Tech Brew Ride Home - Wed. 07/10 – Why Everyone Is Fleeing OpenAI’s Board

Episode Date: July 10, 2024

Suddenly nobody wants to be on OpenAI’s board, even as observers. I’ll tell you why. More on how AI has thrown everybody’s carbon neutral plans into chaos. A monster raise in the AI and robotics space. And a16z has found a way to get chips into the hands of their AI startups. Links: Microsoft and Apple drop OpenAI seats amid antitrust scrutiny (FT) Microsoft gives up observer seat on OpenAI board (Axios) Google Is No Longer Claiming to Be Carbon Neutral (Bloomberg) Amazon Says It Reached a Climate Goal Seven Years Early (NYTimes) Microsoft is hiking the price of Xbox Game Pass Ultimate and launching a new ‘Standard’ tier (The Verge) This $1.5 Billion AI Company Is Building A ‘General Purpose Brain’ For Robots (Forbes) Andreessen Horowitz Is Building a Stash of More Than 20,000 GPUs to Win AI Deals (The Information) 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 Wednesday, July 10th, 2024. I'm Brian McCullough today. Suddenly, nobody wants to be on open AI's board even as observers. I'll tell you why. More on how AI has thrown everybody's carbon neutral plans into chaos. A monster rays in the AI and robotics space and A16Z has

found a way to get chips into the hands of their AI startups. Here's what you miss today in the world of tech. Some days, the stories just cluster themselves into topics even if I don't plan it that way. And today is one of those days. I don't know why, but a bunch of topics have grouped themselves together even when they deal with completely different companies. For example, sources say Apple is not taking an OpenAI board observer role as AI antitrust scrutiny mounts at the same time. Microsoft is flat up giving up an observer seat on OpenAI's board, citing what it calls the new board's significant progress and saying the company is confident in Open AI's direction. Now, OpenAI, to host regular meetings with investors and partners. So, just like the acquisition and all but

name strategy we've been speaking about recently, is this a board of directors and all but name? Quoting the FT. Microsoft, which has invested $13 billion in the maker of the generative AI chatbot chat chbt, said in a letter to OpenAI that it withdraws from its board role effective immediately. Apple has also been expected to take an observer role on OpenAI's board as part of a deal to integrate chat GPT into the iPhone makers' devices, but now will not do so, according to a person with direct knowledge of the matter. Apple declined to comment. OpenAI would instead host regular meetings with partners such as Microsoft and Apple and investors Thrive Capital and CoSla Ventures, part of a, quote, new approach to informing and engaging key strategic

partners under Sarah Fryer, the former Next Door boss who was hired as its first chief financial officer last month, and OpenAI spokesperson said. The move also comes as antitrust authorities in the EU and US examine the partnership between Microsoft and OpenAI as part of a broader concern about competition in the rapidly growing sector. Microsoft accepted a non-voting role on the board following the chaos that engulfed OpenAI last year when its chief executive Sam Haltman was abruptly fired by the board before being reinstated just days later. The attempted boardroom coup threatened Open AI's valuation and with it Microsoft's multi-billion dollar investment in the company, end quote.

And quoting Axios. We appreciate the support shown by OpenAI leadership and the OpenAI Board as we made this decision, Microsoft said. As you know, we accepted the non-voting board observer role at a time when Open AI was in the process of rebuilding its board, Microsoft said in the letter. This position provided insights into the board's activities without compromising its independence, and we appreciated the opportunity to serve as an observer during this period of change. Over the past eight months, we have witnessed significant progress from the newly formed board and are confident in the company's direction. Given all this, we no longer believe our limited role as an observer is necessary, Microsoft added. With Microsoft's

departure, Open AI no longer plans to have any observers on its board, according to a source familiar with the company's thinking. In giving up its observer seat, Microsoft removes a potential sticking point for antitrust regulators who could object to the giant's significant entanglement with Open AI, end quote. We've spoken before about how the AI land grab has led to an explosive growth in cloud computing to the degree that it is endangering various companies' efforts to become carbon neutral. I was just speaking about this to someone yesterday. How does this play out? Do companies just quietly miss their targets and hope no one notices? Do they just say, sorry, business needs, trump everything else? Well, according to a new report,

Google is straight up no longer claiming to be carbon neutral and has stopped buying carbon offsets. But they say they still aim to reach net zero carbon by 2030. And actually, there might be some nuance to this. Quoting Bloomberg, the alphabet unit has claimed that it's been carbon neutral in its operation since 2007. The status was based on purchasing carbon offsets to match the volume of emissions that were generated from its buildings, data centers, and business travel. But in its latest report, the company states, starting in 2023, we are no longer maintaining operational carbon neutrality. It's a sign of a company learning to adapt to strengthening sustainability criteria. Quote, in line with a changing market, including a more robust carbon removal ecosystem, we helped catalyze. We've shifted our strategy.

