Nvidia 13F Reveals a New Battlefield: AI Trading Shifts from Buying GPUs to Acquiring “Bottleneck Assets”
Show original
By:美股研究社
Bitget offers one-stop trading for cryptocurrencies, stocks, and gold. Trade now!
A welcome pack worth 6200 USDT for new users! Sign up now!
Source | U.S. Stock Research Agency
After Nvidia’s latest 13F disclosure, the market easily treats it as a “Jensen Huang portfolio” to copy.
On May 15, Nvidia disclosed its Q1 2026 13F holdings. 13F reports are inherently delayed, only reflecting selected public securities holdings as of March 31. They should not be equated with real-time trading instructions, nor should they simply be understood as “Nvidia buy recommendations.”
However, the portfolio structure is quite interesting: Intel remains the largest position, CoreWeave saw a significant increase, Coherent and Generate Biomedicines appeared as new entries, and Synopsys, Nokia, Nebius, etc. remain. Public data show Nvidia raised its Q1 CoreWeave position to approximately 47.21 million shares, newly added about 7.8 million shares of Coherent and about 833,000 shares of Generate Biomedicines. Nvidia’s official website shows this 13F was submitted on May 15, 2026.
Here, CoreWeave stands for AI computing power operations, Coherent for optical interconnect and the bandwidth wall, Generate Biomedicines for AI spillover into high-value sectors like biopharma. Taken together with Nvidia’s $2B investment in both Coherent and Lumentum this March, and its expanded partnership with Corning for optical connectivity in May, the answer is clear: AI investment is shifting from “who owns the GPU” to “who can deliver GPUs faster, connect them with lower latency, and maximize actual computational efficiency.”
For investors, this 13F’s real market translation is: The AI rally hasn’t left Nvidia, but excess returns are migrating toward the areas where Nvidia is currently shoring up its weaknesses.
The GPU dividend hasn’t ended,
Nvidia is extending control into the AI factory delivery chain
The market’s main AI thesis is clear: buy Nvidia, buy GPU, buy HBM, buy TSMC, buy CoWoS, buy servers. The bet: LLM training and inference are exploding, global computing power is scarce, and Nvidia is the most critical gatekeeper.
This logic hasn’t ended. The problem is, it’s no longer fresh.
Now that Nvidia’s market cap is in the trillions, data center revenue is consensus, and keywords like Blackwell, Rubin, HBM, and CoWoS have been repeatedly traded, simply saying “GPU demand is strong” no longer drives new expectations. The next capital hunt is not about “is AI still strong,” but which parts of the AI value chain most likely form bottlenecks.
Nvidia’s 13F offers a very clear answer: it is shifting from a chip company to the system orchestrator for AI factories.
Intel is the top holding, representing advanced manufacturing, packaging, and U.S. domestic semiconductor capacity; Synopsys stands for EDA and chip design tools; Nokia for network infrastructure; CoreWeave and Nebius for AI cloud and compute operations; Coherent for optical communications; Generate Biomedicines for high value compute use cases. The commonality of this portfolio isn’t “hot sectors,” but critical nodes in the AI infrastructure chain.
This is Nvidia’s most meaningful transformation: it’s no longer just concerned about selling GPUs, but now cares whether GPUs can be deployed, networked, accessed by clients, and ultimately, if they generate real revenue in industry applications.
Why? Because AI infrastructure is no longer a simple commodities business, but a super heavy-asset, ultra-complex delivery system. GPUs are just the entry point. Behind them are racks, power, cooling, networking, optical modules, fiber, data center siting, power contracts, cloud scheduling, software stacks, and customer commitments. Any weak link turns GPUs from “scarce assets” into “inventory on paper.”
This is why Nvidia supports CoreWeave—not just because it’s a client; and why it invests in Coherent—not just for the optical comms stock price. The bigger logic is: Nvidia must make the delivery chain for AI factories more controllable.
This transformation is critical for capital markets.
In the first phase of AI trading, the core asset is the GPU.
In the second phase, the core asset is “infrastructure that maximizes GPU efficiency.”
In the third phase, the core asset will be “applications that continuously absorb compute power and generate high value.”
Nvidia’s 13F covers all three layers.
CoreWeave is the delivery channel for stage two, Coherent the physical link for stage two, Generate Biomedicines the demand outlet for stage three. They’re not “Nvidia shadow stocks” in the traditional sense; they are the strategic nodes Nvidia must preempt to keep the AI growth curve alive.
If investors only see that “Nvidia bought CoreWeave and Coherent,” it’s easy to turn this into a short-term theme. But a more accurate interpretation: Nvidia is telling the market that AI infrastructure bottlenecks are spreading from chip supply to compute operations, optical interconnect, power, fiber, and vertical applications.
This will drive changes in valuation anchors.
In the past, valuations for supply chain companies looked at order visibility, shipments, and margins. Now, there’s a new indicator: are they in Nvidia’s AI factory system? As long as that system grows, they’re not just suppliers, but prerequisites for AI infrastructure expansion.
This is what makes the 13F truly valuable.
Its function isn’t for you to copy its holdings, but to see where the next round of AI capital might seek “bottlenecks not fully priced in.”
CoreWeave and Coherent
are the two most tangible walls in AI infrastructure
CoreWeave is the name in this 13F most likely to be traded by the market.
