Lisa Su Didn't Save AMD From Bankruptcy. She Did Something Harder.

When Lisa Su became CEO of AMD on October 8, 2014, the stock traded for less than a cup of coffee, and the company's entire stockholders' equity - everything the shareholders owned after the debts were counted - had shrunk to $187 million.¹² Against that sliver of equity sat a buried scar most investors never read: an accumulated deficit of $6.646 billion, the running tally of every dollar AMD had lost and never made back.² That is the picture people remember as 'near-bankruptcy.' It is the wrong picture - and getting it right is the whole point.

The official story is that Su rescued AMD from the brink of insolvency. It's a great story and it isn't true. AMD never filed Chapter 11, never lost its going-concern status, and never stopped paying its bills. What it had lost was something subtler and harder to fix: relevance. Intel had taken the data center, the high-margin heart of the industry, and was treating AMD as a fading nuisance. The turnaround that mattered wasn't financial triage. It was a decade-long architectural bet.

Distressed is not the same as dying

The financial reality was grim but specific. AMD lost $403 million in 2014 - and that was an improvement, with full-year revenue actually up 4% to $5.51 billion.² (The $4 billion figure people quote belongs to 2015, the year after Su arrived, not the year she took over.) By the first quarter of 2015, total debt had reached $2.27 billion, free cash flow ran negative $195 million in the quarter, and the company's own 10-K warned in plain federal language that it 'may not be able to generate sufficient cash to service its debt obligations.'³ That is the language of a company on a tightrope. It is not the language of one in receivership. The distinction is not pedantry - it changes what kind of fix was needed. A company days from insolvency cuts everything and prays. A distressed-but-solvent company can do the one thing AMD desperately needed: keep spending on the long-horizon engineering that wouldn't pay off for years.

Su was not a parachuted-in fixer, either. She had already been inside the building for nearly three years - joining in January 2012, running the global business units, then serving as COO from July 2014 before the board handed her the top job that October.⁸ Before AMD she was CTO at Freescale Semiconductor, and before that spent thirteen years at IBM, including a stint running semiconductor R&D.⁸ This matters to the strategy. The fix wasn't financial engineering by an outsider; it was a bet on silicon by someone who had personally led the building of silicon. She knew exactly how long the bet would take to mature, and exactly why Intel couldn't simply match it on command.

The bet was on architecture, not austerity

Here is the move that actually turned AMD around, and it had almost nothing to do with the balance sheet. Su committed the company to a new ground-up CPU design - the Zen microarchitecture - and to building big server chips out of smaller, cheaper interconnected dies rather than one enormous monolithic slab of silicon. That second choice is the quiet genius. When you cut a giant chip into smaller pieces, more of them come off the manufacturing line working, so each one costs less. Intel was still betting on monolithic designs and on its own factories. AMD, already fabless, leaned fully on the most advanced contract manufacturing it could buy. The result was that AMD could field a chip with more cores, at a lower cost to produce, while Intel's own factory stumbles slowed its road map. AMD wasn't winning a sprint. It was changing the shape of the track.

In May 2017 that bet hit the market with a name: EPYC, AMD's re-entry into the server business it had effectively abandoned years earlier. The first EPYC 7001 chips offered up to 32 cores in a single socket, launched in June 2017 - into a market where Intel held roughly 98% share.⁶ Re-entering a market your rival owns 49-to-1 is not a turnaround tactic. It is a declaration that you intend to spend a decade taking it.

Why a 49-to-1 deficit became a 30% share

The compounding is the part that looks like magic and is actually mechanism. EPYC's server CPU unit share climbed from under 2% in 2018 to roughly 30% by early 2026.⁷ EPYC revenue went from under $100 million in 2017 to over $3.5 billion in 2025, while Intel's server shipments slid from about 12 million units in 2019 to under 8.5 million in 2025.⁷ Each generation of architectural lead made the next one easier: more design wins funded more R&D, more cores per socket made AMD the obvious choice for cloud buyers counting performance per dollar, and Intel's factory delays meant it kept arriving late to its own fight. Architecture, unlike a price cut, doesn't reset every quarter. It accrues.

Run the numbers all the way to the present and the scale of the climb is hard to overstate. From $5.51 billion in revenue in 2014, AMD reached $25.8 billion in 2024 - with its Data Center segment alone up 94% that year to $12.6 billion - and a market cap of roughly $261 billion by mid-2024.²⁴ In 2025 it posted a record $34.6 billion in revenue, $4.3 billion in net income, and $16.6 billion of data-center revenue.⁵ The company that once owned $187 million of net equity now earns more than that in profit every two weeks.

But did AMD actually beat Intel?

The honest objection is that 'AMD beat Intel' is a headline, not a finding. It's true that AMD's market cap has at times exceeded Intel's - but market cap is volatile and point-in-time, a snapshot of mood as much as merit. And across all CPU segments combined - desktop, notebook, and server - Intel remained the largest vendor by unit shipments as of early 2026.⁷ So the triumphalist version overstates the case. The fair reading is narrower and more impressive for being true: AMD took a market Intel controlled almost completely and clawed it to roughly a third in the data center, the most profitable arena in the business.⁶⁷ That is not a knockout. It is something Intel feared more - a rival that keeps closing the gap a node at a time, on a structural advantage that doesn't expire.

AMD was never minutes from death; that's the part the legend gets wrong, and it's the part that matters. A company minutes from death cannot place a ten-year bet. AMD could - and did - because it was distressed, not dying, and because the person now running it had built chips before and knew the only way out was through the silicon. The sticky truth of this turnaround isn't that Su rescued a corpse. It's that she stood on a tightrope with $187 million of equity and chose to spend on a war that wouldn't show a result for years. The rescue would have saved the company. The bet rebuilt it.