A Google spokesperson said, we aim to avoid or reduce greenhouse gas emissions to reach our absolute emissions reduction target, end quote. The changes to its carbon credits purchase strategy have coincided with Google and big tech's push on artificial intelligence, which a Bloomberg news investigation has shown is extremely resource intensive. As a result, Google's total planet warming emissions in 2023 are 48% higher than 2019. In that period, its total energy consumption has doubled. It's a similar story for Microsoft, which has seen primarily its AI activities cause emissions to rise by 30% relative to 2020, even as it still aims to be carbon negative by 2030. The definition of carbon neutral is when an entity balances its planet-warming emissions with techniques that draw down the same amount of carbon dioxide from the air. However, most companies making carbon neutral claims today rely on cheap offsets, where one credit equals a ton of emissions that are derived from projects that claim to protect forest or deploy clean energy.

and avoid the production of emissions. While those are actions that should help the world tackle climate change, experts have found that the purchase of the emissions avoidance offsets does little to actually reduce emissions. Many forest projects exaggerate their claims, while most renewable energy plants would have been built without the small sums raised by the offset purchased. That's why most carbon neutral claims from companies and products are questionable. Google was among a few global corporations that was early in thinking about the role that companies should play in fighting climate change after offsets-based claims from companies came under scrutiny. Google stated in a Bloomberg Green story in 2021 that it aimed to reach carbon neutrality without offsets. But now, instead of

bulk purchase of emissions avoidance offsets, Google says it's going to focus on absolute reductions in emissions and buy carbon removal credits for its residual emissions. These tend to be more expensive, but also have greater chance of verifiably drawing down carbon dioxide from the air, end quote. Meanwhile, and this is what I was getting at by stories clustering together for whatever reason, to combat its own AI energy use, Microsoft announced it has agreed to buy a record 500,000 carbon credits from U.S. oil producer Occidental's 1.5 startup to offset its emissions. So they're sticking with the strategy that Google says is no longer relevant, though, is that, as I suggested, just moving the goalposts and hoping people don't notice,

I don't know. I'm not an expert in this particular area. Meanwhile, meanwhile, Amazon claims it has actually reached its goal of 100% clean energy in 2023, seven years ahead of schedule after investing in more than 500 solar and wind projects. But again, experts are skeptical, quoting the New York Times. The energy generated by those projects is equivalent to the electricity consumed by the company's data centers, corporate buildings, grocery stores, and fulfillment centers in 27 countries. But because the solar and wind farms do not all directly power Amazon's operations, most of that energy is sent to electricity grids that serve many businesses and homes. Some critics say that the company's calculations can create a misleading impression of its efforts on the climate. The clean energy projects Amazon has invested in can produce enough energy to power the equivalent of 7.6 million U.S. homes, the company said.

Amazon aims to reach net zero carbon emissions from all of its operations, including its delivery vans, planes, and other means of transportation by 20. 2040, end quote. We've been talking about price rises in video streaming, but what this segment presupposes is, why not gaming too? Microsoft has hiked the price of Xbox GamePass Ultimate by $3 to $1999 a month for new subscribers and plans a standard tier for $1499 a month without day one titles, quoting the verge. The price increases come just before Microsoft plans to add, Call of Duty, Blackop's 6, and

other big first-party Xbox games to its GamePass service later this year. The price increases will largely affect Ultimate Subscribers who make up the vast majority of GamePass subscribers and come just over a year after the last GamePass price hikes raised rates by $1 to $2 per month. Alongside the Ultimate and PC GamePass price hikes, Microsoft will also offer a new option of an Xbox GamePass standard subscription without day one titles that will be priced at $14.99 per month for new users to the service. The existing Xbox GamePass for console subscription will continue for existing subscribers,

but new GamePass subscribers won't be able to select the console option starting July 10th until the new GamePass standard subscription is available. Microsoft says Xbox GamePass standard, quote, will be available in the coming months. The new standard option also includes online console multiplayer access, which is something that Xbox GamePass for console lacked previously. Existing Xbox GamePass for console subscribers will only be able to stack up to 13 months for September 18th, but this won't impact existing stacks, end quote. Monster interesting raise, skilled AI, which is building a foundational model for robotics, has raised a $300 million Series A at a $1.5 billion valuation. The round was led by Lightspeed, SoftBank, Kootoo, and Jeff Bezos.

So again, folks are betting that generative AI will be the breakthrough that makes robotics a thing beyond just manufacturing and the like, but also SoftBank is, stirring, quoting Forbes. Robots, whether they are bipedal humanoids, handling basic factory tasks, or four-legged military robot dogs intended for urban combat, need brains. Historically, these have been highly specialized and purpose-built. But a Pittsburgh-based robotic startup claims it's created a single off-the-shelf intelligence that can be plugged into different robots to enable basic functions. Founded in May 23 by Abanov Gupta and Deepak Pathak, two former Carnegie Mellon university professor, skilled AI has created a foundational model for what it describes as a, quote,

general-purpose brain that can be slotted into a variety of robots and enabling them to do things like climbing steep slopes, walking over objects obstructing its path, and identifying and picking up items. Ravaraj Jane, the lightspeed partner who also led the company's seed round in July 2023, told Forbes, he was wildly impressed with skilled AI's models when he first saw them being pressure tested last April. Robots using them were able to perform tasks in environments that they'd never seen before and hadn't been designed for demos. The robots at that time were able to climb stairs, and I think it's really crazy how well they were able to do it because it's a very complex stability problem, he said. More impressive still, the robots using skilled AI's models also

demonstrated emergent capabilities, entirely new abilities they weren't taught. These are often simple, like recovering an object that slips out of hand or rotating an object, but they demonstrate the model's ability to perform unanticipated tasks, a tendency that occurs in advanced artificial systems like large language models. Skilled has pulled this off by training its model on a massive database of text, images, and video. One it claims is 1,000 times larger than those used by its rivals. To create this massive database, the co-founders, both former AI researchers at Meta, blend a mix of data collection techniques which have been developed and tested over years of research. One way was to hire human contractors to operate robots remotely and collect data about those actions.

