← AI Chips
AI Chips TraderMagz Research Horizon 2 to 5 years

AI Chips Beyond Nvidia: Broadcom and Intel

Broadcom's price needs free cash flow to grow about 10% a year for a decade. Intel's, 58% off the high, still needs about 17%, or about $18 billion a year if the growth is never faster than a normal business.

23 min read Current · 1d
Live terminal state
Trend rank
AVGO
Broadcom Inc.
Unlock
INTC
Intel Corporation
Unlock

The thesis is free. The live trend rank on these names updates continuously in the terminal — create an account to see it.

Everyone will watch Nvidia on Tuesday 17 November, and that print will not tell you which second-order bet is the better one from here. Broadcom (AVGO) designs custom AI chips and the networking around them, and the market already pays for it as a compounder. Intel (INTC) is the factory turnaround whose shares ran a long way on hope and have since fallen 58% from the high, which leaves a real question: a bargain the crash created, or the same hope at a lower price.

General commentary for a general audience, not personalised advice. Share prices are US closes on Friday 13 Nov 2026 (Yahoo Finance): Broadcom $347.01, Intel $59.66. Broadcom figures are from the 2 Sep 2026 results release, the Form 10-Q for the quarter ended 2 Aug 2026, and the earnings call that day. Intel figures are from the 29 Oct 2026 results release, the financial tables published with it, and the earnings call that day. The government stake is from Intel's August 2025 agreement and its 2026 proxy. Anything marked as our calculation or our assumption is ours. The terminal reads are the TraderMagz board as of 16 Nov 2026, about 06:54 UTC.

The quarter behind the prices

Nvidia set the date itself. On 28 Oct it said it will publish third-quarter results for fiscal 2027, the quarter ended 25 Oct, at about 1:20 p.m. Pacific on Tuesday 17 November, with the call at 2 p.m. Pacific. Both of the stocks in this note will move with that tape. Neither of them is the company reporting.

Broadcom's last report is the fiscal third quarter, ended 2 August, published on 2 September. Revenue was $29.6 billion, up 86% from a year earlier. AI semiconductor revenue was $16.7 billion, up 221% from a year earlier and 54% from the prior quarter. Free cash flow, which is the cash left after the spending needed to run and grow the business, was $13.7 billion. Broadcom's version is cash from operations of $14.2 billion minus $0.5 billion of capital spending, and it was 46% of revenue. The company guided the fourth quarter, which ended 1 November, to about $34.8 billion of revenue and $21.7 billion of AI semiconductor revenue. It reports that quarter on Wednesday 9 December, after the US close.

Intel's quarter ended 26 September and was published on 29 October. Revenue was $16.5 billion, up 21%. Adjusted free cash flow was $1.6 billion. Cash from operations was $5.1 billion. Gross capital spending in the quarter was $4.2 billion. The company's own definition of adjusted free cash flow adds back government incentives and partner contributions and subtracts finance-lease payments, so the $1.6 billion includes $847 million of partner money. The guide for the December quarter is revenue of $15.0 to $16.0 billion, below the quarter just printed, and a non-GAAP gross margin of 38%, against 44.1% in the September quarter.

The share prices have not waited for our write-up. Broadcom closed at $347.01, down 4.8% on the day, and Yahoo Finance puts the one-year total return at 2.8% against a three-year return of 279%. The earnings explosion of the last year did not re-rate the stock. It caught up with a re-rating that had already happened. Intel closed at $59.66. The 52-week range on the same Yahoo page is $32.89 to $142.35, so the close is 58% below the high and still 81% above the low. The one-year return is 66%. The hope rally is in the one-year number. The giveback is in the drawdown.

