Forcing journalists to turn over a source’s identity can have a real chilling effect on newsgathering. People will feel discouraged from talking to journalists, and that harms the public’s ability to be informed about things that affect them. We also increasingly consume our news from independent journalists and smaller outlets, which might not have the legal resources to fight a government subpoena for their records. The PRESS Act would provide the same blanket protections to journalists across the United States, and also covers independent journalists and outlets that publish information in the public interest.
Not surprisingly, TikTok is in second place next to Temu with 33.23 million downloads. The short-form video app is popular among users aged 18 to 24 for several reasons, the most obvious being that it serves as a go-to source for fast-paced, easily digestible content. Interestingly enough, the Gen Z-dominated app is also preferred as a search engine over Google.
Another major video platform, YouTube, remains popular among Gen Z young adults. Over a span of around 10 months, the mobile app amassed 14.03 million new installs, Appfigures noted. While this number is not as high as TikTok’s, it still indicates significant engagement among this group.
At this point, the PRESS Act already has bipartisan support in the Senate, with Sens. Ron Wyden, Lindsey Graham, Mike Lee, and Dick Durbin as co-sponsors. The bill is just waiting for a final vote on the Senate floor, with weeks to go before the bill expires at the end of the congressional session.
Since Threads was the most recent to launch, it makes sense that it has more new downloads. Additionally, there seems to be renewed interest in Facebook following the redesign aimed at engaging this social media-savvy generation. (While we cannot definitively determine whether the update directly influenced the spike in downloads, it’s worth noting the potential correlation.)
Google, another tech giant, has also amassed a decent amount of downloads so far this year. The search engine app saw 17.65 million installs, whereas other Google-owned products Chrome and Meet saw 10.19 million installs and 9.63 million, respectively. Meanwhile, Google Drive garnered 7.22 million downloads, and Google Photos had 6.79 million.
While the bill doesn’t affect the tech industry directly, as a news outlet, we’re in favor of protecting and building upon press freedoms. Some of TechCrunch’s most read and impactful reporting has come from readers like you, who have reached out and tipped us off to corporate wrongdoing, unearthing mismanagement in the startup world, detailing human rights abuses, and revealing major breaches and data spills, cyberattacks and criminality that otherwise might have gone unreported.
Another not-so-shocking discovery was that OpenAI’s ChatGPT continues to be favored among 18- to 24-year-olds, gaining 24.63 million app installs. With many Gen Z young adults either starting college or entering the early stages of their careers, ChatGPT has emerged as a favored resource for educational purposes and career guidance. A survey conducted by Best Colleges revealed that 43% of college students have admitted to using ChatGPT or similar AI tools.
Another AI-powered app getting traction was AI study companion Gauth. The app, developed by TikTok parent company ByteDance, achieved 8.37 million new installs. While it has existed since 2021, Gauth saw a surge in downloads earlier this year, becoming the second most downloaded education app for iOS devices in the U.S. in April.
Filmmaker Ole Ege's film The 1972 brothel set off a wave of bedside films that drew hundreds of thousands of Danes in cinemas. The 'Bordellet' was also present at the Cannes Film Festival, where it was sold abroad.
It suddenly becomes legal to go in and watch that type of movie because they are not so hardcore and vulgar, and that allows mom and dad to go in and watch porn movies together.
For example, the film Agent 69 Jensen in Scorpio's Sign was seen by 450,000 in the cinema.
For Gen Z, photo sharing and social media apps are integral to their daily lives. So it’s safe to say that a lot of these types of platforms would make it to the list.
In addition to Threads and Instagram ranking highly, ByteDance’s CapCut is another platform favored by Gen Z young adults. The video editing app is known for its range of effects and filters, making it easy to upload edited content to TikTok and other compatible platforms. Data indicates that the app saw 21.72 million new downloads.
Others on the list include Snapchat with 19.16 million installs, followed by Telegram with 13.12 million, Pinterest at 8.23 million, Reddit at 8.06 million, and X at 7.58 million.
So far this year, Netflix has been the top streaming service to attract the most 18- to 24-year-olds with 15.67 million installs. Prime Video and Disney+ are not far behind, with 12.86 million and 11.68 million downloads, respectively. Max, Peacock, and Tubi each have around 7 million downloads or more.
And it appears the music streaming app capturing lots of interest among Gen Z young adults is none other than the leading platform, Spotify. This app has attracted 10.45 million downloads, highlighting its significant appeal to this age group.
Encore processes over 50,000 searches per month and is seeing 26% month-on-month growth for searches and 15% growth in clicks.
The startup currently relies on affiliate shares for revenue generation. However, the company is also experimenting with a $3 per month subscription providing unlimited searches with advanced models, finding items by uploading images, and offering support via email and chat.
There are in some sense two types of LLMs - frontier models - at the cutting edge of performance (think GPT-4 vs other models until recently), and everything else. In 2021 I wrote that I thought the frontier models market would collapse over time into an oligopoly market due to the scale of capital needed. In parallel, non-frontier models would more commodity / pricing driven and have a stronger opensource presence (note this was pre-Llama and pre-Mistral launches).
Ruber noted that sentence length can vary, sometimes by a lot. Some people just key in a simple sentence like “Show me jeans” while others write a detailed description like “I am a 6’2″ person who skies and looking for skier pants under $100 with no big logos on them.”