Another was to have the robot carry out random tasks, record the results, and learn by trial an error. The AI model was also trained on millions of public videos. As a PhD student at UC Berkeley, Pathec developed a way of instilling artificial curiosity into robots by rewarding the system for producing outcomes that come about when it can't predict the result of its actions. The more uncertain the agent is about the prediction of the effect of its actions, the more curious it gets to explore, he explained. The technique incentivized the AI to navigate more scenarios and collect more data. His research on curiosity-driven learning was published in 2017 and has been cited more than 4,000 times, he said. Pathak also devised a way for robots to use written information from large language models

like GPT of how to open a can of milk, for example, and convert that into actions. In 2022, we figured out a way to put these things together in a single coherent system, Pathek said, the notion of learning from videos, learning from curiosity, learning from real data, but combined with the knowledge from simulation, end quote. skilled AI faces steep competition from a string of robotics companies that have emerged with billions of dollars in venture funding thanks to the AI boom. Industry Behemoth OpenAI recently revived its robotics team to supply models to robotics companies, Forbes first reported. Then there's outfits like humanoid robotics company Figure AI, helmed by billionaire CEO Brett Adcock and Covariant, an open AI spinoff that is building chat GPT for robots and is raised over $200 million to do so, end quote.

Finally, today, investors have a problem when it comes to AI. As we've discussed, if you want to train your own AI model, it can cost as much as $100 million to do so, and that's right now. In the future, it'll be more. And that's also just training the model, which you have to do before you can make a product using that model. So there's a huge upfront cost of compute

just as an entry ticket to the AI sweepstakes. $100 million is a huge amount to need before you have a product, much less product market fit. That's why you've seen these mega-eastern, deals where the major cloud computing companies take stakes in AI startups that are really just, hey, come use our compute, and that counts as us investing in you. The problem is, if you're an AI startup and you can only get your hands on a million or two dollars for your seed round, say, maybe $10 or $20 million tops, what do you do? You can't afford the compute to even get started.

It's the inverse of the trend of the last decade or so where the cost of forming a startup had gone so far down that it was almost trivial to start a billion-dollar idea with very little money invested beyond a few AWS instances, a few hires, and maybe some social marketing spend. Well, apparently, A16Z has figured out a way around this new era of expensive startups. They have apparently secured thousands of AI chips themselves, including Nvidia H-100 GPUs, and are renting those chips to portfolio companies, quoting the information. A16Z plans to expand the clusters to include more than 20,000 GPUs, a person said. That's about the same number.

Elon Musk says his startup XAI uses to train its GROC large language model, for instance. Andresen Horowitz has told startup founders the initiative is called Oxygen. The program highlights Andreessen Horowitz's aggressive moves into generative AI in the last two years. It has likely cost the firm hundreds of millions of dollars based on the current costs of specialized AI chips used to train and run AI models. Andresen Horowitz started providing access to its GPUs earlier this year to a small number of startups in exchange for equity. The startups also pay to rent the chips at a discount to market price, the person said. Small startups have found it difficult to get enough of the premium chips, even as the shortage of Nvidia chips has eased somewhat.

They have smaller budgets and are competing with larger companies that need far bigger quantities and can more easily sign contracts with the cloud vendors that rent the chips. Luma AI was among the first companies to train its models using the venture capital, firms GPUs. In January, Andriesen Horowitz led a $43 million series B investment in the Palo Alto California-based startup, valuing the company at $250 million. Amit Jane, co-founder and CEO of Luma said other venture firms offered Luma money at a higher valuation, but Andreason Horowitz's promise to give Luma access to computing resources helped seal the deal. For infrastructure and foundation model companies, compute is nearly everything, Jane said. The startup is developing multi-modal models, AI that can understand and generate both texts and images. In June, Luma launched

its first product and AI video generator called Dream Machine. And Dresen Horowitz joins a small number of other VC firms that have acquired access to GPUs on behalf of startups. Investing Duo Nat Friedman and Daniel Gross last year acquired 2,512 Nvidia H-100 server chips, estimated to be worth about $100 million, though the pair ran into the supply crunch and did not immediately get all the GPUs up and running. Index Ventures last year struck a deal with Oracle to rent GPU servers for some of its startups, end quote. But Brian, isn't there a Samsung unpacked event going on right now or went on this morning as you're listening to this? There is, there was, but for various reasons related to still being on the road, I can't cover it today, so I'll summarize all of the

announces from that tomorrow. Talk to you then.