Intro
Imran Khan, CIO and founder of Proem Asset Management, works through Nvidia's quarter the morning after it printed, the circular financing web running between Nvidia, OpenAI, Microsoft and the neoclouds, what he heard from memory investors in South Korea, and the Salesforce quarter and Anthropic deal that landed the night before.
Guest: Imran Khan, CIO and founder of Proem Asset Management, Dallas; formerly an internet analyst, then a banker who took Alibaba public, then COO of Snap
Host: Dan Nathan
Published: 28 August 2026 on RiskReversal Pod · recorded 27 August 2026
Watch on YouTube | 1 hr 0 min
Key Takeaways
The market has already discounted Nvidia's circular revenue
Assume a quarter of earnings comes from companies Nvidia funds, strip it out, and Khan still gets under 20 times against a 25-to-30-times history
Nvidia guided next year far above where the street was
70% growth for next year against "the street guidance consensus of 45%", unusually early for a company that size
The ecosystem checks are about supply, not just demand
"he's locking his competitors out" — rivals can design a chip but cannot get memory and components at scale
Owning Nvidia is owning the AI ecosystem, not a chip vendor
"If Nvidia doesn't work, I just don't see how AI works"
The host is not a bear on the technology, he is a bear on the allocation of resources
Memory's story has gone from a 10 to an 8
Revenue growth, not margin expansion, drives the numbers now, and "buying a stock based on multiple expansion is a tough way of making money"
A buyback is not a thesis
In Seoul group meetings, "Every question was about share buyback" — which he reads as a signal to look elsewhere
Salesforce's numbers are unspectacular and getting better
Growth accelerated by a point, and he makes it roughly 20 times GAAP earnings, "a 20% discount to Microsoft"
Valuation is a read on consensus, not a reason to own something
"my biggest mistake was buying stock just because they're cheap"
There are elements of bubble, and Nvidia's multiple is not one of them
New issuance, private marks and acquisition prices all look frothy to him; a 15-times multiple does not
Risk has been spread wide enough that no single company is exposed
"It's not only Meta going to hold the bag. Everybody going to hold the bag"
Risk management deserves as much airtime as stock picking
What Was Actually New in Nvidia's Quarter
Khan calls it an exceptional quarter with an incredible revenue growth rate, and says "they beat the numbers by 4 billion plus at that scale"
The new information was not the beat, it was the forward guide. The company talked about 70% growth for next year and said demand was running higher than that, which he says is unusual disclosure for a company that size at this point in the year
"that is well above the street guidance consensus of 45%"
That guide speaks to the durability question that hangs over the whole AI trade — whether investors wake up one morning and the spending is gone
His analog is the work-from-home unwind: people went back to the office after 2021 and the stocks that had run on the internet craze started suffering
The second new thing was a floor under gross margins, which had been the live worry as memory costs rose
Flowing sell-side numbers through, he gets to roughly $15 of GAAP earnings for fiscal 2028, which is calendar 2027, on 70% revenue growth with falling gross margins but fast-growing gross profit and operating expenses under control
The third thing he took away was a more assertive tone on buying back stock and returning capital
Dan Nathan read out the press framing around all this — Wall Street Journal headlines from the previous two days including "Nvidia's $279 billion supply chain gamble" and "Nvidia has become the banker to the AI boom, putting it on dangerous ground" — and noted the stock was up six or seven percent in the pre-market but still inside the range it had held for six to nine months
The Math That Makes 15 Times Look Cheap
On his own arithmetic the stock was trading at 14 to 15 times pre-market against a 25-to-30-times historical multiple, which he reads as the market refusing to pay for revenue tied to Nvidia's own investments
He runs the discount explicitly: take the $15 number, assume a quarter of those earnings come from partners Nvidia is funding, and earnings fall to about $12 and change — still under 20 times
"I think it's already priced in"
"I think that's why the stock is trading at 15 times" — the market, in his reading, is assigning zero value to the circular piece rather than mispricing the company
On the historical multiple, $12 and change at 25 times is a $300 stock, which is the gap he thinks the market is refusing to close
Nathan pressed on why this was not the quarter the stock rerated, pointing out that even a move to 17 or 18 times would be a large move