The secondhand retail market is on an upward growth curve, with analysts projecting it to reach $73 billion in the U.S. and $350 billion globally by 2028. A report by online thrift store ThredUp notes that online secondhand resale would account for half of the secondhand market in 2025.
Frontier LLMs are likely to be an oligopoly market. Current contenders include closed source models like OpenAI, Google, Anthropic, and perhaps Grok/X.ai, and Llama (Meta) and Mistral on the open source side. This list may of course change in the coming year or two. Frontier models keep getting more and more expensive to train, while commodity models drop in price each year as performance goes up (for example, it is probably ~5X cheaper to train GPT-3.5 equivalent now than 2 years ago)
Swiggy’s IPO will also show how willing investors are to bet on business models that prioritize growth over profits amid challenging global conditions.
For Dutch investor Prosus, Swiggy’s listing could deliver a three-fold return. It will also be the venture firm’s biggest hit from India, where its $1 billion-plus gains from Byju’s have all but evaporated. Accel is expected to see a more than 35-fold return, one of its largest in the past five years.
As model scale has gotten larger, funding increasingly has been primarily coming from the cloud providers / big tech. For example, Microsoft invested $10B+ in OpenAI, while Anthropic raised $7B between Amazon and Google. NVIDIA is also a big investor in foundation model companies of many types. The venture funding for these companies in contrast is a tiny drop in the ocean in comparison. As frontier model training booms in cost, the emerging funders are largely concentrated amongst big tech companies (typically with strong incentives to fund the area for their own revenue - ie cloud providers or NVIDIA), or nation states wanting to back local champions (see eg UAE and Falcon). This is impacting the market and driving selection of potential winners early.
Swiggy’s Instamart is among the top three quick-commerce businesses in the country, which promise deliveries of groceries, wellness and beauty products and much more within 10 minutes. Whether these companies will be able to revolutionize the broader retail market in India remains to be seen, but they have already captured 56% of the online grocery delivery market from e-commerce firms, according to JPMorgan.
Quick-commerce firms such as Instamart, Zomato-owned BlinkIt, Zepto, BigBasket, and Minutes are changing consumer behavior in urban Indian cities, home to about 80 million people. Together, they are on track to record sales of more than $6 billion this year, according to TechCrunch estimates.
It is important to note that the scale of investments being made by these cloud providers is dwarfed by actual cloud revenue. For example, Azure from Microsoft generates $25B in revenue a quarter. The ~$10B OpenAI investment by Microsoft is roughly 6 weeks of Azure revenue. AI is having a big impact on Azure revenue revently. Indeed Azure grew 6 percentage points in Q2 2024 from AI - which would put it at an annualized increase of $5-6B (or 50% of its investment in OpenAI! Per year!). Obviously revenue is not net income but this is striking nonetheless, and suggests the big clouds have an economic reason to fund more large scale models over time.
In parallel, Meta has done outstanding work with Llama models and recently announced $20B compute budget, in part to fund massive model training. I posited 18 months ago that an open source sponsor for AI models should emerge, but assumed it would be Amazon or NVIDIA with a lower chance of it being Meta. (Zuckerberg & Yann Lecunn have been visionary here).
There are a lot of potential areas that Tako could expand into in the future, like the vast world of employee benefits. Gadotti said the company does plan to expand as it grows into building more features like instant payments.
In addition to competing with legacy companies like ADP, there are multiple other HR tech startups in the country like Gupy and Caju, which are both more focused on other areas within HR and employee management. But if Tako expands into these areas, which it likely will, these companies could also become strong competitors.
Are cloud providers king-making a handful of players at the frontier and locking in the oligopoly market via the sheer scale of compute/capital they provide? When do cloud providers stop funding new LLM foundation companies versus continuing to fund existing? Cloud providers are easily the biggest funders of foundation models, not venture capitalists. Given they are constrained in M&A due to FTC actions, and the revenue that comes from cloud usage, it is rational for them to do so. This may lead / has led to some distortion of market dynamics. How does this impact the long term economics and market structure for LLMs? Does this mean we will see the end of new frontier LLM companies soon due to a lack of enough capital and talent for new entrants? Or do they keep funding large models hoping some will convert on their clouds to revenue?
“The strategy we took is that we are not trying to boil the ocean,” Gadotti said. “We want to start in a segment we know before venturing into industrials or more complex areas. We are starting in more simpler segments; as the company evolves, we are going to more complex segments in the future.”
Tako is emerging from stealth with a sizable $13.2 million seed round co-led by Ribbit Capital and Andreessen Horowitz. The round also included ONEVC and the founders of Ramp. Gadotti said the company plans to put the majority of capital toward research and development in addition to doubling or tripling headcount on its R&D team.
Tako is emerging from stealth with a sizable $13.2 million seed round co-led by Ribbit Capital and Andreessen Horowitz. The round also included ONEVC and the founders of Ramp. Gadotti said the company plans to put the majority of capital toward research and development in addition to doubling or tripling headcount on its R&D team.
There are a lot of potential areas that Tako could expand into in the future, like the vast world of employee benefits. Gadotti said the company does plan to expand as it grows into building more features like instant payments.
xAI launched Grok-2 in August with image generation capacities, backed by Black Forest Labs’ FLUX.1 model. Late last month, the company also gave the model the ability to understand images.