Our own terminal is a poor quote source for Intel and a fine one for Broadcom. The board shows Broadcom at $346.62, a verdict of Accumulate, conviction 87, timing 61, outperforming its AI Chips group, with a fundamentals snapshot dated 29 Sep 2026 that the board marks fresh (quality 8.9, value 4.2). That price is the Friday close for any decision that matters, and we use $347.01 in the arithmetic below. The same board shows Intel at $115.93, verdict Tradable, not core, conviction 55, timing 52, flagged extended, lagging the AI Chips group, on a fundamentals snapshot dated 1 Oct 2026 that the board marks stale (quality 5.4, value 0.5). $115.93 is the 29 September close, not a price you can trade. We do not use the board's Intel price or its Intel fundamentals. The market-health read on the same board is healthy with watch items, a score of 59, neutral to mixed, and both the Technology sector and the AI Chips theme sit in the lagging quadrant of the rotation. This is not a tape that is paying up for the group.

What the two companies sell

An ASIC is a chip designed for one job, rather than a general-purpose processor anyone can buy. Broadcom's AI business is ASICs. The company calls the custom accelerators it designs with individual customers XPUs. In the quarter it shipped Google's Ironwood processor, the seventh-generation TPU, in high volume to Google and to Anthropic, and it began production shipments of the next TPU to Google. It shipped OpenAI's first custom accelerator. It expected production shipments of Meta's custom accelerator in the fourth quarter. The chief executive, Hock Tan, said the company has six of these XPU customers, and that four of them will be very large. AI semiconductors were 56% of revenue. The semiconductor segment as a whole was $20.8 billion, 70% of revenue, up 127%. Take out the $16.7 billion of AI and the rest of the semiconductor business was about $4.1 billion (our calculation). Alongside the chips sits infrastructure software, the business that includes VMware, at $8.8 billion, up 29%. Non-GAAP gross margin was 75% of revenue, down 2.1 percentage points from the prior quarter because AI chips were a larger share of the mix, and the semiconductor segment's own gross margin was about 67%. Non-GAAP operating margin was 67.9%, a level the chief executive called a record. That is the company's point: operating costs do not rise with the revenue. GAAP operating income was $16.0 billion. The gap between the two is mostly amortization of acquired intangibles, about $2.0 billion, and stock-based pay of $2.0 billion.

On the same call Tan said the company had supply secured to put fiscal 2026 AI semiconductor revenue at $58 billion, up 186%, to double that to about $115 billion in fiscal 2027, and had line of sight to double again to $230 billion in fiscal 2028. Those are his outlooks, not results. He also said the limit he watches is whether the chips can actually be installed: land, power, and the building. On 26 October, Bank of America, Blackstone, Citi and Morgan Stanley announced a $63 billion financing vehicle that will buy Broadcom accelerators and networking gear and lease them to Anthropic, with deliveries expected to start in calendar 2027. Broadcom is providing residual-value support to that vehicle equal to the first-lien portion of the financing. In plain words, Broadcom keeps a piece of the downside if the equipment is worth less than the senior debt at the end of the lease.

Intel sells two things that the accounts push together. The product groups design CPUs. The foundry is the factory network that makes chips. A foundry, in the usual sense, manufactures for outside customers who design their own chips and do not own a fab. Intel's foundry revenue was $6.7 billion, up 58%, and almost all of it was Intel making Intel's own chips. The intersegment elimination was $6.3 billion. External foundry revenue, the part a real outside customer paid, was $411 million, and $118 million of the sequential increase was a program the company described as nonrecurring. The foundry lost $1.9 billion of operating income. The product groups made $5.3 billion. Inside that, the data-center group was $7.3 billion of revenue, up 78%, and the client group was $8.1 billion, down 5%.

The process technology is where the hope lives. Intel 18A is the company's newest manufacturing process, now in high volume, with yields the chief executive said improved at a double-digit rate in the quarter and reached what he called industry-standard levels, ahead of Intel's own plan. The cost of an 18A wafer is down about 40% since the first quarter. 18A-P is the follow-on, aimed at internal products by year-end. Intel 14A is the node after 18A. The company released a 0.9 process design kit for 14A to selected customers in October, an early version of the design file a customer needs in order to start a chip on that process. Tan said Intel is talking to a couple of highly strategic customers, and that a couple of customers could be ready to be named in the next two quarters. He has not named them.