Khan's answer is that it can, but three things had to be established first: demand is more durable and accelerating, the supply chain advantage holds, and management sounds more willing to return capital
Why Jensen Keeps Writing Checks Into the Ecosystem
Putting on his operator hat, Khan says the investments do two things: diversify the customer base and lock up supply. He notes non-hyperscaler revenue grew faster last quarter, which moves the company away from dependence on five hyperscalers
"he's locking his competitors out" — a rival can design a strong inference chip, but without memory and components there is no product at scale
"Jensen is TSMC's biggest customers", with 85% of their product locked up, in his description
Nvidia's structural oddity is that "his biggest customers are also his biggest competitors", and Khan says "he is sharing his product road map with his biggest customers" anyway
He calls the relationship management the superpower, and Jensen irreplaceable for it — most technology founders, in his account, take the view that they do everything best
On the ecosystem itself: "he built ecosystem with CUDA he's building ecosystem with suppliers he's building ecosystem with neoclouds"
The risk case is twofold: the return on all this investment takes longer than expected, or a macro shock hits — his example is interest rates going to 9%
Nathan's aside: "Where by the way some of these neoclouds are raising money at 9%"
He does not think the strategy is reckless, but he does not know where the limit is: "I understand why he's doing it I probably would have done the same thing if I were him" and "To what extent is he extending too much? Only time will tell"
He also thinks the deeper supplier relationships speed up innovation, and argues that without the circular spending the stock would carry a higher multiple even on 50% growth
Where the Return on AI Shows Up First
Advertising is his cleanest example, from running Snapchat's ad business — "99% of the ad people see is wasted", so lifting impression effectiveness by even 50 basis points drives a large amount of revenue
It also lets a platform carry fewer ads, which helps usage
Nathan added Meta as the proof point of the last couple of years
Khan says the country should be rooting for AI to work, because a productivity boom means higher growth, higher tax income and help with the deficit — "Assuming our politicians don't screw up"
Nathan flagged Khan's slip — he said "tax rate" rather than tax revenue — joked that nobody in the business wants a higher tax rate, and steered away from a macro conversation
Khan's position is a high degree of confidence that AI generates real utility and real return, with the timing and the macro as the two open risks
Buybacks, Concentration, and Who Is Funding the Build-Out
Khan's argument for returning capital is about position size, not sentiment: every investor has a diversification limit, and Nvidia's market value has grown into it
Nathan: "It's 14% of the S&P 500" — one stock out of five hundred
He is explicit that the two are not exclusive — building the moat and returning capital can both happen
Nathan's counter is that the Mag 7 are not returning much: "Google in the last quarter bought back $19 billion worth of stock. That's a rounding error"
The build-out is being funded from cash flow and balance sheets at the expense of buybacks, then with debt, and a lot of that debt is off balance sheet
Nathan's read on why the private credit boom is going crazy: the hyperscalers do not want the debt on their books
He points to the round table Jensen convened with KKR, Apollo, BlackRock and David Solomon, where the lenders signaled willingness to backstop chip purchases
Khan's response is that this is the same ecosystem-building instinct applied to financing
Compute as an Asset Class
Khan thinks GPUs are fungible, which is what makes the idea work — a machine serving one customer at Nebius today can serve a different one tomorrow
The residual life of the asset is longer than people assume because the industry is supply constrained, not demand constrained, and a better chip cannot be built at scale quickly
"what's the lifetime value of this asset classes" is the question he says nobody has answered yet
Lead times on new capacity are long, and he says suppliers are being careful not to overextend
Revenue-sharing changes the economics again — taking 10% or 15% of what these companies earn on the hardware raises the value of the GPU itself
Nathan: "Hey that's a high margin business"
He credits the healthy skepticism in the market, saying the conversation the two of them are having is being had in every boardroom
The Circular Map, Read Out Loud
Nathan walked the whole web on air, prompted by a Bloomberg chart he says keeps getting updated and is genuinely hard to follow: Nvidia is OpenAI's investor and its supplier; Microsoft is OpenAI's largest investor and OpenAI is Microsoft's largest Azure AI customer; Nvidia is CoreWeave's largest investor, CoreWeave gets essentially all its GPUs from Nvidia, Nvidia backstops those sales and has agreed to buy unused compute, and CoreWeave's largest customer is Microsoft
