On June 3, 2026, Marvell Technology's $MRVL stock price touched $290, reaching a new all-time high.
It has risen by 254% in the past 12 months, from less than $40 just three years ago.
If we start counting from when Matt Murphy took over as CEO in 2016—when the stock price was less than $10 and the market cap was not even $20 billion—that's a 30x increase.
However, the price increase itself is not the focus of our discussion in this article.
What this article truly seeks to understand is: What is the market really valuing Marvell at? Are people still using an outdated cognitive framework to understand this company behind this price?
Many on the street refer to Marvell as "Little Broadcom"—the second player in custom AI chips, following in Broadcom's footsteps to pick up what the hyperscalers left behind. While this statement is not entirely wrong, it has a fatal blind spot: It assumes Marvell is a scaled-down version of Broadcom. However, the true value of Marvell lies precisely in the fact that it is nothing like Broadcom.
Marvell occupies a unique niche. The AI infrastructure is transitioning from "piling GPUs" to "building systems," and this position will become increasingly valuable.
This article will attempt to make this clear.
The full article is about 15,000 words long, so please read it carefully~
1. What Exactly Is Marvell Selling?
To understand Marvell, the first step is to discard the label of being a "chip company." It doesn't make GPUs, CPUs, or sell memory. Marvell sells "connections"—enabling data to flow between AI chips, servers, and data centers at the speed of light.
Breaking it down, they have three main business areas:
Section 1: Optical Interconnect—Moat
Marvell is the absolute leader in high-speed optical DSP. In the global 400G and above data center optical module market, approximately 70% of DSP chips come from Marvell.
Every time you see news about a "Surge in AI Data Center Optical Module Shipments," Marvell is quietly raking in money behind the scenes.
Why is this 70% market share so hard to shake? High-speed optical DSP is no ordinary chip. It has to handle signal modulation, demodulation, error correction, clock recovery—all while dealing with the complexities of signal attenuation and noise management at rates like 800G and 1.6T.
After spending around $10 billion to acquire Inphi in 2021, Marvell has accumulated over 5 years of mass production experience in this field, iterating from 5nm to 3nm. Broadcom is also chasing, but this first-mover advantage is not something money alone can erase.
In March 2026, Marvell released four new 1.6T DSP products—Ara T, Ara X, Petra, Aquila M—covering everything from short reach to long reach, from Ethernet to full InfiniBand support. During the FY2027 Q1 earnings call, Murphy made a bold statement: the growth expectation for the optical interconnect business in FY2027 has been raised from 50% to over 70%.
It's not that the market has changed; they underestimated how strong the demand truly was.
Section 2: Custom AI Chips—Expansion
This area receives the most scrutiny from the market. The logic is simple: Amazon doesn't want to pay royalties to NVIDIA for every GPU, so they designed their own AI training chip called Trainium. But Amazon doesn't fabricate chips—so they need help with design and production. That's where Marvell comes in.
Currently, Marvell has 18 custom XPU design projects in hand, covering three hyperscalers—Amazon, Microsoft, Google. The lifetime revenue funnel has reached $75 billion. Custom chip revenue for FY2026 is around $1.5 billion, with expectations to more than double by FY2028.
But there is one uncomfortable point about this business: the gross margin is lower than that of standard products. The FY2027 Q1 non-GAAP gross margin was 58.9%, while Broadcom's was 77.5%. The logic is straightforward—you are working for Amazon, not selling your own standard product, with heavy R&D investment and strong customer bargaining power.
Part Three: Switching Chips and Enterprise Storage—Cash Cow
Ethernet switching chip is expected to exceed $600 million in FY2027 (a doubling year-over-year), benefiting from the rigid demand for high-speed switching when AI clusters scale from hundreds of GPUs to over 10,000. Enterprise SSD and HDD controllers are Marvell's traditional business, contributing to stable cash flow, but their share is shrinking year by year under the pressure of the AI business.
Putting the three parts together, the picture becomes clear: Marvell is not a "chip company that does everything," but a company that has built full-stack connectivity capability around "AI Data Movement." From SerDes inside the chip, to PCIe/CXL switching between chips, to optical DSP between racks, to coherent optical modules between data centers—each link has a hand in it.
Understanding this, it is clear why NVIDIA invested $2 billion in it. It also gradually makes sense why calling Marvell "Little Broadcom" is a misinterpretation.