The balance sheet was rebuilt to pay for this. In early August Intel completed a $23 billion equity offering. Net proceeds in the nine-month cash flow statement were $22.6 billion. Shares issued and outstanding rose from 4.994 billion on 27 December 2025 to 5.310 billion on 26 September, and the company expects 5.44 billion diluted shares in the December quarter. The US government is already on the register. Under the August 2025 agreement the Department of Commerce bought 433.3 million shares at $20.47, a 9.9% stake at the time, paid for with $8.9 billion of CHIPS Act and Secure Enclave money that had been grants. The stake is passive, with no board seat. A five-year warrant at $20 covers a further 5% if Intel ever owns less than 51% of the foundry. The March 2026 proxy still showed 433.3 million shares, 8.4% of the company at that date, and that count assumes escrowed shares are fully released. More stock has been issued since. We do not have a later ownership filing, so we will not update the percentage ourselves.

Broadcom Intel
Quarter Fiscal Q3, ended 2 Aug 2026 Quarter ended 26 Sep 2026
Revenue $29.6 billion, up 86% $16.5 billion, up 21%
The AI line AI semiconductors $16.7 billion, up 221% External foundry revenue $411 million
Profit we watch Non-GAAP operating margin 67.9% Products operating income $5.3 billion; foundry loss $1.9 billion
Cash Free cash flow $13.7 billion Adjusted free cash flow $1.6 billion, including $847 million from partners
Next guide Revenue about $34.8 billion; AI semiconductors $21.7 billion Revenue $15.0 to $16.0 billion
Friday close, 13 Nov $347.01 $59.66

What the price already assumes

A reverse DCF runs a valuation backwards. The ordinary version forecasts cash flows and arrives at a price. This version starts from the price and asks what the business has to deliver for that price to be fair. Ours is deliberately plain, and it is the same shape we used on Palantir and Super Micro. We use a 9% required return and, after year ten, 3% growth forever. Both are our assumptions. Change them and the levels move. The order of the two stocks does not.

Broadcom. The 10-Q cover counts 4,773,629,865 shares as of 28 August 2026, the latest share count the filing gives. At $347.01 that is $1.66 trillion of equity (our calculation). Debt was $59.4 billion and cash was $24.0 billion, so net debt is $35.4 billion, about 2% of the equity. We compare the cash flows to the equity value. Adding the $35 billion of net debt to the hurdle lifts the required growth by less than a percentage point. The starting cash flow is the quarter's $13.7 billion times four, about $55 billion. That is a run-rate, not a trailing twelve-month total. It is a generous base against the first nine months, which produced $31.9 billion of free cash flow, about $43 billion annualized, because the earlier quarters were smaller. It is a modest base against the company's own next quarter, which guides revenue 18% above the quarter just reported.

The growth rate that makes $55 billion, compounded for ten years, worth $1.66 trillion is about 10% a year. Year-ten free cash flow on that path is about $147 billion. If the cash margin stayed near today's 46%, that would be on the order of $320 billion of revenue, about three times the $106 billion you get by adding the $71.1 billion of revenue in the first nine months to the $34.8 billion fourth-quarter guide (our calculation). At an 8% required return the growth rate falls to about 8%. At 10% it rises to about 13%. Subtract the quarter's $2.0 billion of stock-based pay before you annualize, so the dilution is treated as a cash cost, and the required growth rises from about 10% to about 13%. Use the slower nine-month pace as the starting point and it is about 14%.

