Tech Brew Ride Home - Thu. 08/24 – Nvidia Blows The Roof Off The Place

Episode Date: August 24, 2023

Nvidia gives new meaning to the words “earnings beat.” Meta announces Code Llama. TikTok might start banning links to Amazon. SpaceX wants Starlink to be viable in cities too. And turning thoughts into speech via an AI interface becomes real. Links: Nvidia tops estimates and says sales will jump 170% this quarter, driven by demand for AI chips (CNBC) Meta launches own AI code-writing tool: Code Llama (The Verge) TikTok Shop on Track to Lose More Than $500 Million in U.S. This Year (The Information) Epic offers devs 100 percent of net revenue for six months of EGS exclusivity (Engadget) SpaceX Working with Cloudflare to Speed Up Starlink Service (The Information) Brain implants give a voice to people who cannot speak (FT) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Tech meme right home for Thursday, August 24th, 2020. I'm Brian McCullough today. NVIDIA gives new meaning to the words earnings beat.

META announces Codlama. TikTok might start banning links to Amazon. SpaceX wants Starlink to be viable in cities too, and turning thoughts into speech via an AI interface becomes real. Here's what you miss today in the world of tech. Well, as people were anticipating, Nvidia had an earnings call for the ages yesterday, reporting Q2 revenue up 1001% year over year to $13.51 billion versus the estimated revenue number

was $11.22 billion, so that's quite the beat. Also, data center revenue was up 171% year over year to $10.32 billion. Let me pause for a second and underline that. We've spoken about the data center business and how that was a growth area for Nvidia. Also, how in this new AI era, it's kind of all cloud. You don't tend to run an LLM locally, but in the cloud via, say, chat GPT's APIs. Anyway, data center revenue grew 171% year over year, but that hides the really impressive number, which is that Nvidia grew data center sales 141% quarter over quarter, which is to say, two quarters ago data center sales were $4.3 billion. Last quarter, they were $10.3 billion.

They grew that much in three months. Something, something, everyone's trying to buy their chips to deploy LLMs. Quoting CNBC with more superlatives, remember, they just reported $13.51 billion in revenue for this past quarter. $2 billion more than was estimated, and it looks like they want to do something similar. Next quarter, quote, Nvidia said it expects fiscal third quarter revenue of about six. $16 billion higher than the $12.61 billion forecast by refinative. NVIDIA's guidance suggests sales in the current quarter will grow 170% from the year earlier period. Net income jumped to $6.19 billion or $2.48 a share, which was up from $656 million

or just $26 a share a year earlier. $26 a share, $2.48 a share in a year. new in the second quarter doubled from $6.7 billion a year earlier and increased 88% from the prior period. The world has had something along the lines of about a trillion dollars worth of data centers installed in the cloud, enterprise, and otherwise. InVIDIA CEO Jensen Huang said on a call with analysts that trillion dollars of data centers is in the process of transitioning to accelerated computing and generative AI. On the earnings call, finance chief Colette Kress said the company would not be immediately affected by proposed Biden administration export restrictions on chips.

Given the strength of demand for our products worldwide, we do not anticipate that additional export restrictions on our data center GPUs, if adopted, would have an immediate material impact to our financial results, Crest said. Even before Wednesday's report, Nvidia's stock price had more than tripled for the year, making it the top performer in the S&P 500. It jumped to around $500 after hours, a level that would mark a record if it closes there on Thursday. Its prior closing high was $4.74.

$0.94 on July 18. Invidious performance was driven by its data center business, which includes AI chips as cloud service providers and large consumer internet companies like Alphabet and Meta, snapped up next generation processors. The company reported $10.32 billion in revenue for the group, up 171% year over year and above the $8.03 billion estimate according to street account. Nvidia added that it saw its adjusted gross margins increase 25.3 percentage points to 71.2% because of growth in profit-rich data center sales. The gaming division, which used to be Nvidia's core business, saw revenue increase 22% from a year earlier to $2.49 billion,

topping the $2.38 billion average estimate, end quote. Meta has released Code Lama, a tool built on top of Lama 2 to generate new code and debug human-written work under the same community license as Lama 2, quoting the verge. Code Lama, meta said, can create strings of code from prompts or complete and debug code when pointed to a specific code string. In addition to the basic code llama model,

META released a Python specialized version called Code Lama Python and another version called Code Lama Instruct, which can understand instructions in natural language. According to Meta, each specific version of code Lama is not interchangeable, and the company does not recommend the base code Lama or code Lama Python for natural language instructions. Programmers are already using LLMs to assist in a variety of tasks ranging from writing new software to debugging existing code. Meta said in a blog post,

the goal is to make developer workflows more efficient so they can focus on the most human-centric aspects of their jobs, end quote. Meta claims CodeLama performed better than publicly available LLMs based on benchmark testing, but did not specifically name which models it tested against. The company said CodeLama scored 53.7% on the code benchmark Human Eval and was able to accurately write code based on a text description. Meta will release three sizes of CodeLama and said its smallest size fits on a single GPU for more low latency projects. Code generators have been helping developers work for a while now. GitHub launched Copilot in March, powered by OpenAI's GPT4, to quickly write and check code. GitHub Copilot can also rewrite old code to update it.