All these features were available only to Premium and Premium+ users until now. By opening up Grok to free users, xAI is possibly looking for a more significant userbase and faster feedback cycle for its products, so that it can better compete with other models on the market like ChatGPT, Claude, and Gemini.
Does OSS models flip some of the economics in AI from foundation models to clouds? Does Meta continue to fund OS models? If so, does eg Llama-N catch up to the very frontier? A fully open source model performing at the very frontier of AI has the potential to flip a subportion the economic share of AI infra from LLMs towards cloud and inference providers and decreases revenue away from the other LLM foundation model companies. Again, this is likely an oligopoly market with no singular winner (barring AGI), but has implications on how to think about the relative importance of cloud and infrastructure companies in this market (and of course both can be very important!).
According to a researcher who goes by Swak on X, there are limits on the usage for now: 10 queries per two hours with the Grok-2 model, 20 queries per two hours with the Grok-2 mini model, and three image analysis questions per day.
To use Grok for free, your account should be at least seven days old and have a phone number linked to it.
Upstox has been using Equal for about a year and is processing around 350,000 transactions a month. Before that, Kumar said, the platform was relying on existing ID-verification providers.
“Equal has been able to aggregate across a slew of different APIs and ensures very high uptime between all those different connections,” he said.
How do we think about speed and price vs performance for models? One could imagine extremely slow incredibly performant models may be quite valuable if compared to normal human speed to do things. The latest largest Gemini models seem to be heading in this direction with large 1 million+ token context windows a la Magic, which announced a 5 million token window in June 2023. Large context windows and depth of understanding can really change how we think about AI uses and engineering. On the other side of the spectrum, Mistral has shown the value of small, fast and cheap to inference performant models. The 2x2 below suggests a potential segmentation of where models will matter most.
Equal is not alone in the space, as the market already has players such as Perfios (backed by Warburg Pincus and Teachers’ Venture Growth), IDfy (backed by TransUnion), and Bureau (backed by GMO VenturePartners). However, Reddy told TechCrunch that unlike the competition, Equal plays the role of an aggregator and partners even with some of its competitors.
Ravi Kumar, co-founder and CEO of Upstox, who has also invested in Equal’s maiden round and is one of the early customers for its identity verification and account aggregator, told TechCrunch that it’s the cost and uptime that gives the trading platform a reason not to look for building a similar tech in-house.
“Data sharing is still a major problem in this country if it’s not done digitally with consent,” Keshav Reddy, the son of GVK Group’s vice chairman GV Sanjay Reddy, told TechCrunch.
Reddy founded Equal with former Swiggy engineering director Rajeev Ranjan after moving back to India from the U.S.
For over the last two years, Reddy bootstrapped Equal, and the startup has added more than 350 customers, including State Bank of India, HDFC Bank, ICICI Bank, Reliance Jio, Airtel, Uber, and Zoom.
How do architectures for foundation models evolve? Do agentic models with different architectures subsume some of the future potential of LLMs? When do other forms of memory and reasoning come into play?
Do governments back (or direct their purchasing to) regional AI champions?Will national governments differentially spend on local models a la Boeing vs Airbus in aerospace? Do governments want to support models that reflect their local values, languages, etc? Besides cloud providers and global big tech (think also e.g. Alibaba, Rakuten etc) the other big sources of potential capital are countries. There are now great model companies in Europe (e.g. Mistral), Japan, India, UAE, China and other countries. If so, there may be a few multi-billion AI foundation model regional companies created just off of government revenue.
The startup has now raised a Series A round of $10 million at a post-money valuation of $80 million to scale its operations, expand the product suite, and forge strategic partnerships. The round was led by Prosus Ventures, along with Tomales Bay Capital and Reddy himself, and saw participation from other investors, including Blume Ventures, DST Global, Gruhas VC, and Quona VC.
Equal is not alone in the space, as the market already has players such as Perfios (backed by Warburg Pincus and Teachers’ Venture Growth), IDfy (backed by TransUnion), and Bureau (backed by GMO VenturePartners). However, Reddy told TechCrunch that unlike the competition, Equal plays the role of an aggregator and partners even with some of its competitors.
What happens in China? One could anticipate Chinese LLMs to be backed by Tencent, Alibaba, Xiaomi, ByteDance and others investing in big ways into local LLMs companies. China’s government has long used regulatory and literal firewalls to prevent competition from non-Chinese companies and to build local, government supported and censored champions. One interesting thing to note is the trend of Chinese OSS models. Qwen from Alibaba for example has moved higher on the broader LMSYS leaderboards.
“There is a robust regulatory regime that exists in place today that’s been developed over 30 years,” and it’s well-equipped to construct new policies for AI and other tech. It’s true, at the federal level alone, regulatory bodies include everything from the Federal Communications Commission to the House Committee on Science, Space, and Technology. When TechCrunch asked Casado on Wednesday after the election if he stands by this opinion — that AI regulation should follow the path already hammered out by existing regulatory bodies — he said he did.
How good does Google get? Google has the compute, scale, talent to make amazing things and is organized and moving fast. Google was always the worlds first AI-first company. Seems like a wild card.
The counterargument — and one several people in the audience brought up — was that the world didn’t really see the types of harms that the internet or social media could do before those harms were upon us. When Google and Facebook were launched, no one knew they would dominate online advertising or collect so much data on individuals. No one understood things like cyberbullying or echo chambers when social media was young.
Advocates of AI regulation now often point to these past circumstances and say those technologies should have been regulated early on.