Ten percent sounds undemanding next to a quarter that grew 86%. The dollars are the part that is not undemanding. Very few companies have ever produced $147 billion of free cash flow in a year. What the price does not assume is the outlook Tan just gave. A double to $115 billion of AI revenue next year, and another double to $230 billion the year after, is a boom. A 10% decade is a fade from a boom. The paths around the fade are the whole argument:

Illustrative path (our model) Equity value Versus $1.66 trillion
8% free-cash-flow growth for 10 years about $1.37 trillion about 17% lower
About 10% for 10 years today's price in line
15% for 10 years about $2.35 trillion about 42% higher
20% for 10 years about $3.42 trillion about double

Intel. Outstanding shares were 5.310 billion on 26 September. At $59.66 that is $317 billion. Cash and short-term investments were $54.0 billion and debt was $50.1 billion, so the company has about $4 billion of net cash. There is also $16.6 billion of equity that belongs to other investors in subsidiaries. Neither item moves the conclusion. The starting cash flow is the quarter's adjusted free cash flow of $1.6 billion times four, about $6.6 billion. That count includes the partner contributions. Without them the run-rate is about $3.2 billion.

On the $6.6 billion base the price needs about 17% a year for ten years. Year-ten free cash flow is about $30 billion. At an 8% required return the growth rate is about 14%. At 10% it is about 19%. Strip out the partner money and the required growth goes to about 26%. There is a plainer way to say the same thing. If free cash flow never grows faster than 3% a year, the assumptions need a starting free cash flow of about $18 billion a year. The quarter's $1.6 billion is a run-rate of about $6.6 billion, so the price needs that annual cash flow to be nearly three times as large before it is allowed to grow like an ordinary business.

Illustrative path (our model, $6.6 billion base) Equity value Versus $317 billion
10% free-cash-flow growth for 10 years about $192 billion about 39% lower
15% for 10 years about $281 billion about 11% lower
About 17% for 10 years today's price in line
20% for 10 years about $410 billion about 29% higher
25% for 10 years about $593 billion about 87% higher

The bull version of Intel is not crazy, and it should be stated with the numbers it actually has. The product groups earned $5.3 billion of operating income in the quarter, about $21 billion if you annualize it, and $317 billion is about 15 times that figure (our calculation). The company expects industry server-CPU units to grow more than 30% in both 2026 and 2027, and it says it cannot ship enough Xeon. On that telling, the products business already earns, in operating profit, something close to the $18 billion of cash flow a slow-growth price would need, and the foundry loss plus the capital spending is a gap that closes as 18A yields improve. GAAP operating income was $2.5 billion in the quarter, about $10 billion annualized, and on that number the equity is about 31 times. The gap between 15 times and 31 times is the foundry. The reverse DCF is asking whether that gap closes.

The bear case, at full strength

Broadcom's bear is concentration, the financing, the margin mix, and the size of the number at the end of the decade. The product can keep working and the stock can still be the wrong price if the fade is faster than 10%, or if the cash margin does not survive the mix.

The customer bear is in the 10-Q, not in a rumor. Direct sales to one distributor were 50% of revenue in the quarter and 46% over nine months. The top five end customers, through all channels, were about 55% of the quarter and about 50% of the nine months. A year earlier the top five were about 40%. The AI boom has made the company more concentrated, not less. Tan said the financing vehicles are for two of the six customers, Anthropic and OpenAI, and that the other four can fund themselves. That is a useful distinction and it is not a comfort. Half the revenue still runs through one distributor, and the two customers who need help borrowing are the ones he expects to be among the largest.

The financing bear follows from the 26 October announcement, and from a warning that was already in the 10-Q for the quarter ended 2 August. The filing says some AI customers may be unable to pay for the infrastructure and may use alternative financing, including Broadcom's own financing platform, which adds credit risk. Residual-value support means the $13.7 billion of quarterly free cash flow is not the last word on who holds the risk of the equipment. A lease that comes back to the manufacturer is a sale with a string attached. We do not know the size of the first-lien piece, because the announcement does not split the $63 billion, and we do not pretend to. The direction is enough. Broadcom is helping its customers buy Broadcom's chips.