II. In the AI Era, "Connectivity" Takes the Spotlight
For the past two years, all the spotlight has been on GPUs. Compute power is the muscle, the beefier, the better. However, when AI clusters scale from thousands of GPUs to 10,000s and 100,000s, a problem at the level of physical laws arises: Copper cables can only transmit for 3 meters; beyond this distance, the signal degrades to a point of being unusable.
A GPU can be the most powerful brain in the world, but if the signal transmission between neurons can't keep up, high IQ is useless. In a cluster of 100,000 GPUs, the time each GPU spends "waiting for data" may account for 30%-50% of the total runtime.
This is why optical interconnects have taken the spotlight. Light can travel for hundreds of meters or even kilometers with almost no attenuation. The larger the cluster, the more GPUs, the higher the percentage of optical connections—this is not linear growth, it is superlinear.
Barclays estimates that the shipments of optical ports will double in 2026 and double again in 2027. Marvell's optical interconnect business is expected to grow by approximately 90% in the next two years—wait, it's now revised up to 70%+, but based on the actual numbers, even 90% may be underestimated.
The key point is that this trend is not a short-term phenomenon. As long as the AI model parameters continue to expand and the training and inference clusters continue to grow, the demand for optical interconnect will not flatten out.
This is not a matter of the business cycle; it is a long-term structural trend determined by the laws of physics.
Let's make an analogy: AI infrastructure is a rapidly expanding city, GPUs are the buildings themselves, and Marvell sells the plumbing, wiring, and highways. You can change architects for the buildings, but once the infrastructure is laid out, replacing it is much more challenging than building new houses.
III. From $10 to $290: an Underestimated CEO
In 2016, Marvell was a stock that the market had given up on.
Founders Sehat Sutardja and Weili Dai were forced to step down due to an accounting investigation and governance crisis, with the SEC intervening. The business covered a wide range—mobile communications, printers, consumer electronics—but none of them ranked in the top three in their respective industries. The stock price was below $10, and customers began to worry about the company's survival.
That year, activist hedge fund Starboard Value intervened and led a textbook-level management restructuring. They brought in Matt Murphy from Maxim Integrated to be the CEO.
However, I personally think Murphy deserves more recognition: he spent 22 years at Maxim, moving up from frontline sales to Executive Vice President, overseeing the company's entire product development, sales, and P&L. He is not the kind of "tech genius" semiconductor CEO; he is the kind of extremely practical and focused businessman.
One statement of his left a deep impression on me: "My dad was on Apple's first sales team, and I've known since I was young—no matter how great the technology is, if you can't sell it, it's zero."
After Murphy took office, he did three things that may seem simple but were extremely challenging to execute:
First, cut.
Mobile communications cut. Printer chip business cut. Consumer electronics cut. The Wi-Fi/Bluetooth business was sold to NXP for $17.6 billion (2019). The automotive Ethernet business was sold to Infineon for $25 billion (2025).
All Resources Focused in One Direction: Data Center Infrastructure.
Second, Buy.
In 2018, a $6 billion acquisition of Cavium (the ARM server CPU, DPU). In 2021, a $10 billion acquisition of Inphi, the optical DSP — a deal that changed Marvell's destiny. By the end of 2025, a $32.5 billion acquisition of Celestial AI (silicon photonics/optical fabric-related). In early 2026, a $5.4 billion acquisition of XConn (PCIe/CXL switch chip-related).
Four acquisitions, each filling a piece of the "AI Connectivity" puzzle.
Third, Tie.
Murphy pursues something called "long-term visibility" — years of predictable revenue certainty. Signed a multi-year deal with AWS, covering custom AI chips, optical DSP, AEC DSP, PCIe retimer, DCI optical modules, and Ethernet switch chips — not just a transaction, but a comprehensive system-level collaboration.
As disclosed in the 10-K, certain capacity reservation agreements have terms ranging from 4 to 10 years.
The result? Revenue at the time of takeover in FY2016 was around $2.65 billion (reporting $2.32 billion in FY2017), with thin margins. FY2026 revenue reached $8.2 billion (+42% YoY), non-GAAP EPS of $2.84 (+81% YoY).
In a decade, a second-tier chipmaker emerging from a governance crisis has transformed into a core supplier of AI infrastructure.
There's a rule in investing that I have repeatedly validated: "Change CEO → Strategic Focus → Large Acquisitions → Customer Tie-Up." This complete chain has unfolded, with evidence visible in the financial numbers at every step — such a company is worth dedicating time to study seriously.