Its story flows smoothly: AI cloud, GPU clusters, compute shortages, Nvidia ecosystem, client growth. CoreWeave booked $2.078B in Q1 revenue, up from $982M YoY; revenue backlog reached $99.4B; the company said it has surpassed 1GW of active power capacity and expanded total contracted power to over 3.5GW. This shows it’s no longer a “small-time GPU rental shop” but arguably the most representative “new cloud vendor” among AI compute operators.
But CoreWeave’s financials also expose another side: a Q1 net loss of $740M, $536M in interest expense, and an operating loss of $144M. In other words, the company’s growth is fast, it has many orders, but it’s a classic high-leverage, high-capex, high-depreciation model. It’s not a lightweight cloud software company—it’s more like an AI power plant under endless construction.
This is what makes it most controversial—and most valuable.
Why do we need CoreWeave when there’s Microsoft, Amazon, Google? Because AI cloud demand is so specific. LLM companies need rapid delivery, elastic scale, GPU cluster optimization, network topology, storage, and scheduling. Sure, the leading general cloud providers can offer this, but they juggle many business lines and their capex is limited by free cash flow. CoreWeave, as a vertical AI cloud provider, focuses all its resources on AI compute, yielding efficiency but concentrating risk.
Nvidia’s increased stake in CoreWeave is essentially supporting a highly efficient GPU distribution channel.
This is vital to Nvidia. No matter how strong GPU demand is, if compute gets stuck in cloud vendor purchasing cycles and prolonged data center delivery times, both revenue recognition and the ecosystem expansion suffer. Companies like CoreWeave help Nvidia convert chips faster into rental, callable, and billable compute.
However, there’s a risk the market is already discussing: “circular financing.”
Reuters recently reported that Jensen Huang’s family foundation bought $108.3M worth of AI compute from CoreWeave and donated it to research institutions. Also highlighted was Nvidia’s deep financial relationship with CoreWeave, including a $2B investment and a $6.3B agreement involving unused cloud capacity. Investors have raised concerns about the potential for circular financing.
This is not a trivial issue.
If Nvidia is both supplier and investor, and supports client order expansion through its ecosystem, the market will inevitably question: How high is CoreWeave’s revenue quality? How concentrated is its customer base? Can future cash flows cover its debt and interest burdens? If AI cloud demand slows, will CoreWeave be the first high leverage asset to take a hit?
Therefore, CoreWeave is not a “certainty for price appreciation,” but a high-beta representation of the AI compute operations path. Its capital story is compelling, but its financial structure is heavy. Short-term focus: orders and Nvidia ties; medium-term: utilization and interest coverage; long-term: can it transition from training cloud to inference cloud for stabile cash flow.
Coherent, in contrast, represents another wall: the bandwidth wall.
In March, Nvidia announced a multi-year strategic pact with Coherent, investing $2B to support R&D, capacity, and operations, and gaining access to advanced laser and optical network products. Nvidia made it clear: optical interconnect and advanced packaging will be foundational for the next generation of AI infrastructure because they enable ultra-high bandwidth, low-power connections in AI factories.
This marks a major shift in industry logic.
Previously, the market focused on AI servers mainly by GPU count. Now, as GPU clusters grow, training and inference require massive data movement among chips, servers, racks, and data centers. Fast computation is just the beginning; slow transfer drags down efficiency. Copper’s limits in distance, power, and bandwidth have made optical modules, lasers, silicon photonics, CPO, and fiber transition from side components to core AI factory assets.
This also explains why Nvidia announced a $2B investment in Lumentum on the same day, and in May sealed a long-term deal with Corning. Corning pledged to expand U.S. optical connectivity manufacturing by 10x, boost domestic optical fiber capacity by over 50%, and build three more factories. Nvidia’s official release stated: Modern AI workloads need tens of thousands of GPUs, requiring unprecedented levels of high-performance fiber, connectivity, and photonic technology for data movement.
This theme is worth emphasizing.
Because it shows the AI infrastructure bottleneck is shifting from “not enough chips” to “not enough connectivity.” Optical communications used to be seen as cyclical hardware, but now it’s being repriced as AI factory’s physical foundation. Companies like Coherent, Lumentum, Corning are no longer just Nvidia supply chain periphery, but are prerequisites for massive AI cluster expansion.
This is also the sector with the most potential for surprise in the AI industrial chain over the next year.
GPUs are well understood by the market, HBM has been traded over many cycles, and the value of CoWoS and advanced packaging is increasingly clear. But fiber, modules, lasers, CPO, and optical interconnects are still, for many, just “sympathy plays.” Nvidia continues to validate the opposite with investments, procurement commitments, and capacity lock-ups: the larger the AI factory, the less optional optical connectivity becomes.
In my view, CoreWeave and Coherent represent two sides of AI infrastructure value: one solves “how compute is delivered,” the other “how compute collaborates.” The former is cloud operating efficiency, the latter is physical network efficiency. They are different types of assets but both are coming under Nvidia’s next management radius.
GENB:
Ultimately, high-value industries must pay for compute power
Generate Biomedicines holds a relatively small position in this 13F, easily missed by the market.
This company went public on Nasdaq in February, raising $400M at $16/share. Reuters reports Generate Biomedicines is backed by Flagship Pioneering and uses AI-driven tech to develop protein therapeutics, focusing on immunology and oncology. The pipeline includes GB-0895 for severe asthma, expecting full enrollment in early 2028.
Nvidia buying GENB doesn’t mean it’s becoming a biopharma fund. Rather, