Tech Brew Ride Home - Tue. 07/22 – Is Masa Son Being Left Behind In AI Again?

Episode Date: July 22, 2025

Well, just like that, AI’s winning the Math Olympiad is commonplace. Is Stargate struggling to get off the ground, or is Sam Altman just going to do Stargate on his own without Masa Son? The AI comp...any who’s stated mission was to do AI ethically explains why it needs to get its hands dirty. And even if AI isn’t making scientific breakthroughs yet, there are early signs it is shaking things up nonetheless. Links: OpenAI and Google outdo the mathletes, but not each other (TechCrunch) SoftBank and OpenAI’s $500 Billion AI Project Struggles to Get Off Ground (WSJ) Oracle to Supply OpenAI With 2 Million AI Chips for Data Centers (Bloomberg) Leaked Memo: Anthropic CEO Says the Company Will Pursue Gulf State Investments After All (Wired) AI Comes Up with Bizarre Physics Experiments. But They Work. (Quanta Magazine) Learn more about your ad choices. Visit megaphone.fm/adchoices

Transcript
Discussion (0)
Starting point is 00:00:00 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 TechMean right home for Tuesday, July 22nd, 2025. I'm Brian McCullough today. Well, just like that, AI's winning the Math Olympiad is commonplace. Is Stargate struggling to get off the ground, or is Sam Altman just going to do Stargate on his own without Masasana? The AI company whose stated mission was to do AI ethically explains why it needs to get its hands dirty.
Starting point is 00:00:53 And even if AI isn't making scientific breakthroughs yet, there are signs that it is starting to shake things up nonetheless. Here's what you miss today in the world of tech. Maybe this was an advance whose time had just come. Google says an advanced version of Gemini with DeepThink also won gold at the International Mathematical Olympiad, solving five of six exceptionally difficult problems, quoting TechCrunch. The results underscore just how fast AI systems are advancing and yet how evenly matched Google and OpenAI seem to be in the AI race. AI companies are competing fiercely for the public perception of being ahead in the AI race. an intangible battle of vibes that can have big implications for securing top AI talent. A lot of AI researchers come from backgrounds in competitive math, so benchmarks like IMO mean more than others. Last year, Google scored a silver medal at IMO using a formal system, meaning it required humans to translate problems into a machine-readable format. This year, both OpenAI and Google entered informal systems into the competition, which were able to ingest questions and generate proof-based
Starting point is 00:02:06 answers in natural language. Both companies claim their AI models correctly answered five out of six questions on IMO's tests, scoring higher than most high school students and Google's AI model from last year without requiring any human machine translation. In interviews with TechCrunch, researchers behind OpenAI and Google's IMO efforts claimed that these gold medal performances represent breakthroughs around AI reasoning models in non-verifiable domains. While AI reasoning models tend to do well on questions with straightforward answers such as simple math or coding tasks. These systems struggle on tasks with more ambiguous solutions, such as buying a great chair or helping with complex research. However, Google is raising questions around how OpenAI conducted
Starting point is 00:02:50 and announced its gold medal IMO performance. After all, if you're going to enter AI models into a math contest for high schoolers, you might as well argue like teenagers. Shortly after OpenAI announced its feet on Saturday morning, Google DeepMind's CEO and researchers took to social media to slam OpenAI for announcing its gold medal prematurely, shortly after IMO announced which high schoolers had won the competition on Friday night and for not having their models test officially evaluated by IMO. Thang Long, a Google DeepMind senior researcher and lead for the IMO project, told TechCrunch that Google waited to announce its IMO results to respect the students participating
Starting point is 00:03:29 in the competition, end quote. Speaking of suggestions that OpenAI might be bending over backwards to manage its PR, overnight the Wall Street Journal was reporting that the Stargate Initiative has struggled to get off the ground as the various companies involved have sharply scaled back their near-term plans, and SoftBank and OpenAI seemingly are disagreeing on crucial terms of the deal. Quote, six months after Japanese billionaire Masayoshi Son, stood shoulder to shoulder with Sam Altman and President Trump to announce the Stargate Project. The newly formed company charged with making it happen has yet to complete a single deal for a data
Starting point is 00:04:16 center. While the companies pledged at the January announcement to invest $100 billion immediately, the project is now setting the more modest goal of building a small data center by the end of this year, likely in Ohio, the people said. Stargate's lethargic launch is a setback to the vast ambitions of Saan, who despite spending billions of dollars over the years, has been playing catch-up in the fast-evolving AI sector. SoftBank committed $30 billion to open AI earlier this year. It is by far the largest ever startup investment, an enormous wager that has led SoftBank to take on new debt and sell assets.