The margin bear is the sequential comparison the chief financial officer printed. Non-GAAP gross margin fell as AI became a larger share of revenue, and she said memory content is part of that mix. The reply from the company is to watch operating margin, which rose, and the fourth-quarter guide holds non-GAAP operating margin at about 66%. That reply is coherent for a business whose operating costs are close to fixed. It stops being coherent if gross margin keeps giving ground and the support contracts start costing cash. The 10-Q also says the company depends on a limited number of suppliers and on contract manufacturers. A custom chip is only as shippable as the packaging, the memory, and the substrate.

The valuation bear is the right-hand column of the table, read in dollars rather than in percent. You can believe every word of the quarter and still need the company to be producing something like $147 billion of free cash flow in a decade for the present price to have been fair. A future in which growth becomes merely good, 8% for a decade from an already enormous base, is a gap of about 17% on our model. That is not a collapse. It is also not free. And the $115 billion and $230 billion figures are supply the company says it has secured and demand it says it can see, on the condition that buildings and power show up on time. Tan was explicit that an ordered chip which cannot be installed is not revenue he is willing to promise.

Intel's bear is that the crash already happened, and the price that survived it is still a foundry price.

The valuation bear is the second table. A 10% decade, which is roughly the hurdle Broadcom clears, is worth about $192 billion on Intel's generous cash-flow base, about 39% below the equity. On our arithmetic the shares would have to trade at about $37, close to the 52-week low, before today's adjusted free cash flow only had to compound at Broadcom's rate. That $37 uses the definition that counts partner contributions as cash flow. Without them, the same 10% hurdle is about $18 a share (our calculation). Falling from $142 to $60 removed the most extravagant version of the hope. It did not remove the hope.

The foundry bear is a single comparison. Outside customers paid $411 million. The factory lost $1.9 billion. Gross capital spending is on track for $20 billion in 2026, and management said 2027 spending will be up strongly, with 14A positioned to begin production in 2028. The chief financial officer called 2027 a knothole: 18A and Intel 3 ramping at the same time as the 14A investment. Yields improving and wafer cost down 40% are real operating progress. They are progress inside a business that does not yet have an outside customer to show for the investment. "A couple of customers, in the next two quarters" is a sentence a careful manager says when he does not have a name he can print.

The client bear is the company's own outlook. Intel expects PC units to fall about 12% this year, with the second half down more than 20%, and about 10% again in 2027, because memory is scarce and expensive and PC makers are cutting inventory. Client revenue is guided down more than 20% from the September quarter to the December quarter. The data-center strength is the offset, and it has its own crack: an accident shut the Israel fab, and the company estimates a $300 million revenue hit in the December quarter and a similar one in the March quarter, almost all of it 10-nanometer server chips, plus about 2 percentage points of gross margin in the December quarter. The December-quarter guide of $15.0 to $16.0 billion is what all of this looks like together. A server story that cannot ship enough, sitting next to a PC story the company expects to shrink for two years, is a weaker support for a $317 billion equity than the 78% data-center growth rate on its own.

The dilution bear is the offering you can already see. The share count is up by more than 300 million shares since the end of 2025, the December quarter is expected to carry 5.44 billion diluted shares, and the nine-month GAAP result is a loss of $11.6 billion that includes a $12.6 billion mark-to-market on the shares tied to the government escrow. That mark-to-market moves with the stock price and says nothing about the factories. The non-GAAP earnings of $0.52 a share are the cleaner operating number, and they are the earnings of a company that just sold $23 billion of stock because the factories need the cash. The government stake is strategic support. It is also a reminder of who had to be asked.

What would change our mind

For Broadcom, the thing that breaks the preference is a December report in which the boom is no longer the thing being faded. The fourth-quarter guide is $21.7 billion of AI semiconductor revenue and about $34.8 billion in total. If those figures are missed by a wide margin, or if the fiscal 2027 outlook of about $115 billion is walked back, then 10% a year stops being a slowdown from a stated plan and becomes the optimistic case. The same break happens if the residual-value support turns into cash leaving the company, because the $13.7 billion quarter would then overstate what shareholders keep. A higher share price that pushed the required growth into the high teens would change the expression, not the business: we would still regard Broadcom as the better company, and we would stop calling the asymmetry favorable.