Infra companies
There are a few types of infrastructure companies with very different uses. For example, Braintrust provides eval, prompt playgrounds, logging and proxies to help companies move from “vibe based” analysis of AI to data driven. Scale.ai and others play a key role in data labeling, fine tuning, and other areas. A number of these have open but less existential questions (for example how much of RLHF turns into RLAIF).
“You have to have a notion of marginal risk that’s different. Like, how is AI today different than someone using Google? How is AI today different than someone just using the internet? If we have a model for how it’s different, you’ve got some notion of marginal risk, and then you can apply policies that address that marginal risk,” he said.
“I think we’re a little bit early before we start to glom [onto] a bunch of regulation to really understand what we’re going to regulate,” he argues.
While this particular state law is dead, the fact it existed still bothers Casado. He is concerned that more bills, constructed in the same way, could materialize if politicians decide to pander to the general population’s fears of AI, rather than govern what the technology is actually doing.
He understands AI tech better than most. Before joining the storied VC firm, Casado founded two other companies, including a networking infrastructure company, Nicira, that he sold to VMware for $1.26 billion a bit over a decade ago. Before that, Casado was a computer security expert at Lawrence Livermore National Lab.
The biggest uncertainties and questions in AI infra have to do with the AI Cloud Stack and how it evolves. It seems like there are very different needs between startups and enterprises for AI cloud services. For startups, the new cloud providers and tooling (think Anyscale, Baseten, Modal, Replicate, Together, etc) seem to be taking a useful path resulting in fast adoption and revenue growth.
For enterprises, who tend to have specialized needs, there are some open questions. For example:
Does the current AI cloud companies need to build an on-premise/BYOC/VPN version of their offerings for larger enterprises? It seems like enterprises will optimize for (a) using their existing cloud marketplace credits which they already have budget for, to buy services (b) will be hesitant to round trip out from where their webapp / data is hosted (ie AWS, Azure, GCP) due to latency & performance and (c) will care about security, compliance (FedRAMP, HIPAA etc). The short term startup market for AI cloud may differ from long term enterprise needs.
He says that many proposed AI regulations did not come from, nor were supported by, many who understand AI tech best, including academics and the commercial sector building AI products.
“You have to have a notion of marginal risk that’s different. Like, how is AI today different than someone using Google? How is AI today different than someone just using the internet? If we have a model for how it’s different, you’ve got some notion of marginal risk, and then you can apply policies that address that marginal risk,” he said.
How much of AI cloud adoption is due to constrained GPU / GPU arb? In the absence of GPU on the main cloud providers companies are scrambling to find sufficient GPU for their needs, accelerating adoption of new startups with their own GPU clouds. One potential strategy NVIDIA could be doing is preferentially allocating GPU to these new providers to decrease bargaining power of hyperscalers and to fragment the market, as well as to accelerate the industry via startups.
And this is why we’ve seen a steady rise in funding initiatives come to the fore. This includes reactive programs, such as 2022’s Big Tech-driven $30 million pledge to bolster open source security in the wake of the Log4Shell security flaw that wreaked havoc on the software supply chain. But we’re also seeing more proactive efforts, driven from all corners of industry.
Silicon Valley VC Sequoia Capital launched an open source fellowship in 2023 to support project maintainers with equity-free capital to cover living expenses for up to 12 months. Its inaugural fellow was Colombian software developer Sebastián Ramírez Montaño, creator of FastAPI, an open source web framework for building APIs.
When does the GPU bottleneck end and how does that impact new AI cloud providers? It seems like an end to GPU shortages on the main clouds would be negative for companies whose only business is GPU cloud, while those with more tools and services should have an easier transition if this were to happen.
In February, Sequoia revealed it would start accepting applications from any developer leading an open source project, with plans to provide funding for up to three qualifying projects annually. Nine months on, and the first two fellows from Sequoia’s expanded program have now been revealed: Chatbot Arena, a popular open source AI model benchmarking tool used by many of the industry’s biggest names, including OpenAI, Meta, and Google; and vLLM, an open source library focused on memory management to power faster and cheaper LLM serving.
What else gets consolidated into AI clouds? Do they cross sell embeddings & RAG? Continuous updates? Fine tuning? Other services? How does that impact data labelers or others with overlapping offerings? What gets consolidated directly into model providers vs via the clouds?
Chatbot Arena, which spun out of a broader research organization called LMSYS, is the handiwork of doctorate students Wei-Lin Chiang and Anastasios Angelopoulos from Berkeley’s Sky Computing Lab. With north of 1 million monthly users, Chatbot Arena is all about helping LLM developers validate claims around their models’ performance, while anyone can test these models and vote for their preferences. Companies such as OpenAI often share versions of their models with the Chatbot Arena team ahead of the models’ release to help fine-tune things before their formal launch.
It is important to note there are really 2 market segments in the AI cloud world (a) startups (b) mid-market and enterprise. It seems likely that “GPU only” business model default works with the startup segment(who have fewer cloud needs), but for large enterprises adoption may be more driven by GPU cloud constraints on major platforms. Do companies providing developer tooling, API endpoints, and/or specialized hardware, or other aspects morph into two other analogous models - (a) “Snowflake/Databricks for AI” model or (b) “Cloudflare for AI”? If so, which ones adopt
Forcing journalists to turn over a source’s identity can have a real chilling effect on newsgathering. People will feel discouraged from talking to journalists, and that harms the public’s ability to be informed about things that affect them. We also increasingly consume our news from independent journalists and smaller outlets, which might not have the legal resources to fight a government subpoena for their records. The PRESS Act would provide the same blanket protections to journalists across the United States, and also covers independent journalists and outlets that publish information in the public interest.