Starting point is 00:04:48 The investment was made alongside the plans for Stargate, giving SoftBank a role in the physical infrastructure needed for AI. Altman, eager to secure the computing power to support the next generation of his company's signature product, ChatGBT, has plowed ahead without SoftBank, signing deals for data centers with other operators. The leaders of both companies say all is well in their joint effort. Last week, they appeared on video at a soft bank event. And Altman said they have an initial goal of building 10 gigawatts of data centers together.
Starting point is 00:05:18 It is a, quote, wonderful partnership, he said. In a joint statement, the two companies said they were advancing projects in multiple states and were moving at hyperscale and speed to deliver the AI infrastructure that will power the future and serve humanity, end quote. Altman's OpenAI recently struck a data center deal with Oracle that calls for OpenAI to pay more than $30 billion a year to the software and cloud computing company, starting within three years, according to people familiar with the transaction. That deal, which doesn't involve SoftBank totals 4.5 gigawatts of capacity and would consume the equivalent power of more than two Hoover dams, enough to power about 4 million homes. The data centers are spread among locations around the U.S., people familiar with the deal said. taken with a smaller deal, OpenAI struck with CoreWeave, OpenAI now has completed data center deals for nearly as much capacity as Stargate promised for this year in January. OpenAI has said
Starting point is 00:06:13 $100 billion roughly equates to five gigawatts of data centers. Despite Stargate's slow start, Sahn has told Associates he is bullish on Open AI and would like to invest even more in the company a person familiar with the matter said. Sahn has long craved a prominent seat at the AI table over the past decades, he devoted the two largest startup funds ever raised, more than $140 billion, to finding the AI companies of the future, only to miss out on OpenAI and all of its competitors before the launch of ChatGPT, while being tarred by high-profile flops such as WeWork and Construction Startup Katera. Stargate is not formed yet, Oracle Chief Executive Safra Katz said on an investor call last month. One recent complication between OpenAI and SoftBank has been over how extensively to build
Starting point is 00:06:57 data centers on sites tied to SB Energy, a softbank-backed energy developer, according to people familiar with the matter. Altman has used the Stargate name, shared with a 1994 Kurt Russell film about aliens who teleport to ancient Egypt, on projects that aren't being financed by the partnership between OpenAI and SoftBank. The trademark to Stargate is held by SoftBank, according to public filings. For instance, OpenAI refers to a data center in Abilene, Texas, and another it agreed in March to use in Denton, Texas as part of Stargate, even though they are being done without SoftBank, some of the people familiar with the matter said, end quote.
Starting point is 00:07:34 Well, I said managing PR because, boom, this morning, OpenAI and Oracle announced plans to develop 4.5 gigawatts of additional U.S. Data Center capacity in an extended partnership, hitting 5 gigawatts in total and running 2 million AI chips. Quoting Bloomberg, Open AI has yet to name the data center sites it will co-developed with Oracle, but states including Texas, Michigan, Wisconsin, and Wyoming are under consideration, together with its facility being built in Abilene, Texas, the company said it will have more than 5 gigawatts total in capacity running on more than 2 million chips for AI work. Bloomberg earlier this month reported OpenAIs plans to rent the additional data center capacity from Oracle.
