AMD & Meta – Carbon Copy

Good deal. Bad Sign.

• Meta and AMD have signed a supply agreement where AMD shareholders will pay for Meta’s silicon chips, but at the same time, it is demonstrative of Meta’s weakening financial position.
• Meta has agreed to purchase 6GW of AI chips from AMD (roughly 6m chips) that will be delivered in 1GW tranches over the next 5 years.
• AMD is issuing Meta 160m warrants (options on newly issued shares) with a strike price of $0.01 that vest along virtually the same conditions as the deal with OpenAI.
• As Meta deploys each GW of capacity, 26.7m warrants become available for exercise as long as the share price of AMD is above a certain threshold.
• The price threshold for the last tranche is $600, and so I am estimating that the threshold increases by around $66 each time.
• This means that the threshold for 2026 will be around $266, increasing to around $333 in 2027 and so on.
• This is a fantastic deal for both AMD and Meta because if all goes well, AMD shareholders pay for part of the deal through 10% dilution, with the other part being paid for by marginal buyers of AMD shares whose purchases drive the price beyond the required thresholds.
• As AMD’s share price crosses the thresholds, Meta can exercise the options, sell the shares in the market and use the cash to buy the next tranche of AMD chips.
• It also works reasonably well if the AI bubble pops because AMD’s share price will not hit the thresholds, and Meta will no longer be under so much pressure to invest everything that it has in trying to keep up with OpenAI, Anthropic, Google, etc.
• This is why this deal (and all of the huge deals announced) are conditional, and there is no penalty on either side for not fulfilling the terms of the arrangement, as we saw when OpenAI’s $100bn deal with Nvidia collapsed.  
• However, while this is a good deal for Meta, as all being well, it gets the chips for free, I take it as a sign of the deterioration in both Meta’s financial performance and its balance sheet.  
• In Q4 2025, revenues grew by 24% YoY, with 28% YoY expected in Q1 2026, of which, 4% is due to the weakness in the USD.
• If I assume that this growth rate persists for FY 2026, then revenues would be expected to be $257bn for the full year, which is broadly in line with the consensus estimate of $250bn.
• Expenses of $165.5bn gives operating profit of $91.5bn, which amounts to net income of $79.6bn after tax of 13%.
• This is the equivalent of $30.93 per share, giving YoY EPS growth of 31%, but this is flattered by the one-off higher tax rate incurred in 2025.
• This is substantially better than I had previously forecast, as I made the error of misinterpreting Meta’s expenses forecast as OPEX as opposed to all expenses, including cost of goods sold which is what Meta meant.
• However, even fixing this error does not paint a great picture.
• If I reverse the abnormal tax charge out, the underlying EPS growth for 2026 is 6.5%, clearly demonstrating the impact of the large increase in expenses and decline in profitability as a result of participating in the AI spending race.
• The impact on cash flow will be more significant as 2026 capex is expected to be $125bn, which will consume all of the incremental growth in cash flow from excellent top-line performance.
• Hence, I expect that Meta will come to the bond market to finance some of its capex, given that cash flow is unlikely to be able to cover all of the expenditure.
• Raising debt is precisely what one does not want to see in this environment, as debt has to be paid back no matter what happens with AI.
• It was debt issuance that exacerbated the internet bubble in 2001, and seeing companies raise debt to finance excessive capital spending fills me with disquiet.
• Debt issuance is not limited to Meta, as Alphabet has also recently come to the market, and I expect to see Amazon also come back for more.
• All in all, Morgan Stanley has estimated that $400bn in new debt will be raised by US companies to spend on AI infrastructure, which is great for the suppliers but raises the risk and consequences of a correction substantially.
• Meta is on 21x 2026 PER, but frankly, I don’t want to be anywhere near this even though it is now one of the cheapest of the Mag 7 as it has no direct route to earn a return on its capex, and it remains a laggard in AI, having lost open source to China.
• I would prefer to be positioned in the suppliers where I have positions in Qualcomm and Samsung Electronics and nuclear power, where I hold a range of uranium names.
• The software and services sector has been hammered by the “AI Scare Trade”, and this is rapidly becoming both a great investment opportunity in select names as well as a way to hedge the AI bubble risk.
• I am looking closely at this sector for a few high-quality names that can capitalise on the AI trend rather than be decimated by it.