Not surprisingly, TikTok is in second place next to Temu with 33.23 million downloads. The short-form video app is popular among users aged 18 to 24 for several reasons, the most obvious being that it serves as a go-to source for fast-paced, easily digestible content. Interestingly enough, the Gen Z-dominated app is also preferred as a search engine over Google.
Another major video platform, YouTube, remains popular among Gen Z young adults. Over a span of around 10 months, the mobile app amassed 14.03 million new installs, Appfigures noted. While this number is not as high as TikTok’s, it still indicates significant engagement among this group.
At this point, the PRESS Act already has bipartisan support in the Senate, with Sens. Ron Wyden, Lindsey Graham, Mike Lee, and Dick Durbin as co-sponsors. The bill is just waiting for a final vote on the Senate floor, with weeks to go before the bill expires at the end of the congressional session.
Since Threads was the most recent to launch, it makes sense that it has more new downloads. Additionally, there seems to be renewed interest in Facebook following the redesign aimed at engaging this social media-savvy generation. (While we cannot definitively determine whether the update directly influenced the spike in downloads, it’s worth noting the potential correlation.)
Google, another tech giant, has also amassed a decent amount of downloads so far this year. The search engine app saw 17.65 million installs, whereas other Google-owned products Chrome and Meet saw 10.19 million installs and 9.63 million, respectively. Meanwhile, Google Drive garnered 7.22 million downloads, and Google Photos had 6.79 million.
While the bill doesn’t affect the tech industry directly, as a news outlet, we’re in favor of protecting and building upon press freedoms. Some of TechCrunch’s most read and impactful reporting has come from readers like you, who have reached out and tipped us off to corporate wrongdoing, unearthing mismanagement in the startup world, detailing human rights abuses, and revealing major breaches and data spills, cyberattacks and criminality that otherwise might have gone unreported.
Another not-so-shocking discovery was that OpenAI’s ChatGPT continues to be favored among 18- to 24-year-olds, gaining 24.63 million app installs. With many Gen Z young adults either starting college or entering the early stages of their careers, ChatGPT has emerged as a favored resource for educational purposes and career guidance. A survey conducted by Best Colleges revealed that 43% of college students have admitted to using ChatGPT or similar AI tools.
Another AI-powered app getting traction was AI study companion Gauth. The app, developed by TikTok parent company ByteDance, achieved 8.37 million new installs. While it has existed since 2021, Gauth saw a surge in downloads earlier this year, becoming the second most downloaded education app for iOS devices in the U.S. in April.
Filmmaker Ole Ege's film The 1972 brothel set off a wave of bedside films that drew hundreds of thousands of Danes in cinemas. The 'Bordellet' was also present at the Cannes Film Festival, where it was sold abroad.
It suddenly becomes legal to go in and watch that type of movie because they are not so hardcore and vulgar, and that allows mom and dad to go in and watch porn movies together.
For example, the film Agent 69 Jensen in Scorpio's Sign was seen by 450,000 in the cinema.
For Gen Z, photo sharing and social media apps are integral to their daily lives. So it’s safe to say that a lot of these types of platforms would make it to the list.
In addition to Threads and Instagram ranking highly, ByteDance’s CapCut is another platform favored by Gen Z young adults. The video editing app is known for its range of effects and filters, making it easy to upload edited content to TikTok and other compatible platforms. Data indicates that the app saw 21.72 million new downloads.
Others on the list include Snapchat with 19.16 million installs, followed by Telegram with 13.12 million, Pinterest at 8.23 million, Reddit at 8.06 million, and X at 7.58 million.
So far this year, Netflix has been the top streaming service to attract the most 18- to 24-year-olds with 15.67 million installs. Prime Video and Disney+ are not far behind, with 12.86 million and 11.68 million downloads, respectively. Max, Peacock, and Tubi each have around 7 million downloads or more.
And it appears the music streaming app capturing lots of interest among Gen Z young adults is none other than the leading platform, Spotify. This app has attracted 10.45 million downloads, highlighting its significant appeal to this age group.
Encore processes over 50,000 searches per month and is seeing 26% month-on-month growth for searches and 15% growth in clicks.
The startup currently relies on affiliate shares for revenue generation. However, the company is also experimenting with a $3 per month subscription providing unlimited searches with advanced models, finding items by uploading images, and offering support via email and chat.
There are in some sense two types of LLMs - frontier models - at the cutting edge of performance (think GPT-4 vs other models until recently), and everything else. In 2021 I wrote that I thought the frontier models market would collapse over time into an oligopoly market due to the scale of capital needed. In parallel, non-frontier models would more commodity / pricing driven and have a stronger opensource presence (note this was pre-Llama and pre-Mistral launches).
Ruber noted that sentence length can vary, sometimes by a lot. Some people just key in a simple sentence like “Show me jeans” while others write a detailed description like “I am a 6’2″ person who skies and looking for skier pants under $100 with no big logos on them.”