Starting point is 00:08:15 In January, OpenAI, Oracle, and SoftBank Group announced they would invest $500 billion in 10 gigawatts of AI infrastructure in the U.S. in the next four years. years. OpenAI said Tuesday that it now expects to exceed that commitment. A gigawatt is enough to provide electricity for about 750,000 U.S. homes. While OpenAI labeled the data center expansion with Oracle as part of its Stargate project, SoftBank isn't financing any of the new capacity, the AI company said. In May, Bloomberg reported that SoftBank was hitting snags and financing talks amid broader economic uncertainty around global tariffs. The chat GPT maker also shared an update on its initial Stargate site in Abilene, saying that the first data center building there
Starting point is 00:08:56 now powers some of OpenAI's compute workload for running and training its algorithms. Oracle began delivering the first racks of Nvidia GB200 chips last month, and part of the data center has been up and running for a few weeks, the company said. We feel pretty good about our ability to move quickly on this because in many regards, Abilene was that beta test to prove out that you could build these at scale and at speed, said Chris Lahane, opening I vice president of global policy, end quote. So is Masa getting left behind on AI again? As I said, all of the sudden, the only numbers that matter in tech are megawatts and gigawatts, not megabytes and gigabytes. Lots of chatter this morning around the leaked memo wherein
Starting point is 00:09:43 Dario Amo Dai reportedly told staff at Anthropic that the company plans to seek United Arab Emirates and Qatar funding, likely enriching, quote, dictators and saying, quote, a no bad person rule is impractical. Quoting Wired, weighing the pros and cons. Amadai acknowledged in his note that accepting money from Middle East leaders would likely enrich, quote, dictators. Quote, this is a real downside, and I'm not thrilled about it, he wrote. Unfortunately, I think no bad person should ever benefit from our success is a pretty difficult principle to run a business on. The message comes as AI companies race to secure the massive amounts of capital required to train and develop frontier AI models. In January, OpenAI announced a $500 billion data center project called Stargate with financial
Starting point is 00:10:28 backing from MGX, a state-owned Emirati investment firm. Four months later, the company announced it was planning to build a data center in Abu Dhabi as part of a push to help foreign governments, quote, build a sovereign AI capability in coordination with the U.S. Quote, as an American company at the frontier of AI development, we have always believed the supply chain of frontier AI model development should be on American soil in order to maintain America's lead, said Anthropic spokesperson Christopher Nulte in a statement, quote, as Dario has said before, we believe fundamentally in sharing the benefits of AI and serve the Middle East and regions around the world commercially in line with
Starting point is 00:11:02 our usage policy. In May, President Donald Trump toured the United Arab Emirates and Saudi Arabia as part of a four-day trip focused on economic investments. A cabal of tech leaders, including Elon Musk, Sam Altman, and Nvidia chief Jensen Huang, joined him for a meeting with the Crown Prince of Saudi Arabia, Anthropics' leadership was notably absent. In his memo, Amo Dye acknowledged that the decision to pursue investments from authoritarian regimes would lead to accusations of hypocrisy. In an essay titled Machines of Loving Grace, Amadai wrote, quote, democracies need to be able to set the terms by which powerful AI is brought into the world, both to avoid being overpowered by authoritarians and to prevent human rights abuses within authoritarian countries. In
Starting point is 00:11:47 In 2024, Anthropic decided not to accept money from Saudi Arabia, citing national security concerns per CNBC. The news came as FTX. The failed cryptocurrency exchange went into bankruptcy proceedings, and its nearly 8% stake in Anthropic went up for sale. Ultimately, a majority of those shares went to A-TIC third international investment, a UAE firm. At the time, the stake was worth about $500 million. Now it appears Anthropic is poised to accept Gulf State money, though the company hasn't said whether it has changed its stance. on Saudi Arabia in particular. There is a truly giant amount of capital in the Middle East. Easily $100 billion or more, Amadai wrote in the memo. If we want to stay on the frontier, we gain a very large benefit from having access to this capital. Without it, it is substantially
Starting point is 00:12:32 harder to stay on the frontier. By pursuing a, quote, narrowly scoped, purely financial investment from Gulf countries, the company hopes to avoid the risks associated with allowing outside investors to gain leverage over the company, the memo says. The basis of our opposition to large training clusters in the Middle East or to shipping H20s to China is that the supply chain of AI is dangerous to hand to authoritarian governments, since AI is likely to be the most powerful technology in the world. These governments can use it to gain military dominance or to gain leverage over democratic countries, Amadai wrote in the memo referring to Nvidia chips. Still, the CEO admitted investors could gain soft power through the promise of future funding.
