Latest / Super Intelligence News and Report / Canada Bets on Cohere, Mistral Opens a Trillion-Parameter Model, Wikipedia Draws a Line
Transcript
- Super Intelligence News and Report — Canada Bets on Cohere, Mistral Opens a Trillion-Parameter Model, Wikipedia Draws a Line Host 1: Welcome to Super Intelligence News and Report for Friday, October 30th. Host 2: Yeah, excited to jump in. Host 1: Today we're doing a deep dive into the latest source material covering the AI industry and for you listening we know you follow AI and business closely so we aren't going to bog things down by defining terms today. Host 2: Right. We're just getting straight to it. We're looking at how the AI race is basically evolving from a pure tech sprint into this, you know, complex web of geopolitics and sovereign control. Host 1: And massive financial engineering. Host 2: Yeah. Host 1: I mean, think of it like a national power grid at this point. No country wants a foreign entity holding the off switch. Host 2: Exactly. Which brings us to our first headline today about Cohere. Host 1: Right. The Canadian deal. Host 2: Yeah. So they're targeting 7 billion Canadian dollars at a 35.7 billion dollar valuation. But the standout detail here is the the government of Canada stepping in with a a commercial equity investment of up to 1.6 billion. Host 1: Which, I mean that is definitely not your standard venture capital play looking for a massive massive exit. Host 2: No not at all. It is pure industrial policy. Canada is basically buying an equity stake to ensure data sovereignty. Host 1: So they can legally dictate dictate where the data lives. Host 2: Right. Right and who accesses it. That 1.6 billion is mechanically tied to keeping the compute within Canadian borders and ensuring the tech talent doesn't just drain down to Silicon Valley. Host 3: That makes sense. But you know not every country has 1.6 billion dollars to just just buy their own sovereign AI. Host 3: AI. Host 2: Right. European enterprises for instance essentially have to buy sovereignty off the shelf. Host 1: Which is exactly the gap Mistral is trying to fill with their new large four model they just opened the waits for this one trillion parameter model right Host 2: Yep they did. Host 1: And while the performance is really strong Mistral's real pitch here is actually jurisdiction heavily regulated European industries so finance healthcare defense they cannot legally rely on on an American or Chinese API Host 2: Right Host 1: so Mistral is giving them a massive model that they can run can run entirely inside their own corporate firewalls Host 2: Okay, but I do have to push back a little on Mistral's motives here. Host 1: Oh, how so? Host 2: Well, are they championing this whole auditability thing simply simply because they kind of have to? I mean, they've raised around $6 billion, which is a lot, but Host 1: but they physically cannot match the hundreds of billions the big hyperscalers are burning Host 2: Exactly. exactly Host 1: are they leaning into this local sovereignty angle just because it's their only viable lane to survive Host 2: i mean it might be born out of financial reality sure but you shouldn't discount the engineering it takes to actually pull that off Host 1: fair point Host 2: yeah Host 1: compressing it is hard Host 2: yeah getting a one trillion parameter model to run efficiently inside a local corporate firewall is a a massive technical feat it requires completely different optimization than you know just running it on a limitless cloud cloud cluster Host 1: so it's not just a just a marketing bit of it Host 2: right and while we're seeing this fierce battle to control where models physically run there's an equally fierce battle over the cognitive resources extracted to train them Host 1: ah which brings us to wikimedia's amicus brief amicus brief in that massive copyright litigation against OpenAI Host 2: Yeah, they are drawing a very specific mechanical line here. Wikimedia is stating that using their 67 million Wikipedia articles to train a base model should be considered fair use. Host 1: Because they want open source foundation tools to thrive. Host 2: Yes. But they draw a hard line at unlicensed chatbot regurgitation. Host 1: Right. So basically saying it's fine for an AI to read the encyclopedia to learn how to speak. Host 2: Exactly. Host 1: But it's not fair use to rip the pages out and spit them back as as direct answers without publisher credit. Host 2: Because that destroys the human knowledge production pipeline. If the AI replaces the synthesizer and and strips away citations the secondary sources that generate the original data will just financially dry up entirely. Host 1: So true. Okay so sovereignty solves the software side data access feeds the model but it all eventually crashes into hard physics. Host 2: Oh absolutely. Securing the concrete the the copper the sheer gigawatts to run these systems. Host 1: The cost of AI compute is forcing some truly wild financial engineering right now. Host 2: It is. Host 1: Which explains the news and we need to stress that this this is reported not confirmed that NVIDIA is in talks to guarantee billions of dollars in debt for an AI cloud provider called Nscale. Host 1: scale. Host 2: Wait, N scale? Host 1: They're building that massive