The secondhand retail market is on an upward growth curve, with analysts projecting it to reach $73 billion in the U.S. and $350 billion globally by 2028. A report by online thrift store ThredUp notes that online secondhand resale would account for half of the secondhand market in 2025.
Frontier LLMs are likely to be an oligopoly market. Current contenders include closed source models like OpenAI, Google, Anthropic, and perhaps Grok/X.ai, and Llama (Meta) and Mistral on the open source side. This list may of course change in the coming year or two. Frontier models keep getting more and more expensive to train, while commodity models drop in price each year as performance goes up (for example, it is probably ~5X cheaper to train GPT-3.5 equivalent now than 2 years ago)
Swiggy’s IPO will also show how willing investors are to bet on business models that prioritize growth over profits amid challenging global conditions.
For Dutch investor Prosus, Swiggy’s listing could deliver a three-fold return. It will also be the venture firm’s biggest hit from India, where its $1 billion-plus gains from Byju’s have all but evaporated. Accel is expected to see a more than 35-fold return, one of its largest in the past five years.
As model scale has gotten larger, funding increasingly has been primarily coming from the cloud providers / big tech. For example, Microsoft invested $10B+ in OpenAI, while Anthropic raised $7B between Amazon and Google. NVIDIA is also a big investor in foundation model companies of many types. The venture funding for these companies in contrast is a tiny drop in the ocean in comparison. As frontier model training booms in cost, the emerging funders are largely concentrated amongst big tech companies (typically with strong incentives to fund the area for their own revenue - ie cloud providers or NVIDIA), or nation states wanting to back local champions (see eg UAE and Falcon). This is impacting the market and driving selection of potential winners early.
Swiggy’s Instamart is among the top three quick-commerce businesses in the country, which promise deliveries of groceries, wellness and beauty products and much more within 10 minutes. Whether these companies will be able to revolutionize the broader retail market in India remains to be seen, but they have already captured 56% of the online grocery delivery market from e-commerce firms, according to JPMorgan.
Quick-commerce firms such as Instamart, Zomato-owned BlinkIt, Zepto, BigBasket, and Minutes are changing consumer behavior in urban Indian cities, home to about 80 million people. Together, they are on track to record sales of more than $6 billion this year, according to TechCrunch estimates.
It is important to note that the scale of investments being made by these cloud providers is dwarfed by actual cloud revenue. For example, Azure from Microsoft generates $25B in revenue a quarter. The ~$10B OpenAI investment by Microsoft is roughly 6 weeks of Azure revenue. AI is having a big impact on Azure revenue revently. Indeed Azure grew 6 percentage points in Q2 2024 from AI - which would put it at an annualized increase of $5-6B (or 50% of its investment in OpenAI! Per year!). Obviously revenue is not net income but this is striking nonetheless, and suggests the big clouds have an economic reason to fund more large scale models over time.
In parallel, Meta has done outstanding work with Llama models and recently announced $20B compute budget, in part to fund massive model training. I posited 18 months ago that an open source sponsor for AI models should emerge, but assumed it would be Amazon or NVIDIA with a lower chance of it being Meta. (Zuckerberg & Yann Lecunn have been visionary here).
There are a lot of potential areas that Tako could expand into in the future, like the vast world of employee benefits. Gadotti said the company does plan to expand as it grows into building more features like instant payments.
In addition to competing with legacy companies like ADP, there are multiple other HR tech startups in the country like Gupy and Caju, which are both more focused on other areas within HR and employee management. But if Tako expands into these areas, which it likely will, these companies could also become strong competitors.
Are cloud providers king-making a handful of players at the frontier and locking in the oligopoly market via the sheer scale of compute/capital they provide? When do cloud providers stop funding new LLM foundation companies versus continuing to fund existing? Cloud providers are easily the biggest funders of foundation models, not venture capitalists. Given they are constrained in M&A due to FTC actions, and the revenue that comes from cloud usage, it is rational for them to do so. This may lead / has led to some distortion of market dynamics. How does this impact the long term economics and market structure for LLMs? Does this mean we will see the end of new frontier LLM companies soon due to a lack of enough capital and talent for new entrants? Or do they keep funding large models hoping some will convert on their clouds to revenue?
“The strategy we took is that we are not trying to boil the ocean,” Gadotti said. “We want to start in a segment we know before venturing into industrials or more complex areas. We are starting in more simpler segments; as the company evolves, we are going to more complex segments in the future.”
Tako is emerging from stealth with a sizable $13.2 million seed round co-led by Ribbit Capital and Andreessen Horowitz. The round also included ONEVC and the founders of Ramp. Gadotti said the company plans to put the majority of capital toward research and development in addition to doubling or tripling headcount on its R&D team.
Tako is emerging from stealth with a sizable $13.2 million seed round co-led by Ribbit Capital and Andreessen Horowitz. The round also included ONEVC and the founders of Ramp. Gadotti said the company plans to put the majority of capital toward research and development in addition to doubling or tripling headcount on its R&D team.
There are a lot of potential areas that Tako could expand into in the future, like the vast world of employee benefits. Gadotti said the company does plan to expand as it grows into building more features like instant payments.
xAI launched Grok-2 in August with image generation capacities, backed by Black Forest Labs’ FLUX.1 model. Late last month, the company also gave the model the ability to understand images.
All these features were available only to Premium and Premium+ users until now. By opening up Grok to free users, xAI is possibly looking for a more significant userbase and faster feedback cycle for its products, so that it can better compete with other models on the market like ChatGPT, Claude, and Gemini.
