Booth, Nobel laureate
'We need to increase scrutiny of acquisitions of related businesses, e.g. WhatsApp by Facebook' José Scheinkman, Columbia
Since 2010, in a winner-takes-all market, all Big Tech firms have quickly bought double-digits of competing AI startups. This creates 'kill zones' which makes investing in them unprofitable.[1] Indeed, worldwide, VCs funded 500 new AI startups in 2013, then 1200 in 2018, but only 746 by 2021. Likewise, US VC funding peaked in 2018 with new 500 AI startups; by 2021, it was only 300.[2] Start- and scaleups using AI are valued at $2.3T, down 26% in 2021.[3] This is not in sync with the general macroeconomic trends.[4]
Startups won't adapt to a new platform if bought so quickly, making it harder for them to win users.[5] As a rule, conglomerate M&A involving ecosystems reduce profits on entry.[6] Thus, while M&A has short-term benefits, long-term, it harms competition (also see major cases before the EC[7]).
Meanwhile, Big Tech's large P/E ratios show that, despite relatively little revenue so far, markets expect huge profits. Regulators should assess the long-term effects of M&A and shift the burden of proof to firms to show that M&A won't harm the market.
Indeed, we have oligopol where smaller AI firms are forced to pair with Big Tech:
Microsoft bought 49% of OpenAI for just $13B,[8] despite ChatGPT beingthe fastest growing product ever[9] (though profits are capped at $1.3T i.e. a 100x return). They also partner by using GPT in MS products and provide cloud services.
Google owns DeepMind and bought 10% of Anthropic for $.3B, with which it has a partnership with Anthropic to train, scale and deploy its AI systems.
Amazon has implemented Anthropic's language model in their cloud and might have bought 49% of it for $4B.[10] Google and others might invest another $2B.[11]
Facebook plans to offer free commercial AI models to pressure Google and OpenAI.[12] In turn, as competition intensifies, OpenAI readies open-source models.[13]
Anthropic raised $5.5B.[14] (note that the Amazon investments might include billions in AWS compute credits)
Inflection raised some of its $1.5B from Microsoft and Nvidia.
Tiger Global Management, Index Ventures, Inovia Capital
Adept
$415M
Spark Capital, Greylock, General Catalyst, Addition
Stability AI
$89M
Coatue, Lightspeed Venture Partners
Given Big Tech's trillion dollar valuations, they bought up large parts of the top AI firms (e.g. DeepMind, OpenAI, and Anthropic), relatively cheaply. If the market would be more competitive, smaller AI firms like DeepMind and Anthropic could have used their superior AI to just buy the compute and data they needed, implement their software on existing open ecosystems and pull ahead of Big Tech. Now, neither Demis Hassabis nor Dario Amodei nor Sam Altman are billionaires.
Also, Big Tech has common ownership in several of these AI startups (e.g. Google owns both DeepMind and parts of Anthropic, Microsoft owns shares in OpenAI and inflection). In turn, Big Tech is owed in large part by institutional investors (e.g. 75% of Microsoft is owned by institutional investors).
Such common ownership can have anti-competitive effects and reduce innovation.[15] Recently, institutional investors have pressured Big Tech to spend less on 'moonshot R&D'.[16] But when large investors own technologically related firms it mitigates this problem as their firms act in their interest and can profit by innovating via R&D as a 'public good to their competitors' that might otherwise be unprofitable.[17] When technology spillovers are relatively large, an increase from the 25th to the 75th percentile of common ownership is associated with an increase of +13% in citation-weighted patents.[18]
Feature/Company
Microsoft
Google
Amazon
Facebook
AI CIS Strategy
Frenemies
University
Secrecy
Application-centered
AI Conference Presentations
➕➕➕
➕➕➕
➕
➕
Participation in AI Conference Committees
➕
➕➕➕
➕➕
➕➕
Content of AI Research
General topics with focus on AI and functional applications for language
Maximum diversity with general and specific AI, including reinforcement learning
Highly diversified, skewed towards AI for language with a focus on time series and transfer learning
Few direct links; specific focus on 'action recognition' in computer vision
Acquisitions
➕➕
➕➕➕
➕
➕➕
Top Investor
➕➕➕
➕➕
➕
-
AI Patents (Count)
➕ (less important now)
➕➕➕
➕
-
AI Patents (Content)
Focus on virtual assistants and healthcare; more general machine learning terms
Focus on general machine learning, computer storage (possibly cloud-related), and autonomous vehicles
Most diverse in terms of AI functional applications
Focus on AI related to existing platforms and the Metaverse; uses multi-terms associated
Double Affiliations
➕➕➕ (less US concentrated; significant in China)
➕➕➕ (highly US concentrated)
➕
➕
Job Posts
➕➕
➕➕
➕➕➕
➕
AI CIS Space
Central, global, geopolitically strategic (connecting China and the West)
Central, global (mainly outside Asia)
Limited to leading AI organizations doing frontier research
Narrow focus driven by application/platform-specific AI
AI CIS Scope
General research with a focus on generative AI and reinforcement learning; more application-focused than Amazon
Similar to Microsoft but more focused than Amazon
Diverse in applications, no explicit focus on generative models or reinforcement learning; applies frontier AI when economically beneficial
Focuses on developing AI for specific applications and platforms
While Facebook has a narrow, applied focus on AI for its platforms, Microsoft, Google and Amazon build a more generic AI corporate strategies connected to their clouds:[20]
Google's 'University' strategy emulates top universities, developing a cutting-edge AI CIS with a central place outside Asia. But, like universities, we don't know how Google will profit from AI R&D.
Microsoft's 'Frenemies' strategy. MS has successfully integrated into its CIS rivals globally. In the AI frontier research network, it is the bridging organization between Asia, especially China, and the rest of the world. Microsoft controls not by buying but also by building a CIS with more organizations that are de jure independent but de facto controlled (and sometimes highly funded) by Microsoft (cf OpenAI)
Amazon 'secrecy' strategy. Amazon has developed what looks like the most diverse AI strategy in terms of functional applications, privileging secrecy and highly connected to its businesses. Secrecy tries to create and apply frontier AI to benefit users concretely, which then turns into more profits and data.
Regulators have also stopped Nvidia from buying the chip manufacturer Arm on national security grounds since the chip industry is critical for international security: the chip export ban on China is 'the most aggressive US foreign policy of the last 20 years'[21]. Regulators must work with other departments (e.g. defense) to investigate international security implications of Arm's future IPO (e.g. foreign ownership etc.).
Executive summary: The post argues that big tech firms like Google, Facebook, Microsoft, and Amazon are acquiring many AI startups, which reduces competition and innovation in the long run.
Key points:
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.