Here is the short answer to a question readers keep sending: The Magnificent Seven is a 2023 stock-market label, not a ranking of AI companies. That is why AMD, Qualcomm, and Oracle sit outside it and still do serious AI work. The more interesting story sits in the funding. The seven have turned to the bond market to pay for AI build-outs, and the coupons they pay now shape what every chip vendor and cloud buyer can charge. Oracle reports September 10.
A Bank of America analyst coined the “Magnificent Seven” in 2023 to name the exact seven mega-cap stocks then supplying most of the S&P 500’s return: Apple, Microsoft, Amazon, Alphabet, Nvidia, Meta, and Tesla. The label froze on that date, the way FAANG once excluded Microsoft. Three AI companies that matter to silicon teams sit outside it, for three different reasons.
AMD missed the inclusion on arithmetic and on timing. When membership settled, the seven carried valuations in the trillions or high hundreds of billions, and AMD ran far below that line. Nvidia had taken the AI hardware narrative, and Wall Street filed AMD as the challenger. AMD now carries a market cap around $826 billion, ranking 19th among public companies. The Instinct MI300 and MI350 accelerators put it in production at hyperscale, with OpenAI, Meta, and Anthropic committing to a combined 14 GW of AMD GPUs. AMD qualifies as an AI principal. It does not qualify for a 2023 branding exercise.
Qualcomm misses on scale by two orders of magnitude. Its market cap sits near $185 billion, roughly 106th in the world. The seven together clear $23.5 trillion as of September 9, 2026. Several members individually passed $4 trillion. Qualcomm sells components and licenses patents. The label rewards ecosystem owners, from Apple’s devices to Nvidia’s accelerators. Qualcomm’s core handset business also absorbs the loss of the Apple modem socket. Its September 8 agreement with Amazon, with a warrant laddering to $60 billion of purchases, changes Qualcomm’s data center prospects. Entering the top seven takes a hundredfold re-rating that no single agreement delivers.
Oracle presents the sharpest case, because it owns platform-scale AI capacity and still sits outside the list. Oracle reports fiscal first-quarter results after the close on Thursday, September 10, 2026. Its fiscal 2027 plan calls for roughly $70 billion of net cash capital spending, and customer prepayments and timing push reported CapEx toward $90 billion to $95 billion. Fiscal 2026 CapEx hit $55.7 billion against $21.2 billion in fiscal 2025, a 162% increase, and free cash flow ran to negative -$23.7 billion. Oracle plans to raise about $40 billion of debt and equity in fiscal 2027 after issuing $43 billion of senior notes in fiscal 2026. The share price fell as investors weighed component costs, competition from capacity landlords including SpaceX, and concentration in OpenAI. Oracle’s database and enterprise software still generate cash, and their share of the business shrinks each quarter.
Table 1. Three AI heavyweights the 2023 label leaves out.
Neoclouds, including CoreWeave, buy specialized AI chips and rent that capacity out to customers directly. They rank among the fastest-growing companies in the AI industry today, and they also rank among the most vulnerable. Many neoclouds started as crypto miners originally, then repurposed their existing data centers for AI workloads specifically. Their exclusive focus on AI, instead of a broader range of cloud services, let them concentrate their capital investments directly. Nvidia has provided various forms of financial support to help them buy more of its chips.
Figure 1. Neocloud debts. (Source: The Economist)
Their biggest customers now include model-makers like OpenAI and the hyperscalers themselves, who frequently lease extra capacity from the neoclouds to supplement their own infrastructure.
The debt tells the real story
Follow the funding and the seven look different. Group bond issuance reached $134 billion in 2026 through late May, against $87.5 billion for all of 2025, on Dealogic figures, with Alphabet, Amazon, and Meta driving the increase. One widely circulated analyst tally put the group’s aggregate debt near $400 billion at the end of 2025, a level it had not carried before, with Alphabet’s burden rising 2.8 times year over year. That figure reflects one analyst’s aggregate, and no company filing states it.
The pricing carries the signal. Alphabet raised $31.51 billion across currencies in February, including a £1 billion (US $1.35 billion) 100-year tranche at a 6.125% coupon, tech’s first century bond since the 1990s, with demand near 10 times the amount offered. Meta priced a $25 billion package on April 30 with 2056 bonds at a 6.30% coupon. Amazon returned on July 7 for $25 billion, taking its 2026 issuance to $92 billion. Investment-grade AI-related issuance from hyperscalers, data center developers, and related firms reached $218 billion through July 8, against $80.5 billion for 2025, and high-yield AI paper added $31.9 billion. Investors demand more for the thinner credits: SpaceX 30-year bonds priced at 6.65%, and CoreWeave’s six-year notes carry a 9.625% coupon and traded at 96.50 for a 10.42% yield.
Table 2. What AI money costs, by borrower and maturity.
Capital spending from Alphabet, Microsoft, Amazon, and Meta will total at least $630 billion in 2026. Goldman Sachs expects big tech to fund more than a third of its AI investment with debt in 2027. Operating cash flow paid for the first wave of AI CapEx. Bondholders pay for the next one.
What silicon teams and IT buyers should take from it
Silicon teams should read the coupon curve as a demand forecast. A hyperscaler paying 6.3% for 30-year money underwrites accelerator purchases against a fixed interest bill, which raises the value of performance per watt in every socket decision. Vendors that improve tokens per dollar meet buyers holding a hard number to beat.
CIOs and IT decision-makers get a pricing signal with a two-year fuse. Debt-funded capacity carries a fixed service cost that lands in cloud rate cards, and providers with negative free cash flow have an incentive to fill capacity through longer commitments.
The Magnificent Seven label describes a moment in 2023 and answers no question worth asking in 2026. The list to watch runs on different criteria: who owns the accelerator roadmap, who owns the power interconnect, and who can service the coupon when the build slows. AMD, Qualcomm, and Oracle each hold a real position in AI infrastructure, and none of them holds a seat at the branded table. The seats matter less each quarter, and the bond covenants matter more.
What do we think?
The label has outlived its usefulness; the balance sheets replaced it. Watch coupon spreads between Alphabet at 6.125% and CoreWeave at 9.625%, because that gap prices the market’s confidence in who fills the halls. AMD and Oracle hold real AI positions. Qualcomm holds a credible new one. None of that shows up in a 2023 acronym. Read the debt.
Inflection point. The financing marks the inflection point. AI CapEx has outgrown operating cash flow, so the seven and their peers now issue 30-year and 100-year paper to fund three-year silicon cycles. Fixed interest bills convert AI capacity into a utility that has to run at high utilization for decades. That discipline reaches every vendor: performance per watt and tokens per dollar stop being marketing claims and start showing up in bond covenants and cloud rate cards. Expect covenants to shape roadmaps.
LIKE WHAT YOU’RE READING? TELL YOUR FRIENDS; WE DO THIS EVERY DAY, ALL DAY.