Does OSS models flip some of the economics in AI from foundation models to clouds? Does Meta continue to fund OS models? If so, does eg Llama-N catch up to the very frontier? A fully open source model performing at the very frontier of AI has the potential to flip a subportion the economic share of AI infra from LLMs towards cloud and inference providers and decreases revenue away from the other LLM foundation model companies. Again, this is likely an oligopoly market with no singular winner (barring AGI), but has implications on how to think about the relative importance of cloud and infrastructure companies in this market (and of course both can be very important!).
According to a researcher who goes by Swak on X, there are limits on the usage for now: 10 queries per two hours with the Grok-2 model, 20 queries per two hours with the Grok-2 mini model, and three image analysis questions per day.
To use Grok for free, your account should be at least seven days old and have a phone number linked to it.
Upstox has been using Equal for about a year and is processing around 350,000 transactions a month. Before that, Kumar said, the platform was relying on existing ID-verification providers.
“Equal has been able to aggregate across a slew of different APIs and ensures very high uptime between all those different connections,” he said.
How do we think about speed and price vs performance for models? One could imagine extremely slow incredibly performant models may be quite valuable if compared to normal human speed to do things. The latest largest Gemini models seem to be heading in this direction with large 1 million+ token context windows a la Magic, which announced a 5 million token window in June 2023. Large context windows and depth of understanding can really change how we think about AI uses and engineering. On the other side of the spectrum, Mistral has shown the value of small, fast and cheap to inference performant models. The 2x2 below suggests a potential segmentation of where models will matter most.
Equal is not alone in the space, as the market already has players such as Perfios (backed by Warburg Pincus and Teachers’ Venture Growth), IDfy (backed by TransUnion), and Bureau (backed by GMO VenturePartners). However, Reddy told TechCrunch that unlike the competition, Equal plays the role of an aggregator and partners even with some of its competitors.
Ravi Kumar, co-founder and CEO of Upstox, who has also invested in Equal’s maiden round and is one of the early customers for its identity verification and account aggregator, told TechCrunch that it’s the cost and uptime that gives the trading platform a reason not to look for building a similar tech in-house.
“Data sharing is still a major problem in this country if it’s not done digitally with consent,” Keshav Reddy, the son of GVK Group’s vice chairman GV Sanjay Reddy, told TechCrunch.
Reddy founded Equal with former Swiggy engineering director Rajeev Ranjan after moving back to India from the U.S.
For over the last two years, Reddy bootstrapped Equal, and the startup has added more than 350 customers, including State Bank of India, HDFC Bank, ICICI Bank, Reliance Jio, Airtel, Uber, and Zoom.
How do architectures for foundation models evolve? Do agentic models with different architectures subsume some of the future potential of LLMs? When do other forms of memory and reasoning come into play?
Do governments back (or direct their purchasing to) regional AI champions?Will national governments differentially spend on local models a la Boeing vs Airbus in aerospace? Do governments want to support models that reflect their local values, languages, etc? Besides cloud providers and global big tech (think also e.g. Alibaba, Rakuten etc) the other big sources of potential capital are countries. There are now great model companies in Europe (e.g. Mistral), Japan, India, UAE, China and other countries. If so, there may be a few multi-billion AI foundation model regional companies created just off of government revenue.
The startup has now raised a Series A round of $10 million at a post-money valuation of $80 million to scale its operations, expand the product suite, and forge strategic partnerships. The round was led by Prosus Ventures, along with Tomales Bay Capital and Reddy himself, and saw participation from other investors, including Blume Ventures, DST Global, Gruhas VC, and Quona VC.
Equal is not alone in the space, as the market already has players such as Perfios (backed by Warburg Pincus and Teachers’ Venture Growth), IDfy (backed by TransUnion), and Bureau (backed by GMO VenturePartners). However, Reddy told TechCrunch that unlike the competition, Equal plays the role of an aggregator and partners even with some of its competitors.
What happens in China? One could anticipate Chinese LLMs to be backed by Tencent, Alibaba, Xiaomi, ByteDance and others investing in big ways into local LLMs companies. China’s government has long used regulatory and literal firewalls to prevent competition from non-Chinese companies and to build local, government supported and censored champions. One interesting thing to note is the trend of Chinese OSS models. Qwen from Alibaba for example has moved higher on the broader LMSYS leaderboards.
“There is a robust regulatory regime that exists in place today that’s been developed over 30 years,” and it’s well-equipped to construct new policies for AI and other tech. It’s true, at the federal level alone, regulatory bodies include everything from the Federal Communications Commission to the House Committee on Science, Space, and Technology. When TechCrunch asked Casado on Wednesday after the election if he stands by this opinion — that AI regulation should follow the path already hammered out by existing regulatory bodies — he said he did.
What happens with X.ai? Seems like a wild card.
How good does Google get? Google has the compute, scale, talent to make amazing things and is organized and moving fast. Google was always the worlds first AI-first company. Seems like a wild card.
The counterargument — and one several people in the audience brought up — was that the world didn’t really see the types of harms that the internet or social media could do before those harms were upon us. When Google and Facebook were launched, no one knew they would dominate online advertising or collect so much data on individuals. No one understood things like cyberbullying or echo chambers when social media was young.
Advocates of AI regulation now often point to these past circumstances and say those technologies should have been regulated early on.