IV. NVIDIA's $20 Billion: Endorsement or Acquisition?
On March 31, 2026, NVIDIA announced a $2 billion strategic investment in Marvell—subscribing to 2 million shares of convertible preferred stock, with an initial conversion price of approximately $91.84, representing about 2.4% ownership after full conversion.
On the day of the announcement, Marvell's stock price surged by 13%. However, the market's excitement at that time was slightly different from what I later pondered.
What the market saw was NVIDIA putting down a substantial amount of cash—a stamp of approval saying, "This guy is my endorsed partner." The logic was sound. The $2 billion was not a PR expense but a strategic investment. In 2026, NVIDIA made a series of intensive investments in optical interconnect companies—Coherent ($2 billion), Lumentum ($2 billion), Marvell ($2 billion)—channeling $6 billion into the same track, a signal more powerful than any analyst report.
However, what we should pay more attention to is the collaborative framework—NVLink Fusion.
NVLink Fusion is an NVIDIA-promoted "semi-custom AI infrastructure" platform. Third-party vendors (such as Marvell) can provide custom XPU accelerators that directly connect to NVIDIA's high-speed interconnect network. NVIDIA itself offers the Vera CPU, ConnectX network card, BlueField DPU, NVLink interconnect, and Spectrum-X switch.
In plain terms: "You hyperscalers want to use your own designed custom chips instead of GPUs? No problem, I'll use NVLink Fusion to integrate your chips into my ecosystem. You can have your chips made by whoever, but the connectivity layer remains mine."
It can only be described as extremely clever. Turning the "enemy" into a "customer"—the more hyperscalers want to move away from NVIDIA's GPUs, the more they need NVIDIA's network. And Marvell, precisely the one helping hyperscalers create custom chips and also assisting NVIDIA in building an interconnect ecosystem.
When the snipe and the clam fight, the fisherman wins.
The Next Web had a sharp but accurate analysis headline: "NVIDIA's $2 billion investment in Marvell is not an investment, it's a toll booth."
Through this investment, NVIDIA has set up a barrier at every entrance to the ecosystem. However, looking from a different perspective, Marvell itself is part of that toll booth. They help cloud providers make chips with one hand and help NVIDIA build the network with the other.
Both sides can't live without it, both sides are pouring money into it.
Of course, this also means a never-ending tension: NVIDIA $NVDA is both a partner and a competitor. It makes its own network chips and is also involved in silicon photonics. The boundary between cooperation and competition has always been relatively blurred.
But in my personal opinion: in the AI infrastructure "building system" stage, NVIDIA needs Marvell more than Marvell needs NVIDIA. Because the hyperscaler's custom chip demand is structural and irreversible, NVIDIA's refusal to cooperate is equivalent to handing the entire cake to Broadcom.
5. How Does Marvell Compare to Broadcom?
Many people will quickly come to the conclusion after seeing this table: "Broadcom is better—ten times larger in scale, twenty percentage points higher gross margin, and the valuation is not much higher."
It's not entirely wrong, but two key things have been overlooked.
First, the gross margin gap has a structural reason and does not mean "Marvell cannot make money."
Broadcom's 77.5% is not a pure semiconductor number: it includes VMware's software revenue, with an EBITDA margin of 67%, significantly inflating the combined gross margin.
Looking solely at the semiconductor segment, it's probably around 60%-65%. Marvell's 58.9% is indeed lower, but the difference is not as exaggerated as it seems. Furthermore, with the increasing scale of Custom ASIC mass production and R&D cost dilution, there is a clear path to margin improvement— the company's target mid-term non-GAAP operating margin is 38%.
Second, Marvell is not the "runner-up" in optical DSP, it is the "leader."
With a 70% market share, in this field, Broadcom is the one playing catch-up. Optical DSP is precisely the most beneficial link in the era when AI infrastructure is transitioning from "training-centric" to "inference-centric"— the distribution density of inference clusters is much higher than that of training clusters, requiring a higher demand for optical interconnects.
Duan Yongping has a classification framework that I always find very useful: "Class B business" is something you do better than others, but others are doing it too; "Class A business" is something you do that others simply cannot do, or even if they do, they can't catch up. Marvell's optical DSP is closer to Class A; Custom ASICs are closer to Class B, but the depth of customer binding and switching costs are extremely high, so the quality of Class B is actually not low.
The market evaluates Marvell using the "Mini Broadcom" framework and naturally concludes that it is "not worth $180 billion." However, if we switch to a three-dimensional framework of "Optical Interconnect Leader + Custom Chip Runner-Up + NVIDIA Ecosystem Partner," the valuation logic changes.