data center in West Virginia, right? Host 2: Yeah, the one meant for Anthropic. Host 1: Wow. So NVIDIA isn't just selling chips anymore. They are effectively acting as a central bank for AI. Host 2: Basically. Because traditional lenders just won't touch billions in data center debt. Host 1: Oh because the physical depreciation of GPUs is way too fast I mean a server server bot today is practically obsolete in two years Host 2: Exactly. So Nvidia has to step in to keep their own sales pipeline moving. If you synthesize all of this you have Canada throwing in 1.6 billion in sovereign equity right you have Microsoft pricing massive bond offerings and now Nvidia acting as a potential lender of last resort Host 1: The industry is desperately finding creative ways to fund the concrete and power this requires. Host 2: Exactly. The narrative has shifted. it's no longer just about who has the smartest algorithm it's about who can secure the power bill and which government or hardware monopoly is willing to backstop the backstop the debt. Host 1: Right so to sum up the AI AI industry is fragmenting into locally controlled tech stacks backed by by really unprecedented financial maneuvering and we want to leave you with a final thought to ponder on your own. Host 2: Yeah think about this. Host 1: If national governments and hardware monopolies are the only entities wealthy enough to backstop this staggering infrastructure will the future of AI truly be open or will it be fundamentally restricted by the physical borders where the servers sit? Host 2: Because if AI is the new national power grid who really controls your off switch? Host 1: Definitely something for you to mull over. We'll leave it there and hand things off to the main story. Host 1: Imagine watching half a trillion dollars evaporate in like a single afternoon. Host 2: Just completely gone. Host 1: Yeah. Wiped out. And that is the brutal market reality Alphabet woke up to yesterday Thursday October 29th 2026 Host 2: Right. Host 1: I mean a 12.6% drop down to $283.82 wiping out nearly 500 billion in market value it was their worst single single day performance since its 2004 listing okay let's unpack this Host 2: Let's do it Host 1: because the AI build out is accelerating but equity markets are aggressively I mean violently violently punishing the hyperscalers for the cash it burns Host 2: yeah and today you know we are examining the hyperscaler cash flow crack Alphabet Amazon and Microsoft are all showing immense cloud growth growth the demand is there Host 1: Totally Host 2: but the infrastructure cost is is triggering a full blown market revolt Alphabet had its first joint revenue and profit miss in nearly four years breaking a massive run of beats. Host 1: It's wild. And to understand the the sheer scale of the panic, you really have to look at the massive gap between the money coming in and the money going out. Host 2: Right. The fundamentals. Host 1: Exactly. So let's bounce through the key numbers. Alphabet brought in 127.6 billion dollars in revenue which is up 25% Host 2: but that was below the 128.2 billion expected Host 1: Right a miss. But then you have this massive win on the other side, Google cloud revenue hit 30.7 billion dollars Host 2: Wow which is up 103% right Host 1: Exactly. And operating income income for cloud tripled to 11.1 billion dollars Host 2: Okay Okay so the revenue is growing but here comes the capex shock Alphabet spent 51.4 billion dollars in the quarter Host 1: just in one quarter Host 2: yeah up from 24 billion a year ago Host 1: incredible Host 2: and that resulted in a negative free cash flow of 9.1 1 billion dollars which is its second consecutive negative quarter Host 1: just bleeding cash Host 2: yeah and Amazon showed an identical pattern from the same day Amazon swung to a trailing 12 month free cash flow outflow of 42.4 billion dollars Host 1: and they were positive before that right Host 2: yeah down from a positive 14.8 billion a year earlier. And that was after a staggering quarterly capex of 77.3 billion dollars. Host 1: Unbelievable. But there is a lifeline in all this spending. Amazon reported a 578 billion dollar AWS contracted backlog. Host 2: That is massive. Host 1: Right it's more than double a year earlier. So it sets up this crazy tension of future promises versus current costs. It's like well it's like financing the world's most expensive multi-lane toll road but right now you're only collecting pennies from bicycles. Host 2: Right exactly Host 1: while pouring billions into concrete for flying cars that arrive in 2027. Host 2: That is a perfect way to put it the it the bill for the concrete is due today this raises an important question though if the backlog is that massive but the cash drain is this severe who is reading the situation correctly here is it the tech optimists or the panicked panicked market Host 1: well I'm going to take the bull case here I think the optimists have it right Host 2: okay let's hear it Host 1: the demand will absolutely pay off look at what Sundar Pichai is saying he noted a five year high query growth Host 2: sure Host 1: and they have 1 billion Gemini app users users right now their APIs are handling nearly 32 billion tokens per minute you you don't just ignore that kind of volume Host 2: volume is great