Infra companies
There are a few types of infrastructure companies with very different uses. For example, Braintrust provides eval, prompt playgrounds, logging and proxies to help companies move from “vibe based” analysis of AI to data driven. Scale.ai and others play a key role in data labeling, fine tuning, and other areas. A number of these have open but less existential questions (for example how much of RLHF turns into RLAIF).
“You have to have a notion of marginal risk that’s different. Like, how is AI today different than someone using Google? How is AI today different than someone just using the internet? If we have a model for how it’s different, you’ve got some notion of marginal risk, and then you can apply policies that address that marginal risk,” he said.
“I think we’re a little bit early before we start to glom [onto] a bunch of regulation to really understand what we’re going to regulate,” he argues.
While this particular state law is dead, the fact it existed still bothers Casado. He is concerned that more bills, constructed in the same way, could materialize if politicians decide to pander to the general population’s fears of AI, rather than govern what the technology is actually doing.
He understands AI tech better than most. Before joining the storied VC firm, Casado founded two other companies, including a networking infrastructure company, Nicira, that he sold to VMware for $1.26 billion a bit over a decade ago. Before that, Casado was a computer security expert at Lawrence Livermore National Lab.
The biggest uncertainties and questions in AI infra have to do with the AI Cloud Stack and how it evolves. It seems like there are very different needs between startups and enterprises for AI cloud services. For startups, the new cloud providers and tooling (think Anyscale, Baseten, Modal, Replicate, Together, etc) seem to be taking a useful path resulting in fast adoption and revenue growth.
For enterprises, who tend to have specialized needs, there are some open questions. For example:
Does the current AI cloud companies need to build an on-premise/BYOC/VPN version of their offerings for larger enterprises? It seems like enterprises will optimize for (a) using their existing cloud marketplace credits which they already have budget for, to buy services (b) will be hesitant to round trip out from where their webapp / data is hosted (ie AWS, Azure, GCP) due to latency & performance and (c) will care about security, compliance (FedRAMP, HIPAA etc). The short term startup market for AI cloud may differ from long term enterprise needs.
He says that many proposed AI regulations did not come from, nor were supported by, many who understand AI tech best, including academics and the commercial sector building AI products.
“You have to have a notion of marginal risk that’s different. Like, how is AI today different than someone using Google? How is AI today different than someone just using the internet? If we have a model for how it’s different, you’ve got some notion of marginal risk, and then you can apply policies that address that marginal risk,” he said.
How much of AI cloud adoption is due to constrained GPU / GPU arb? In the absence of GPU on the main cloud providers companies are scrambling to find sufficient GPU for their needs, accelerating adoption of new startups with their own GPU clouds. One potential strategy NVIDIA could be doing is preferentially allocating GPU to these new providers to decrease bargaining power of hyperscalers and to fragment the market, as well as to accelerate the industry via startups.
And this is why we’ve seen a steady rise in funding initiatives come to the fore. This includes reactive programs, such as 2022’s Big Tech-driven $30 million pledge to bolster open source security in the wake of the Log4Shell security flaw that wreaked havoc on the software supply chain. But we’re also seeing more proactive efforts, driven from all corners of industry.
Silicon Valley VC Sequoia Capital launched an open source fellowship in 2023 to support project maintainers with equity-free capital to cover living expenses for up to 12 months. Its inaugural fellow was Colombian software developer Sebastián Ramírez Montaño, creator of FastAPI, an open source web framework for building APIs.
When does the GPU bottleneck end and how does that impact new AI cloud providers? It seems like an end to GPU shortages on the main clouds would be negative for companies whose only business is GPU cloud, while those with more tools and services should have an easier transition if this were to happen.
In February, Sequoia revealed it would start accepting applications from any developer leading an open source project, with plans to provide funding for up to three qualifying projects annually. Nine months on, and the first two fellows from Sequoia’s expanded program have now been revealed: Chatbot Arena, a popular open source AI model benchmarking tool used by many of the industry’s biggest names, including OpenAI, Meta, and Google; and vLLM, an open source library focused on memory management to power faster and cheaper LLM serving.
How do new AI ASICS like
Groq
impact AI clouds?
What else gets consolidated into AI clouds? Do they cross sell embeddings & RAG? Continuous updates? Fine tuning? Other services? How does that impact data labelers or others with overlapping offerings? What gets consolidated directly into model providers vs via the clouds?
Chatbot Arena, which spun out of a broader research organization called LMSYS, is the handiwork of doctorate students Wei-Lin Chiang and Anastasios Angelopoulos from Berkeley’s Sky Computing Lab. With north of 1 million monthly users, Chatbot Arena is all about helping LLM developers validate claims around their models’ performance, while anyone can test these models and vote for their preferences. Companies such as OpenAI often share versions of their models with the Chatbot Arena team ahead of the models’ release to help fine-tune things before their formal launch.
It is important to note there are really 2 market segments in the AI cloud world (a) startups (b) mid-market and enterprise. It seems likely that “GPU only” business model default works with the startup segment(who have fewer cloud needs), but for large enterprises adoption may be more driven by GPU cloud constraints on major platforms. Do companies providing developer tooling, API endpoints, and/or specialized hardware, or other aspects morph into two other analogous models - (a) “Snowflake/Databricks for AI” model or (b) “Cloudflare for AI”? If so, which ones adopt