Starting point is 00:13:12 The implicit promise of investing in future rounds can create a situation where they have some soft power making it a bit harder to resist these things in the future. In fact, I actually am worried that getting the largest possible amounts of investment might be difficult without agreeing to some of these other things, Amadai writes. But I think the right response to this is simply to see how much we can get without agreeing to these things, which I think are likely still many billions, and then hold firm if they ask, end quote. In a section titled Erosion of Standards, Amadai notes that the reason Anthropic vociferously pushed for not allowing big data centers in the Middle East was because, quote, without a central authority blocking them,
Starting point is 00:13:50 there's a race to the bottom where companies gain a lot of advantage by getting deeper and deeper in bed with the Middle East. Unfortunately, having failed to prevent that dynamic at the collective level, we're now stuck with it as an individual company, and the median position across the other companies appears to be outsourcing our largest five-gigawatt training runs to UAE. Saudi is fine, the CEO said, likely referring to the United States Agreement to build a new AI data center in the UAE, powered by 5 gigawatts of electricity. Quote, that puts us at a significant disadvantage, and we need to look for ways to make up some of that disadvantage while remaining less objectionable.
Starting point is 00:14:26 I really wish we weren't in this position, but we are, end quote. Finally, today, if you listen to last weekend's bonus episode, you know that folks are growing frustrated by the fact that for all of the advances AI has made, it hasn't made the advances its biggest boosters have always promised, at least not yet, promises like, you know, curing cancer and stuff. Well, Quantum Magazine says that while AI hasn't yet led to new physics discoveries, the tech is proving powerful in that field, at least by aiding in experiment design and spotting patterns in complex data. For example, LIGO, the laser interferometer gravitational wave observatory pushes measurements to extremes. It's twin detectors.
Starting point is 00:15:17 in Washington and Louisiana use four-kilometer-long arms arranged in an L-shaped with lasers bouncing back and forth when a gravitational wave passes one arm lengthens or shrinks by less than the width of a proton, a sensitivity like measuring the distance to Alpha Centauri to the width of a hair. Built over decades and upgraded for years, LIGO first detected a wave in 2015 from two colliding black holes. But recently, physicists made efforts to optimize LIGO and later explored how to extend its capabilities. The team turned to AI to do so, feeding it every conceivable optical component. At first, the AI's designs were chaotic and alien, but after refining the outputs, the team uncovered a breakthrough. An additional three-kilometer
Starting point is 00:16:00 light circulation ring that reduced quantum noise. This principle long theorized, but never tested, could improve LIGO's sensitivity by up to 15%. AI is now reshaping experimental physical physics beyond LIGO, revealing patterns in complex data, guiding new quantum experiments, and even producing better models for dark matter. Again, as the leap labs people outlined for us, it's about finding new things to test, not necessarily just spitting out novel answers to hypotheses, quoting the conclusion of this piece. Cranmer and you point out that while such models are good at discovering patterns, making sense of those patterns, and coming up with hypotheses or the physics to explain them, remains elusive for today's AI models.
Starting point is 00:16:41 But Kranmer thinks that the advent of large language models like chat GPT could change that. I think there's a huge potential for language models to be useful to help automate that construction of hypotheses, he said. It's kind of around the corner. Steinberg agrees that while AI has yet to invent new concepts, AI-aided discoveries of new physics could conceivably become reality. We really might be crossing that threshold, which is exciting, he said, end quote. Nothing more for you today. Talk to you tomorrow.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.