By | Beyond the Headlines, Written by | Banjun
On August 2, 2024, Intel's stock plummeted 26% overnight, falling back to levels seen a decade ago, marking its worst single-day decline in years.
Three months later, a more symbolic turning point arrived. On November 8, 2024, NVIDIA officially replaced Intel as a component of the Dow Jones Industrial Average.
An index adjustment was also a verdict of the era. GPU replaced CPU, NVIDIA replaced Intel. This industry honor, which Intel had held for 25 years, finally changed hands, becoming the most vivid symbol of its decline.
On April 23, 2026, after Intel released its earnings report, its stock surged nearly 20% in after-hours trading, staging a strong rebound. Less than two years had passed in between.
From being judged obsolete by the era to rebounding against the trend, what exactly did it go through?
Behind this lies a tumultuous business history.
I. The Years of Being Kicked Out
Intel's decline is a textbook case of path dependency.
In 1999, Intel's market capitalization peaked at over $500 billion, making it the world's highest-valued semiconductor company. In those days, one out of every two PCs globally ran on Intel chips; the "Intel Inside" sticker was plastered throughout the industry.
It wasn't just a chip company; it was more like the infrastructure of the computing era.
Then, the mobile internet arrived.
Relying on its PC dominance, Intel essentially missed the smartphone赛道 (market). ARM architecture swept the mobile端 (end). Apple, Qualcomm, MediaTek. All the winners in mobile chips were not Intel.
It was almost completely defeated in the mobile era but still maintained its composure relying on the PC and server markets.
After the mobile internet, AI experienced a massive explosion.
Starting around 2020, training large models became the core computing need. This requires the parallel computing power of GPUs, with thousands of cores running simultaneously, repeatedly feeding data to a model, adjusting parameters, and iterating optimizations.
NVIDIA's GPUs are天生 (naturally) made for this. Intel's CPUs were彻底 (completely) relegated to a supporting role.
2024 was Intel's most difficult year.
It reported a net loss of $1.6 billion in the second quarter, announced layoffs of over 15,000 employees, accounting for 15% of its workforce. On August 1st after the market closed, the Q2 earnings were released, and the stock price plummeted about 26% the next day, wiping out approximately $32 billion in market value.
The cumulative annual decline was nearly 60%.
With this report card, Intel CEO Pat Gelsinger was "retired" on December 1, 2024.
II. Three CEOs in Five Years, None Could Save It
Gelsinger was no ordinary professional manager.
He joined Intel at 18, worked there for 30 years, rose to become Chief Technology Officer, and was widely recognized as a son of Intel. He left in 2009 to become CEO of VMware, where he performed very well.
In 2021, Intel's board brought him back, pinning their hopes on this veteran to salvage the situation.
His plan was ambitious: plan to invest $200 billion+ in building wafer fabs, transforming Intel from a design company back into a manufacturing company, competing with TSMC for the foundry market, making Intel both a chip designer and a chip manufacturer.
Four years later, Intel's market capitalization evaporated by $150 billion during his tenure.
Gelsinger's problem wasn't a lack of effort. His bet, the wafer fabs, required time, money, yield rates, and customer trust. These were not things that could be built in two or three years.
And the market wouldn't wait.
Before Gelsinger, there was Robert Swan, a finance-oriented CEO, good at managing money, but the chip industry needed more than just financial discipline. Before him was Brian Krzanich, during whose tenure Intel repeatedly fell behind TSMC in process technology and was overtaken by AMD.
Five years, three CEOs, a full cycle, but Intel's core困境 (dilemma) remained unchanged: in the era most needing bets on AI computing power, it bet on the wrong direction.
In March 2025, the ninth CEO, Lip-Bu Tan, walked into Intel headquarters.
III. The Ninth CEO's Three Trump Cards
Lip-Bu Tan was not groomed by Intel.
Born in Malaysia, he moved to the US early on and worked in the semiconductor EDA field for decades. During his tenure as CEO of Cadence Design Systems, the company's revenue more than doubled, and its stock price rose over 3200%.