He sees two points of failure in the picture: Nvidia and OpenAI
The sharpest version of the problem, in his framing: OpenAI is building a chip to compete with what it buys from Nvidia, and Bloomberg had just reported OpenAI claiming its new chips outperform Nvidia's processors in tests
Nathan's own position, in answer to a founder friend who texted to ask if he was still a bear: "I'm not a bear on the technology. I'm a bear on the allocation of resources"
He calls the technology magic, says the dinner table conversation about how people use it was amazing, and expects agents inside businesses to consume far more compute than consumer use ever did
Khan's answer starts from scarcity: "the world is compute shortage like there's an incredible shortage in compute", which is why buyers take compute wherever they can find it — his example is Anthropic doing a deal with SpaceX
"everybody is trying to control their destiny" — OpenAI does not want to be wholly dependent on Nvidia, and Nvidia in turn is buying into open-weight models, which he calls super smart because it opens a different vertical of customers
The conclusion he wants investors to take: "when you are buying Nvidia you are not buying GPU maker, you are buying AI ecosystem", and "If Nvidia doesn't work, I just don't see how AI works"
Why He Doesn't Own TSMC
"we don't own TSMC", and the reason is not a view on the business — "I just don't think I have a differentiated viewpoint" on it
He calls it incredibly important to the ecosystem, a great business, and one many of his investor friends own, and says he pays very close attention to it as a data point
The fund is concentrated, so a position has to earn its place by being a view the market does not hold — on Nvidia, he says the differentiated view is that the circular flow is already in the price and that the company is an ecosystem rather than a chip maker
His framing for any purchase: are we thinking something differently that people are not appreciating today but will appreciate tomorrow
Micron, Memory, and Why the Cycle Caught Them
Nathan's setup: a deeply cyclical company that went from negative gross margin to 85% margin, and "we've never seen a company like this have that sort of pricing power" in such a business
He asked why Micron, watching Nvidia's growth up close and knowing better than almost anyone that the GPU has to be paired with high bandwidth memory, did not bring capacity on sooner — "they're just they're digging a hole right now in Clay New York" on a plant discussed three years ago, near his upstate hometown
The cost of that miss, in his telling, is the "1,000% run in the stock over the last year"
Khan's answer is that Jensen saw it and they did not: "He saw the AI revolution before anybody else", and can arguably be said to have helped create it
The memory makers were leaning on consumer demand and were in a defensive posture, and he points out how hard it is to invest through negative gross margins
He also spreads the blame: "you can blame Apple to create some of the problems too", because the memory suppliers were squeezed hard by their largest customer
When a Story Goes From a 10 to an 8
His central stock-picking idea, stated plainly: "a lot of great companies could be bad stock and a bad company not so great company could be phenomenal stock"
"When the story goes from 10 to 8, those stocks tend to underperform" — and a story going from one to three can produce a phenomenal stock
Memory is now a 10 becoming an 8, in his reading. A year ago the move came from explosive demand and expanding margins together; from here the numbers are driven by revenue rather than margin expansion
That does not make it a bad company or mean the stock cannot rise, but the buyer today is arguing for a higher multiple on better visibility
"buying a stock based on multiple expansion is a tough way of making money"
The easy way, which he calls Imran's rule: revenue up, margins up, market share gains
The 60-second test he has carried since his analyst days: "if I can't explain a story to an investors in 60 seconds or less" it does not clear the bar
"When you have to explain an LTA, you cannot explain it in 60 seconds"
"Complicated things. Again, it works" — he is not saying the complicated trade fails, only that there is an easy way and a difficult way to make money
The Tape While They Were Talking
Nathan used the Roundhill DRAM ETF, launched in April, as a sentiment marker — people called it the top, and it has since taken in tens of billions of dollars as a way to get access to Samsung and SK
Shortly after the open he read the board out: Nvidia "It's up 7%", Apple down, and SanDisk, Micron, Seagate and Western Digital all lower on the day