Six, Numbers Don't Lie
Financial data is the "sole criterion" to evaluate all narratives. Let's look at Marvell's latest report card and guidance:
Key figures for FY2027 Q1:
· Quarterly revenue of $24.18 billion, YoY +28%, QoQ +9%, reaching a historical high. Exceeds mid-range guidance by $180 million.
· Data center revenue of $18.33 billion, accounting for 76%, YoY +27%, QoQ +11%.
· Non-GAAP EPS of $0.80, meeting expectations. Operating cash flow of $6.39 billion—a historical high as well.
· Q2 guidance: Revenue of around $27 billion (midpoint), +12% QoQ / +35% YoY. Expecting mid- to high-double-digit sequential growth in data centers.
Several trends are highly notable:
The growth rate is accelerating.
FY2026 full-year growth +42%, FY2027 guidance +40%, FY2028 target +45%. The acceleration on an $80 billion base indicates this is not inventory replenishment or a cyclical rebound—it's the structural demand ramp-up.
Operating leverage is being unleashed.
EPS growth rate (81%) far outpaces revenue growth (42%). The scale effect of custom chips, improved yield on optical DSP mass production, and Murphy's strict cost control are all contributing factors.
Custom chips have a "hidden goldmine"—attach. At the 2025 Custom AI Investor Event, the company disclosed an easily overlooked data point: by 2028, the TAM for custom XPU is around $40.8 billion, with the TAM for attach (network card, scale-up fabric, security coprocessor, memory pooling, and other supporting chips around the XPU) at approximately $14.6 billion—this latter part has a compounded annual growth rate of up to 90%. Many people only focus on "who designed the most expensive AI chip," but the real profits lie in the supporting roles.
A rough PEG calculation: Forward PE around 23-24x, revenue growth around 40%, PEG around 0.6.
Comparatively, Broadcom has a Forward PE of around 30-41x, revenue growth of about 20%, PEG around 1.5-2.0. Looking through the simple lens of PEG, Marvell's current valuation is not considered expensive. Of course, PEG is just a rough starting point—whether the growth rate can be sustained, if the gross margin can improve, and whether the competitive landscape will worsen are the real variables.
Seven, The Story of Light: Celestial AI and the Next Journey
Light DSP is Marvell's present, Celestial AI is its future.
In December 2025, Marvell announced the $32.5 billion acquisition of Celestial AI—a startup working on "photonic mesh" technology. $32.5 billion is the initial payment, and if Celestial AI achieves cumulative revenue of $20 billion by FY2029, the total consideration could reach $55 billion.
The price is not cheap. Why is Murphy willing to pay?
Because Celestial AI addresses the next physical bottleneck in AI chip interconnects: the limit of copper cables has been reached.
Currently, within AI servers, GPUs are interconnected using NVLink, which is fast but limited by distance. With one acceleration card containing 8 GPUs, one rack having 4 cards, and a cluster comprising hundreds of racks—the physical limitation of copper cables becomes the bottleneck for the entire system.
Celestial AI's Photonic Fabric replaces electricity with light, enabling chip-to-chip direct optical interconnects—each chiplet has 16 Tbps bandwidth, halving power consumption, with nanosecond-level latency.
In simple terms: Light DSP is expanding the data center's internal highway from two lanes to eight, while Celestial AI is installing teleportation devices directly between each building.
This acquisition has an interesting detail: Amazon supported this transaction.
Marvell even issued a stock warrant to Amazon, allowing Amazon to purchase up to $90 million of Marvell's stock based on the purchase volume of the Photonic Fabric product. Amazon does not casually endorse suppliers' acquisitions—it endorses it because it truly needs this technology.
Marvell expects Celestial AI to start contributing meaningful revenue in the second half of FY2028, with Q4 FY2028 annualized revenue reaching $5 billion and Q4 FY2029 annualized revenue reaching $10 billion. If the roadmap is realized, the photonic interconnect business will transition from the "first-tier business" to the "super business"—a full-stack photonic interconnect platform spanning DSP, silicon photonics, and photonic fabric.
With the addition of the $540 million XConn acquired in early 2026, Marvell now has a complete "Electron + Photon" interconnect puzzle in hand: on-chip SerDes → inter-chip PCIe/CXL switches → intra-rack optical interconnects → inter-rack optical DSPs → inter-data center coherent optical modules.
In the AI interconnect race, no second company has a similarly comprehensive layout.