For Intel, the thing that flips the ranking is a named external customer for 18A or 14A, with a volume large enough to put into a cash-flow line. That would mean the $1.6 billion is the wrong base, and the option value we are refusing to pay $317 billion for would have become a contract. The same flip happens if adjusted free cash flow steps up, without leaning on partner contributions, far enough that the required growth falls toward 10%. A further rally on a process update that does not name a customer would not do it. The shares already travelled from the low $30s to $142 on that kind of news, and they travelled back.

Ranked by asymmetry

The ranking is by asymmetry: the reward if you are right, set against what the price already assumes. It is not a ranking by which business is finer, and it is not a ranking by which stock has fallen more.

Broadcom is first. The price assumes free cash flow slows to about 10% a year for a decade, off a quarter that grew 86% and off an outlook that describes two doublings. You do not need the $230 billion to earn a return. You need the slowdown to be merely sharp rather than brutal. On our table, 15% growth for a decade is worth about 42% more equity than today's price, and 8% is worth about 17% less. That is a favorable shape. The risk attached to it is the one in the bear case: half the revenue through one distributor, a financing structure that leaves Broadcom on the hook for residual value, and a year-ten cash flow of $147 billion that has to actually arrive.

Intel is second. The 58% fall from the high is why the question feels close, and it is the wrong measure. At $60 the shares still need about 17% growth on a cash-flow definition that includes partner money, or about $18 billion a year if growth is never faster than 3%. The product groups are a real business. The foundry, measured by outside revenue, is not yet one. The asymmetry improves on the day a customer is named. It does not improve because the chart has gone down.

The board's verdicts line up with that order for a different reason. Accumulate, at a conviction of 87, against Tradable, not core, at 55. We agree with the order. We do not use the board's Intel price to get there. The extended flag is pinned to $116. The live question is $60, and $60 is still a hope price relative to cash.

What we are watching

The item that will move the prices first is Nvidia's print on Tuesday 17 November, after the close. It sets the mood for every AI silicon stock. It does not contain Broadcom's custom-chip revenue or Intel's foundry pipeline. A Nvidia guide that says customer spending is still accelerating supports both names. A guide that cracks the spending narrative hurts both. We would treat the move as information about the customer, and we would not rewrite either cash-flow hurdle from it.

The item with the most information about Broadcom is the 9 December report. The questions are narrow. Did AI semiconductor revenue land near $21.7 billion? Did total revenue land near $34.8 billion? Does management repeat, raise, or hedge the fiscal 2027 figure of about $115 billion? Any of those three answers changes the fade we are underwriting.

The item with the most information about Intel is a customer, not a quarter. Tan said a couple of customers could be ready to announce over the next two quarters. A name, a process node, and a volume would change the model. A yield update, a kit release, or another description of conversations would not. Behind that, the December-quarter report will show whether the CPU franchise held together through the Israel shutdown and the PC inventory correction. That tells you about the business that already makes money. It does not answer the factory question the equity is paying for.

Bottom line

Broadcom is the better asymmetry. The price assumes a slowdown from a boom the company has already put numbers on, and the risk is that the boom is concentrated in a few customers and partly financed with Broadcom's own residual-value support. Intel at $60 is a good CPU business attached to a foundry option the share price still treats as likely. We would rather own the cash flow that is already in the accounts, at a hurdle of about 10% a year, than the cash flow that has to be built, at a hurdle of about 17%, until a customer makes the build visible. None of this is a recommendation to buy or sell either share. It is a ranking of what the two prices already believe.

Names in this note AVGO INTC

The thesis is half of it.

Every name we write about is scored continuously in the terminal — market regime, sector rotation, fundamentals and technicals resolved into one Conviction × Timing verdict.

Open the terminal