Khan's explanation for Apple's weakness ties back to the same squeeze — the company no longer gets the access to compute and components it once had, and it needs memory to build iPhones and iPads
He also suspects large-cap managers are reallocating within the Mag 7
Nathan noted the DRAM fund was lagging both the Mag 7 fund and the semiconductor index, with Nvidia driving much of the latter
What Seoul Was Asking About
Khan avoids one-on-one company meetings on purpose: "I rarely rarely request one-on-one meetings"
He prefers group meetings where a company presents to 25 investors, because the questions other people ask tell him more than the answers do — "management going to tell you what they're going to tell you", and nothing beyond what has been said publicly
In a meeting with 30 or 35 investors, roughly 70% of the questions were about share buybacks and 30% about long-term agreements
Nathan noted SK had just launched a buyback he put at around 25 billion
His read on that mix is a warning: "if you're buying the stock because they're going to buy back shares, there's other place to invest time and find better ideas"
What he actually hunts for is inflection: "I'm always looking for where business underlying business fundamentals are seeing inflection" — revenue improving, margins improving, share gains, which is what memory looked like nine months ago
His best investments over seven and a half years running the fund were businesses nobody was paying attention to where the story quietly got better
Asked about performance, he declined: "Let's not talk about performance. We don't talk about it"
Domain Expertise Is a Blind Spot
"being a domain expertise doesn't make you a good stock picker" — and the deeper the expertise, the bigger the blind spot, because you can talk yourself into a position
He points to Stanley Druckenmiller as the model, and says the great investors he has met are not domain experts in any single industry
What they understand instead is what has to happen for a company to become a better stock
The overthinking failure he names is arguing about whether a company buys back another 10% of its shares rather than asking what is actually changing in the business
His example of fundamentals turning: the hyperscalers will be phenomenal stocks when the capital expenditure cycle starts showing real return, the way Microsoft's stock worked when revenue growth accelerated, and the way Google's did
Salesforce, Anthropic, and the Toll Software Now Pays
Nathan's setup: a stock that was down more than 70% from its all-time high a few months ago, a CEO he calls one of the best salesmen technology has produced, and a quarter whose numbers were not eye-popping
The thing that moved investors was Marc Benioff sitting in a conference room with Anthropic's Dario Amodei, interviewed by Jim Cramer, announcing a deal together
Khan agrees the numbers are unremarkable and says that is not the point: "their growth rate accelerated by one point", and every metric is pointing the right way
Growth is now 12, 13, 14% against the single-digit business he refused to pay a multiple for
Seat count is improving, margins are improving, and "They were the biggest abuser of stock-based compensation" — that is improving too
He will not call it an inflection point, but he will say things are getting incrementally better, which is what he thinks puts a floor under the multiple
His structural point about software is the one investors should carry away: the industry paid a toll to cloud providers when it moved off premise, and "now we are have to pay a toll to Dario"
The question for every software company is whether it can charge customers more to cover that cost, cut elsewhere, or both
On the hundreds of millions of dollars in tokens Salesforce said it would spend with Anthropic, he says the money has to be made up somewhere
Nathan's observation: the company keeps cutting staff, and has had layoffs this year
On the arithmetic: guidance around $10 this year, roughly $12 next year on 20% growth, which at 240 is about 20 times GAAP earnings — "I don't look at non-GAAP. It's a GAAP EPS, which is trading at a 20% discount to Microsoft"
Nathan's price history: "the stock traded $360-some dollars" at the mid-2024 high, "It traded as low as 150" a few months ago, and "it's basically trading at 245" now, up 18% on the day
On whether leadership rotates back into software: he says "software was probably under-owned in the market" while the semiconductor names are widely held, and that improving stories tend to mean those stocks do well
He would not predict the move — the underlying fundamentals in semis are still very strong and valuations there have come down since the summer, so he expects some of those names to work too
What Valuation Is Actually For
"my philosophy is you don't buy a stock just because of valuation", and "my biggest mistake was buying stock just because they're cheap"