8. After Rising 254%, Of Course, Don't Forget About the Risks
When researching investments, the most important thing is not to find "why it will rise"—bull markets are filled with reasons. What's important is to find "what will make it fall," and then assess whether you are willing to bear it. I think @aleabitoreddit's risk control concept is very much worth everyone's serious consideration.
Risk One: Loss of Trainium3, Higher Customer Concentration Than You Think
Marvell recently lost the primary design rights for Amazon's next-generation Trainium3, which was taken by Alchip from Taiwan, China. The company emphasizes that Trainium2.5 will continue to be done by Marvell, with "no revenue interruption."
However, what the market sees is another story: the biggest custom chip customer did not choose the old partner for the next-generation product. This is not a good sign.
In FY2026, the top ten customers contributed 82% of revenue, with two accounting for over 10%. The 10-K candidly states, "The current level of capital expenditure on AI infrastructure may not be sustainable in the long term."
If Amazon or Microsoft scale back their custom chip plans, Marvell's revenue will take a direct hit.
Risk Two: Gross Margin Ceiling
Non-GAAP gross margin is 58.9%, nearly 20 percentage points lower than Broadcom. This is not temporary; it is structural. Custom ASICs are essentially a service business—helping customers design exclusive chips, with customers owning the final products, limiting your bargaining power by nature.
The scale effect can improve, but cannot fundamentally resolve.
If future revenue growth is mainly driven by custom ASICs (low margin) rather than by silicon photonics (high margin), there may be a trade-off between revenue growth and margin expansion. The market's valuation multiple may not be as generous as the bulls expect.
Risk Three: NVIDIA's "Toll Booth" Could Turn Into "Toll Booth + Competitor"
NVIDIA has invested $2 billion, but has also built its own network chip team. Technologies like the Spectrum-X switch, BlueField DPU, and NVLink interconnect pose direct or potential competition to Marvell's switch chips and custom ASICs. A 2.4% shareholding is not control but ecosystem alignment.
If NVIDIA decides in the future to internalize more value chain segments of NVLink Fusion — such as producing more silicon photonics chips internally — Marvell's position could become delicate.
Risk Four: Insider Selling
Since 2026, CEO Murphy has sold approximately $5.3 million (three sales, prices ranging from $98.70 to $177.26), CFO Willem Meintjes has sold about $4.7 million, COO Chris Koopmans has sold about $2.73 million, and CDO Sandeep Bharathi has sold about $13.14 million.
No insider has made any purchases.
The absolute amounts relative to shareholding are not significant (even after Murphy's sales, he still holds about $131 million in stock), and all sales are being executed through a pre-designed 10b5-1 plan.
However, the signal is clear: as the stock price hits an all-time high, a group of people most familiar with the company are selling, with no one buying.
At the very least, this should make you question yourself once more: Am I buying this stock because I understand its value, or because I see it has risen by 254%?
Risk Five: Supply Chain
10-K disclosures require capacity to be secured 26-52 weeks in advance, with some agreements lasting 4 to 10 years.
TSMC's 5nm/3nm capacity is in fierce competition with GPU manufacturers (such as NVIDIA, AMD), and the lead time for DSP delivery has been extended to 6 months.
Marvell has identified a demand miscalculation—either overcommitting capacity resulting in declining demand, or facing higher-than-expected demand but insufficient capacity—both scenarios will directly impact the financial report.
In this supercycle of AI infrastructure, rewards are not only given to the "technologically savvy," but also to those who maintain a flawless supply chain.
After discussing all these risks, what is my conclusion?
All these risks are real.
The loss of Trainium3 is significant, resolving the structural gross margin issue will take time, and insider selling is concerning. However, I have not taken a bearish stance on Marvell for three reasons:
First, the bad news of losing Trainium3 has been covered by the guidance of doubling the revenue from FY2028 custom chips.
If losing the largest customer's next-generation product can still lead to a revenue doubling guidance, it indicates that the pipeline of other customers (Microsoft Maia, Google Axion, and that "undisclosed new hyperscaler") is stronger than the market perceives.
Second, the moat of optical interconnect is real and expanding.
70% DSP market share + Celestial AI silicon photonics + XConn PCIe/CXL switch = a unique end-to-end capability that others cannot replicate quickly.
Competitors may snatch a few custom chip orders, but no one can catch up with Marvell's accumulated expertise in optical interconnect within three to five years.
Third, a PEG of 0.6 provides a certain margin of safety.