Amazon's AWS also has Code Whisper, which also writes checks and updates code. And yes, Google also has a code writing tool in AlphaCode, but that isn't out yet, end quote. As we've learned recently, if it's your social network, you can technically do whatever you want. Sources are telling the information that TikTok plans to ban links to Amazon and other e-commerce sites in order to, shall we say, encourage people to use TikTok shop instead. Quote, TikTok Shop, the app's nascent shopping service, is expected to lose more than $500 million in the U.S. this year, according to a person with direct knowledge of the matter. The loss reflects TikTok's heavy investment in hiring, building a delivery

network and subsidizing merchants that offer double-digit percentage discounts and free shipping. The company can easily afford it. TikTok's parent company ByteDance last year earned $18 billion in operating profits, as the information has previously reported. Still, the enormous investment shows how the Chinese tech giant is continuing to pour money into a U.S. expansion despite the longstanding threat of a government ban. While the prospect of a ban may have receded from the headlines, the threat hasn't disappeared. The expansion, particularly if it succeeds in making TikTok a retailer with the skill to compete with Amazon is likely to reinforce anxieties about bite dance's influence in the U.S. Meanwhile, TikTok has drawn up plans to bar links to outside e-commerce

sites, such as Amazon as part of an effort to force people to use TikTok shop if they want to buy something they see on the app, say, two people familiar with the situation. That would stop creators from promoting furniture, clothing, or kitchen items available on Amazon and other sites. TikTok's plan to bar external e-commerce links is a page out of the playbook of TikTok's Chinese sibling app, Doyen, which has become a shopping powerhouse. In October 2020, Doyen barred links to other Chinese e-commerce sites such as Alibaba's Taubao. It's not clear when TikTok plans to impose the ban. Once links are barred, TikTok users will be only able to buy products they see on creator-made videos directly in the app through TikTok shop. Some of these videos or live streams contain links that

let people buy a product directly, while some sellers and influencers also have TikTok shop storefronts embedded in their profiles. Once TikTok adds a shop tab to its app, merchants that sign up to sell through the service will be able to list their products without a video. Consumers aren't yet spending much money on TikTok shop around $3 to $4 million a day in the U.S. currently up from between half a million and $1 million a day in June, said four people familiar with the situation. By the end of the year, TikTok staffers expect the daily shopping volume could exceed $10 million in the U.S. TikTok shop is further ahead in Southeast Asia, where it has operated since 2021. In that region, TikTok shops' current daily gross merchandise volume is around $50 million to $60 million,

and the company aims to increase that to around $90 million by the end of the year, or nearly half the volume of local rival Shoppy, according to two people familiar with the matter, end quote. Epic Games has announced the Epic First Run program offering developers 100% of net revenue for six months if they launch games exclusively on the Epic Games store, quoting a gadget. The company which says it's launching it. and store has 68 million monthly active users will also give participants of the Epic first-run program extra exposure to get their wares in front of more eyes. The games and apps will receive exclusive badging and spots on the homepage. Epic will include them in various collections and

promotions too. The program will be open to developers and publishers who are releasing eligible products on or after October 16th. A product will be eligible if it hasn't appeared on another third-party PC store or subscription services on said storefronts. Games and apps are, that already have exclusivity deals with Epic aren't eligible. Developers and publishers will still be able to sell games and apps that are included in the program directly to users via their own launchers and websites. They can also list their products on stores such as Green Man Gaming and Humble Store via the Epic Games Store's keyless redemption program. Epic already offered developers and publishers a more generous split of game and app sales than the likes of Valve, which takes a

30% cut of Steam sales. The promise of vacuuming up all net revenue for six months, and getting extra promotion on the Epic Game Store might be compelling enough for more developers and publishers to take the exclusivity plunge there, end quote. Also from the information whose sources are reporting that, SpaceX is working with Cloudflare to increase Starlink's terrestrial network of mini data centers, which could help the service delivers faster speeds, but also increase its potential market footprint. Quote, such improvements could help overcome one obstacle to Starlink's long-term growth, as future bandwidth-hungry applications become commonplace.