You buy because the story and the fundamentals are improving, and you sell because they are deteriorating. Valuation is the guideline that tells you what the market currently believes
A price-to-earnings or free-cash-flow multiple is a reflection of consensus
Nvidia at 15 times says the market is heavily discounting the earnings; Salesforce at 150 on $10 of earnings said the market thought the company had something close to a going-concern problem
"buying or selling stock just based on narrative is could be very dangerous because you miss out the incrementality"
He is careful to add that the reversal does not mean Salesforce is the old Salesforce, and the criticisms of it have not gone away
Let the numbers lead: "Let the numbers guide you as opposed to narrative guide" is what he says has worked over seven and a half years
The Dentist's Question
Nathan set up the question that gets asked outside the business — by someone watching data centers go up nearby, hearing about strain on energy supply, and seeing their retirement account rise: are we in a bubble?
He notes that professionals do not actually talk this way among themselves; they drill into themes and look for ways to express them
Khan's answer is that it is not an easy one, and that the valuations of the biggest and most durable businesses do not say bubble — but that elements of bubble are clearly present
The three places he sees it: new capital issuance, private market valuations, and prices being paid in acquisitions
On issuance, his reference point is "in 1999 1% of the capital market was new issue capital", and he thinks the debt and equity being raised now is approaching that share of market value
Nathan's caveat, which Khan accepted: the deals back then were much smaller
On private marks, "when Kalshi is valued at 40 billion", with Robinhood named alongside it, is the kind of number he means
He also cites a chief executive projecting a trillion-dollar outcome and the stock rising 25% on it for an already sizable business
The conclusion, and the line the episode opened on: "it's hard to argue that the company who's building the AI ecosystem Nvidia is trading at a 15 times" is a bubble
Who Ends Up Holding the Bag
Nathan's closing case is about structure rather than valuation: Meta building a Louisiana data center through a special purpose vehicle with Blue Owl in the middle of it, KKR providing capital and banks lending in, with Meta holding a four-year option on the compute and keeping it off its balance sheet
"who's left holding the bag if there is a pullback in demand"
"Supposedly, there's two trillion dollars of, private credit offbalance sheet sort of stuff"
His historical contrast: "The Nasdaq went up 85% in 1999" despite the new issuance, against Khan's interjection that "Nasdaq is up only 10% this year"
Khan treats the spread of the risk as the interesting development, and says he hopes it is a differentiated view: AI investment has expanded into mainstream industries now that Goldman Sachs, Blackstone and KKR are providing the money
He has not read every contract, but his conclusion is blunt: "It's not only Meta going to hold the bag. Everybody going to hold the bag"
The corollary is that a failure would not be contained to five companies — it becomes a macro problem
"we are heavily invested in AI and I think rightfully so" as a country, and he thinks growing faster is the best route to bringing the debt level down
On probability and risk: "anything can happen there's always a probability of life", and anyone who tells you something cannot happen does not know what they are talking about
"the risk management is equally important as stock picking", and he says podcasts including this one spend their time on stock picking instead
"this risk has been spread out to so many organizations" is precisely why the media framing that Nvidia or Meta or Google or Amazon carries all the risk is wrong — the off-balance-sheet structures exist because other people are taking some of it
Nathan's parting view is darker than his guest's: "the financialization of this whole thing is what is really different", and it brings him back to the late 90s, Lucent, Nortel, Sun Microsystems and the circular arrangements of that era
He expects the eventual unwind to combine an equity blowup with something like the financial crisis, and says the earlier episode will "look like a walk in the park" by comparison
They ran out of time before China, national security and the competitive picture, which Nathan flagged for a future episode
Khan's bottom line is that the thing everyone is calling circular is already in Nvidia's price, and that the risks worth watching are not the ones in the chart of who invests in whom but the ordinary ones — rates, a macro shock, and a build-out whose return arrives later than the capital did.