A 40% revenue growth corresponds to a 23x forward P/E ratio—this valuation does not suggest "the market already views it as the next Broadcom in hype," but rather "the market is still pondering whether it is worth less than Broadcom."
9. Some Thoughts on the Era
Peter Thiel raised a point in "Zero to One" that made many entrepreneurs uncomfortable: Competition is for losers; true companies create monopolies.
“All failed companies are the same — they failed to escape competition.”
Applied to investing, this framework forces you to ask a sharp question: Is the company you are researching struggling in a competitive market, or does it hold a monopoly in a market it defines?
Marvell's interesting aspect is that it is doing both at the same time.
In the custom AI chip field, it is a follower of Broadcom — a "participant."
In the optical interconnect and high-speed DSP field, it is an absolute market leader — a "monopolist."
NVIDIA's $2 billion investment essentially confirms Marvell's monopoly value in the "connection" dimension with real money and silver.
Thiel says one characteristic of a monopoly is that the market is smaller than it seems — “Monopolies usually disguise their monopoly position to avoid regulatory attention.”
Marvell is quite the opposite: Its monopoly position has been overlooked by the market because everyone is focusing on its gap with Broadcom in custom chips.
Do a thought experiment: Marvell's current market value is about $250 billion, corresponding to around $11.5 billion in FY2027 revenue, approximately 21.7x price-to-sales ratio. However, the data center segment of the $11.5 billion is about $9.2 billion, growing at a rate of over 50%.
Valuing only based on Broadcom's valuation multiples (about 25-30x price-to-sales ratio), this segment alone is worth $230-276 billion.
The market value of $250 billion essentially discounts the data center business and gives away the other businesses.
Of course, this “segment valuation” is too rough — Marvell's data center business will not really receive Broadcom's multiples, its gross margin structure is different, customer concentration is higher, and its position in custom chips is weaker than Broadcom.
But it at least provides a starting point for reflection: the market's pricing of Marvell likely remains stuck in the old narrative of "this is the company that lost Trainium3," rather than the new reality of "this is the only company globally with meaningful revenue in all three battlegrounds of optical DSP, silicon photonics, and custom AI chips."
My assessment could be wrong as well. The custom chip competition may be fiercer than I anticipated, the growth in optical interconnect demand may not be as optimistic as the models predict, and Celestial AI's $1 billion annualized revenue target may not materialize.
But as of now, I am willing to bet on the "AI Connectivity" direction. Not because Marvell is the best company, but because it is in the rightest position.
Epilogue: Light and Civilization
At this point, I want to step out of the investment framework and say a few words on a grander scale.
Every leap in human civilization, in retrospect, has not been due to a singular breakthrough but because of an upgrade in "connections."
Writing allowed ideas to transcend time, the printing press enabled knowledge to move across social strata, the telegraph allowed information to cross oceans, and the Internet interconnected the global brain for the first time.
Each time, what truly changed the world was not the "content" itself but the speed and scale of content flow.
The AI era is reenacting the same story.
We have focused too much on the "brain" — bigger models, stronger computing power, smarter reasoning. But the brain has never existed in isolation.
No matter how high one's intelligence is, if they cannot communicate with others, they are just an isolated island. The same is true for a cluster of 100,000 GPUs — if data cannot flow freely between them, no matter how powerful the computing capacity, it is just silent silicon.
Light is the messenger of this era.
From a physics perspective, light is the speed limit for information transmission in the universe. It took us thousands of years to learn to harness it — from beacons to fiber optics, from Morse code to 1.6T DSP signal processing. Now, as humanity attempts to build a true "silicon-based brain" for the first time, we are once again faced with the same ancient question: How do we enable thoughts, whether carbon-based or silicon-based, to flow at the speed of light?
The Marvell story is, on the surface, a decade-long turnaround story of a chip company. But digging deeper, it touches on a more fundamental proposition: in any complex system, the value of "connection" will eventually surpass the value of "nodes."
In the Internet age, the total value of routers and fiber optics eventually exceeded that of any single server.
In the social media age, the value of the platform surpassed that of any single content creator.
In the AI age, the same logic is replaying—while everyone is vying for the title of the "most powerful brain," the true winner may be the one quietly weaving neural networks.
We are standing at a magical yet subtle moment in history. For the first time, humans have the ability to build something smarter than themselves, and whether this entity can truly become "intelligent" depends on whether we can solve what seems like a mundane engineering problem: allowing light to freely shuttle between chips.
In itself, this carries a kind of poetic beauty.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