Starlink is SpaceX's most ambitious new effort after its rocket launch business, but the service which had 1.5 million subscribers as of May is still relatively small, and at least in areas with larger populations, faces tough competition from terrestrial internet services. More specifics of the arrangement between SpaceX and Cloudflare, including its financial terms, couldn't be learned. SpaceX didn't respond to a request for comment. A Cloudflare spokesperson had no comment. Cloudflare, a company that offers network security and content delivery services, has data centers in 300 cities in more than 100 countries according to its website.

That network helps it offer services that speed up internet connections and approve website performance for its customers. While known for its 4,000 or so satellites circling the globe and low Earth orbit, Starlink also relies on earthbound infrastructure. It uses antenna farms known as ground stations to wire into fiber cables that carry data around the globe. Those fiber connections linked to the mini data centers, which enables Starlinked. to service larger population centers and increase the speed of service for users. But the service has historically struggled to provide the same speed of service to dense urban areas where there are more people competing for bandwidth. That makes Starlink unattractive in those areas, particularly as people in major cities typically have lots of choices for

high-speed internet services, often at lower costs than what Starlink charges. Last June, SpaceX CEO Elon Musk said in a tweet that Starlink was working to address some issues with latency, delays, and network transmissions, which can affect applications like online video games. In response, Cloudflare CEO Matthew Prince said that he would be interested in working with Musk to improve Starlink speeds by putting antennas on the roofs of Cloudflare's data centers. It couldn't be learned whether that is part of the agreement between the two companies, end quote. Finally, today, research teams from the University of California, San Francisco, and Stanford

University have reportedly used AI and electrodes to turn thoughts into speech via a lifelike avatar for the first time. Quoting the Financial Times. The details of both brain computer interfaces were jointly published in the journal Nature on Wednesday. Both studies are a big leap forward towards decoding brain activity with the speed and accuracy required to restore fluent communications to people with voice paralysis, said Frank Willett, one of the Stanford team. The two implants were significantly different in design.

The UCSF team placed a paper-thin rectangle with 253 electrodes on the surface of the cortex to record brain activity from an area known to be critical for speech. The Stanford device inserted two smaller arrays with a total of 128 microelectrodes deeper into the brain. Each team worked with a single volunteer. Stanford with Pat Bennett, 68, who has a myotrophic lateral sclerosis and UCSF with a 47-year-old stroke patient known as Anne. Despite the differences between the implants and research participants, the results from the two studies were broadly similar. They achieved average speech rates of about 60 to 80 words per minute, almost half the speed of a normal conversation, but at least three times faster

than any previous brain computer interface has achieved. Both projects use an artificial intelligence algorithm to decode electrical signals from their subjects' brains, teaching itself to distinguish the distinct pattern associated with individual phenomes, the subunits of speech that form spoken words. The system needed long training sessions, 25 in Bennett's case, each lasting four hours, during which she repeated in her mind different sentences chosen from a large data set of phone conversations. These initial results have proven the concept and eventually technology will catch up to make it easily accessible to people who cannot speak, Bennett wrote. For those who are nonverbal, this means they can stay connected to the bigger world, perhaps continue to work, maintain friends

and family relationships, end quote. The UCSF scientists working with colleagues at UC Berkeley created a personalized voice for Anne based on a recording of her speaking at her wedding. They also created an avatar for her using software that simulates face muscle movements from speech graphics, a company in Edinburgh that uses facial animation software. Signals from Anne's brain, as she tried to speak, were converted into corresponding movements on the avatar's face. When the subject first used the system to speak and move the avatar's face in tandem, I knew this was going to be something that would have a real impact, said Kalo Littlejohn

of UC Berkeley. Much more development work will be needed to translate the laboratory proof of concept into devices simple and safe enough for patients and their carers to operate at home. An important step, the researchers say, will be to produce a wide-rector. wireless version that would not require the user to be wired to the implant, end quote. Quick reminder that the Ride Home AI Fund will be raising money through the end of October. More info and actually the ability to invest right away can all be found at ridehomefund.com or email me at Brian at Ridehomefund.com.

But if you'll recall, the minimum check size to invest in the AI fund is $100,000, while the minimum investment for the Ride Home Rolling Fund is only $4,000. $5,000 a quarter for four quarters or $20,000 in total. I am reminding you of this because the rolling fund has to co-invest in all the deals the AI fund invest in over the next two years. The AI fund has already written three checks so far to some exciting AI companies. So if you want your money to be a part of these deals via the rolling fund, you have until August 31st, one week from today to invest in the rolling fund. Now, it is a rolling fund so you can come in at any time. You can come in next month, next year, still be a part of the AI deals as long

as they last. But if you want it to be a completist and participate in all the deals, head over to ridehomefund.com before the end of the month to invest in the rolling fund option. Talk to you tomorrow.