Products, Companies & Tools Mentioned
Nvidia (The episode's spine: an exceptional quarter, a 70% growth guide against a 45% street consensus, and a multiple Khan says already writes off a quarter of the earnings)
OpenAI (Nvidia's largest customer, one of its investees, and now building its own inference chip — the second point of failure in Nathan's map)
Microsoft (OpenAI's largest investor, one of Nvidia's largest customers, CoreWeave's largest customer, and Khan's example of a stock that worked when the capital expenditure cycle turned into revenue growth)
CoreWeave and Nebius (The neoclouds in the middle of the circular financing, and Khan's example of why a fungible GPU can serve a different customer tomorrow)
TSMC (Locked up by Nvidia at 85% of product, in Khan's account; a great business he does not own because he has no differentiated view on it)
Micron (From negative gross margin to 85%, a 1,000% run, and the plant it is only now digging in Clay, New York)
Samsung and SK (The Korean memory makers behind the supply question, and the buyback SK had just launched)
SanDisk, Seagate and Western Digital (The rest of the memory and storage complex, all lower on the day they recorded)
Salesforce (Growth accelerating by a point, a stock down from $360-some to 150 and back to 245, and about 20 times GAAP earnings on Khan's math)
Anthropic (The Benioff–Amodei appearance with Jim Cramer that moved the stock, the hundreds of millions in token spend Salesforce disclosed, and the SpaceX compute deal Khan cites as evidence of scarcity)
Apple (Squeezed the memory suppliers for years, and now cannot get the access to components it used to)
Meta (The advertising proof point for AI's return, and the Louisiana data center built through a special purpose vehicle with Blue Owl, KKR and bank lending)
Alphabet and Amazon (The hyperscalers Nvidia is diversifying away from; Google's $19 billion quarterly buyback is Nathan's example of a rounding error)
Snap (Khan ran the Snapchat advertising business, which is where his "99% of the ad people see is wasted" example comes from, and took the company public in 2017)
Alibaba (The listing he banked, one of the largest ever at the time)
Kalshi and Robinhood (The private-market marks Khan reaches for when asked where the bubble is)
KKR, Apollo, BlackRock, Blue Owl, Goldman Sachs and Blackstone (The credit providers behind the build-out; Nathan points to the round table Jensen convened with David Solomon and the others willing to backstop chip purchases)
Roundhill's DRAM and Mag 7 ETFs (Nathan's sentiment markers — the DRAM fund was called the top in April and has since taken in tens of billions)
Lucent, Nortel and Sun Microsystems (Nathan's late-1990s comparison for circular arrangements between suppliers and customers)
Books & Resources Mentioned
Bloomberg's chart of the circular AI relationships (Nathan says it keeps being updated and that following the lines is genuinely confusing)
Proem Asset Management's 13F filings (Khan points listeners to them twice, for what the fund owns and does not own)
Get the latest market chatter and takes as they happen:
X | Threads | Instagram | YouTube | TikTok | Facebook