Instacart reportedly offered itself to Uber. Uber reportedly said no. As an Uber holder, that news didn't surprise me. On Thursday, it came out that Instacart, worth about $12B, is exploring a sale and had early talks with Uber and DoorDash. The stock jumped 9% to $56.44. Uber and DoorDash both fell about 3.5%. Then after the close it was reported that Uber had already walked away, and an Uber official said there are no talks right now. That leaves DoorDash as the only named buyer. Here's why I think passing was the right call. Uber says openly that it measures every acquisition against their own hurdle rate, including buying back its own stock. So a deal only happens if the price is good enough to beat that. If Uber said no, my guess is the price was too high for the upside it saw. At the right price, I think Uber would still have done it. Uber also already has a lot on its plate. It's buying Delivery Hero for $14.8B and ezCater for $2.3B, both in cash. A typical 30-40% premium would put Instacart at roughly $15.6-16.7B on top of that. Together, all three would come to about $33B, more than three years of Uber's free cash flow. Buybacks already dropped to $510M last quarter from $1.4B a year earlier. And Uber is already beating Instacart in grocery without buying it. Uber's grocery business is growing about 40% a year, Instacart's about 13%. Why pay a premium for the rival you're taking share from? The bull case for a deal is real though. Grocery is the biggest delivery category left. Instacart brings 2,200 retailers, an ads business doing around $300M a quarter, and about $1.3B of EBITDA expected this year. It would have closed the grocery gap in one move. The risk now is DoorDash. If it buys Instacart, it leads US restaurant delivery and grocery delivery at the same time. That's a much harder antitrust case, but if it clears, Uber Eats is a clear No. 2 in the US. Uber is priced at about 13x free cash flow, whereas DoorDash commands closer to 37x. To me, that gap reflects an opportunity. Uber has the scale and breadth to steadily siphon share from DoorDash over time. It is global, it is not a one-trick delivery business, and it can lean into platform network effects. That could mean outspending on customer acquisition or rolling out differentiated perks and functionality that pure-play delivery apps cannot easily match. UBER is one of the main stocks in my Milk Road portfolio. You can see it here: link.milkroad.com/Martin

Morgan Stanley's worst case for Galaxy is $16. The stock closed Friday at $14.68. TeraWulf is even further off. The bank's bear case is $17.50, and it closed at $11.32. I own both. Morgan Stanley reiterated Overweight last week, with base targets of $65 on TeraWulf and $43 on Galaxy, and called the selloff a buying opportunity. Four of the seven names in its table (TeraWulf, Galaxy, Hut 8 and Riot) now trade below the bank's own bear case. So either the bear case is too generous, or the market is pricing something worse than a bad year. I think it's pricing the second one: money for building drying up. Elon said it bluntly on Sunday. Debt markets have "almost entirely shut off access to money for any company that doesn't have substantial revenue independent of AI data centers." These landlords build mostly with borrowed money, so that sentence covers them too. He's right about the debt markets. But I think the risk is smaller here than for the neoclouds. These companies raise money to buy land, put up buildings, and connect them to power. They keep owning the land and buildings, and buildings depreciate over 25 years or more. Whatever happens to the AI trade, those assets don't go away. The rent they build against usually comes from hyperscalers or neoclouds, so the tenant risk is real. But I find it hard to imagine we'll use less of those services in the future than we do today. Whoever owns the land, buildings, and grid connection owns a valuable asset. Still, this capital-intensive phase is touching everyone, and anyone looking for funding will likely have to pay a steeper price for it, assuming they can secure any at all. But here is the other side of this trade: Morgan Stanley's TeraWulf target is about double the Street average of around $33. It also leads TeraWulf's credit line, which just got upsized to about $400M. They might be biased. And Galaxy's only Helios tenant is CoreWeave, whose funding gap banks put anywhere from $23B to $76B. The next tenant will definitely help. For all the noise, the market is pricing these PSPs at levels, and with implied upside, that are hard to look past. This is not a free lunch, but the asset base puts a floor under the downside, while the ceiling is essentially set by how much pricing power they can extract at the table. Galaxy reports Q3 tomorrow before the open. I want to see a profitable crypto business and any news on a second Helios tenant. PSPs are 13.6% of my portfolio on Milk Road. You can see the rest here: link.milkroad.com/Martin

SK hynix is reportedly weighing a special dividend of about 16T won (~$12B). If it happens, SK Square holders would get a bigger slice than direct SK Hynix shareholders. The payout would be about 23,000 won per share, roughly 1.8% of SK hynix's market value. Nice, not life-changing. SK Square owns 20.68% of SK hynix, so it would collect about 3.31T won (~$2.5B). But SK Square trades at a 46.6% discount to its holdings. So that same cash equals about 3.2% of SK Square's market value, close to double what a direct holder gets. For scale, SK Square's whole 2026 shareholder return plan is 310B won. This one payment would be about 10x that. The money would come from SK Hynix's old Kioxia stake, roughly 80% of an estimated 20T won gain. Nothing is decided yet. SK Hynix says it's reviewing options; nothing is final, and the board will decide on Wednesday with Q3 results. Back in August I wrote that special dividends could land on top of the record buyback, with the full plan coming at Q3. That's this week. The bear case is that the market won't pay for peak-cycle earnings, and a big one-off payout is what companies do at the top. Goldman just cut its KOSPI target from 12,000 to 9,000, though it stayed overweight, and says memory stocks are priced for a ~65% drop in 2027 earnings. Goldman calls that too pessimistic. I agree. In a flat-stock scenario, SK hynix trades under 2x forward earnings by the end of 2027. The catch for SK Square is that the cash lands at the holding company first. Shareholders only see it if SK Square passes it on or buys back stock, and so far its return plan says nothing about special dividends. I think this is about to change. Wednesday's SK Hynix board decision, alongside its buyback program, will be a key thing to watch. The second is whether SK Square says what it does with the money. I’m still firmly bullish on memory. It’s the largest slice of my Milk Road portfolio. See the full breakdown here: link.milkroad.com/Martin

Companies keep using more AI every month, but for two months now, they haven't paid more for it. Spending on AI tokens (what businesses pay the labs to run their models) grew just 1% in August and 0.2% in September, the slowest since last November. And it doesn't look like a summer lull. The price per token is down about 41% from the March peak, so companies run a lot more through the models while the bill barely moves. No surprise that once AI got cheaper, people started relying on it far more; that is the predictable twist known as the Jevons paradox. What bugs me is who's paying. Apollo found that 80% of the top 1% of AI spenders are VC-backed. So the biggest token buyers are mostly paying with investors' money, not profits. The bull side says this is fine. A cheap token burns roughly the same compute as an expensive one, so more usage on a flat bill is great news for chips and memory. I mostly agree. Volume is what memory sells into.. But volume only keeps growing if someone keeps funding it. If VC money slows, flat spend turns into falling spend, and the first to feel it are the neoclouds renting GPUs to labs and startups. The questions I keep coming back to is who is going to spend hundreds of billions on AI? Hyperscalers use most of their compute today for their own products, and EPS growth has been strong every quarter. But the rest of the market hasn't caught up yet. One caveat: the data below comes from a small sample of companies, so I'd treat the exact numbers loosely and the direction seriously. Microsoft and Google report Wednesday; Amazon on Thursday. I want to hear whether cloud growth is holding and who is actually renting all that compute. It is less about the backlog number and more about the quality of that customer, and how that contract is financed. Follow me @m0xt_ for more market updates and insights.