He had no emotional baggage with Intel, nor did he have Gelsinger's style revenge narrative.
The first thing he did upon arrival was to make the company lean again.
In recent years, Intel had gone too far trying to do everything. Gelsinger's strategy was to simultaneously pursue chip design and wafer foundry. The former requires product innovation, the latter requires manufacturing precision. Trying to兼顾 (manage) both spread resources thin and bloated the organization.
Tan's assessment was to stop the bleeding first. He cut R&D and SG&A expenses by 8% and clearly stated that 2026 would see further reductions. This was no small move; the money saved in Q1 directly showed on the income statement, with non-GAAP net profit surging 156% year-over-year.
The second thing was to redefine Intel's core battlefield.
Tan placed his bets on two areas: Data Center Xeon processors and the foundry business.
The former is Intel's deepest moat; in the global server market, Xeon remains the mainstream CPU, and no company can shake this ecosystem in the short term. The latter is the heavy asset left by Gelsinger, but Tan didn't discard it. Instead, he spent time improving the yield rates.
This quarter, the 18A process yield exceeded expectations, the EUV wafer share in the Intel 3 process increased, foundry revenue reached $5.4 billion, up 16% year-over-year and 20% sequentially, beginning to show an independent growth curve.
The third thing was to actively pursue collaborations others deemed impossible.
NVIDIA's investment, Google's order, Musk's invitation—these three events happened in less than a year, all pointing to the same conclusion: Tan chose to open the doors and welcome allies when Intel was at its weakest.
This was drastically different from the closed strategy of the Gelsinger era. In the past, Intel tried to do everything itself, with poor results in every direction. Tan's logic is to find the few things Intel truly has value in and then have others validate that value.
These three trump cards supported the better-than-expected Q1 2026 performance. Total revenue was $13.6 billion, up 7% YoY; non-GAAP net income was $1.5 billion, a massive 156% increase YoY; non-GAAP gross margin was 41%, steadily improving.
IV. Its New Friends Are All Former Rivals
In September 2025, the entire industry was shaken.
NVIDIA announced it would purchase $5 billion worth of Intel shares at $23.28 per share.
The two companies had been competing for thirty years. Intel's CPUs and NVIDIA's GPUs had long vied for dominance in the computer market. Now, Jensen Huang chose to invest in the old rival and signed a joint development agreement.
Huang's logic was clear: NVIDIA handles training, Intel handles scheduling. The stronger the GPU, the more it needs a powerful CPU to manage it. They are天然互补 (naturally complementary). NVIDIA's investment in Intel was laying the foundation for its own ecosystem.
In December 2025, the US FTC approved the deal.
Just half a year after the NVIDIA investment news fermented, Google followed suit.
On April 9 this year, Google announced an expanded collaboration with Intel, committing to deploy multiple generations of Xeon 6 processors in AI data centers for AI inference and general workloads, and to jointly develop custom ASICs. Intel's stock rose 4.7% that day.
The significance of Google's order far exceeded the procurement itself. Remember, just a few years ago, Google was one of the world's most active companies in developing its own chips; the TPU was Google's product to seize control of computing power from NVIDIA. Now, while大力推进 (vigorously advancing) TPU, Google also explicitly bets on Intel's Xeon CPU.
In AI inference and Agent workloads, both CPU and GPU are indispensable; Google isn't betting on a single side in this issue.
Musk's move was an even more powerful助攻 (assist).
On April 7 this year, Intel announced it would join the Terafab project initiated by Musk, partnering with SpaceX, xAI, and Tesla to build a super chip factory with an annual production capacity of 1 terawatt of computing power. Musk revealed that Terafab would futurely use Intel's 14A process, with Tesla responsible for building the pilot production line and SpaceX负责 (responsible for) mass production.
What Musk is doing is integrating Intel's foundry capabilities into his empire spanning space, automobiles, and AI.