but Host 1: and wait Philip Schindler made a great argument that that AI mode queries are incremental they aren't cannibalizing old search they'll monetize later Host 2: later being the key word Host 1: true but it's backed up by Amazon too Andy Jassy is projecting AWS growth over 40% in 2027 and 2028 Fortune 500 bedrock spend is up nearly 7x 7x and token volume is up 17x Host 2: huge numbers I'll admit Host 1: yeah even the Jeffries analyst Brent Thill pointed out that cloud was the only real beat meaning the market is just overreacting to ad timing the infrastructure demand is is solid Host 2: okay I hear you but I have to take the bear case here Host 1: bring it on Host 2: monetization and financing are deteriorating much faster than the revenue improves you mentioned the AI search AI search queries but Alphabet itself admits AI mode searches earn less than classic search right now Host 1: well yeah for now Host 2: but that's a problem when their long term debt has ballooned six fold in a single year to 125.1 billion dollars Host 1: six fold wow Host 2: yeah they are funding this with massive debt and to counter the amazon Amazon optimism Amazon missed on on earnings, $1.89 versus $1.96 expected. Host 1: Okay, that hurts. Host 2: They missed on ads and they guided guided Q4 light. The market verdict? Alphabet down 12.6%. Amazon Amazon down more than 6% after hours. It proves investors now actively discount cloud beats when free cash flow is deeply negative. Host 1: Because they're just burning too much cash to get there. Host 2: Exactly. In 2027 spending is guided blindly higher. No ceiling. Host 1: Okay, weighing it all out, the infrastructure demand is irrefutably real. We know that. But the debt funded low margin nature of the initial build out is completely breaking Wall Street's tolerance. Host 2: Totally breaking it. Host 1: So if the equity markets won't fund this without a fight, how are these companies actually going to pay for the future of AI? Host 2: If we connect this to the bigger picture, this isn't just just a stock market correction. It exposes the financing fragility fragility of the entire AI ecosystem you operate in. Host 1: Because everyone relies on this infrastructure. Host 2: Exactly. And the creative financing we're seeing is wild. Host 1: Like Microsoft. They just priced a $25 billion bond offering to maintain flexibility. It's their first major US investment grade sale since 2017. Host 2: Right. Host 1: And they still have positive free cash free cash flow. Host 1: They're just hoarding cash because of macro pressures. Host 2: And you have to look at Microsoft's OpenAI revenue loop. Their commercial arrangements produce $10 billion in Q3 revenue. Host 1: A massive chunk of their total revenue. Host 2: Yes, but OpenAI still owes them 7.2 billion dollars in receivables. Host 1: So it's basically circular money at this point Host 2: Pretty much. much. And speaking of ecosystem financing, Nvidia is stepping in too. Host 1: oh right they are in talks to give credit support for Enscale's deal with Anthropic they need to be clear though this is reported not confirmed by the information Host 2: still it shows how far these companies will go to grease the wheels but there's broader regulatory pressure pressure building on these companies too Host 1: yeah Washington is watching Host 2: objectively speaking Senator Warren is is pressing Anthropic for answers about IPO impacts amidst these existential financial and societal questions Host 1: the pressure is coming from everywhere Wall Street and Washington so what don't we know yet let's explicitly list the unknowns first what will the actual 2027 capex be for Alphabet Amazon and Microsoft Host 2: right Host 1: and how much of that will be debt versus leases Host 2: that's a huge blind spot also will AI search search monetization actually close the gap with classic search Host 1: we don't know know yet and how much of this cloud growth is profitable enterprise inference versus lower margin capacity resale Host 2: or hardware pass-throughs Host 1: right like Alphabet's TPU system sales Host 2: exactly Host 1: and finally will the third party GPU and power delays that forced Microsoft to trim its 2026 capex target to 170 billion ease or worsen by early 2027 Host 2: These are the massive unknowns dictating the market right now. So what to watch? November 4th, 2026, Microsoft's 25 billion dollar bond closing. Host 1: Yes. Keep an eye on that. Host 2: Watch the final rates and demand across the 10 tranches as a real time read on the debt market's appetite for AI spending. Host 1: And second, TBA. Alphabet's fourth quarter call. Host 2: Why that one specifically? Host 1: CFO CFO Anant Ashkenazi has promised to finally detail exactly how significant the 2027 capex increase will be. Host 2: Everyone will be holding their breath for that number. I'll leave you with a final provocative thought. Host 1: Let's hear it. Host 2: If public equity markets simply refuse to tolerate the negative cash flow required to build AGI, will we eventually see big tech forced to spin out their AI infrastructure arms into massive separate debt carrying utilities? Host 1: Wow. Like the telecom build build outs. We'll see you back here tomorrow at the same time and please check the show notes for all sources.