For Intel, the significance of Terafab is far more than just an order. The computing power demand from Musk's system will grow along with xAI, Starlink, and autonomous driving. Intel's position in this supply chain will become increasingly difficult to replace.
Three companies, three motivations, converging into the same conclusion: Intel, at this point in time, has suddenly become indispensable.
V. The Inference Era, the CPU is Back
Lip-Bu Tan said one sentence in the Q1 earnings call worth pondering. He said that as the focus of AI workloads shifts from training to inference, the ratio between CPU and GPU is changing from 1:8 to 1:1, driving a surge in CPU demand.
The first half of AI was training: stacking GPUs, feeding data, repeatedly iterating model parameters. The CPU was almost a marginal role in this stage, responsible for managing memory and I/O, that's it. This was the most painful period for Intel in recent years and the most glorious era for NVIDIA.
The second half of AI is inference and Agents. When models are deployed into enterprises, embedded into products, and called billions of times, the scenario changes completely. Every agent completing a task needs to call multiple tools, switch multiple contexts, and coordinate multiple sub-tasks.
Who does this scheduling work? The CPU.
Alphabet (Google's parent company) CEO Sundar Pichai said at the recently held Google Cloud Next that Google processes 16 billion tokens per minute. Behind these 16 billion tokens, every tool call, every context switch, involves a CPU participating. (Extended reading: Google Doesn't Want to Beat NVIDIA)
In the Agent era, both GPU and CPU are used. The GPU handles inference execution, the CPU handles task orchestration.
This demand is also reflected in the financial report. Intel's First Quarter DCAI (Data Center and AI Group) revenue was $5.1 billion, a sharp increase of 22% YoY. The earnings report disclosed that ASIC revenue grew over 30% sequentially and nearly doubled year-over-year. Xeon 6 secured orders from both Google and NVIDIA, with demand for the full product line far exceeding supply.
Intel lost in the AI training era, an era that required the parallel computing power of GPUs, pushing CPUs aside. It took nearly five years to find its place in the AI era: in inference, in scheduling, in the orchestration hub of Agents.
VI. Out of the ICU, But Not Discharged Yet
But this doesn't mean Intel has suddenly risen again. Hidden dangers on the books remain.
On a GAAP basis, Intel had a net loss of $3.7 billion in Q1, mainly due to $4.07 billion in restructuring charges and Mobileye goodwill impairment. The foundry business, though growing 16%, still needs time to move from loss to profitability. The 18A process yield "exceeded expectations," but there's still a long way to go before mature mass production.
Tan is still cleaning up the mess left by Gelsinger.
Another key question is: Is the CPU demand in the inference era structural or cyclical? There's no definitive answer yet. As AI's Agent capabilities advance further, chip architecture might change again. AMD is catching up, Arm is catching up, and NVIDIA itself is also布局 (laying out plans) in the CPU direction.
Intel has won a window of opportunity, but the window won't stay open forever.
Tan said one sentence in the earnings report: The Intel of today is drastically different from the one I joined just over a year ago. We have returned to our roots, are data-driven, maintain a sense of crisis, and are engineering-centric.
This statement doesn't sound like it's from someone who has won a great victory, but more like someone who has just climbed out of the deepest valley and begun to regain their footing.
Intel struggled for nearly five years, under three CEOs, with over 15,000 people laid off, nearly $200 billion in market value evaporated, removed from the Dow Jones index, and suppressed by competitors from all directions. At its lowest point, the stock price fell back to where it was a decade ago.
Then, the AI inference era quietly began. Its rebound relied not entirely on itself, but also on waiting for a turn in the industry cycle.
Fortunes change, but many companies don't live to see that day.
Words from 【Beyond the Headlines】:
Every major technological wave in history creates winners and victims.
Sometimes, the same technological wave will first knock you down, then pick you up.
Intel was defeated in the training era of AI. Whether it can win in the inference era of AI remains to be seen.
But that nearly 20% surge at least说明 (shows) one thing:
The market believes it still has a fight left in it.
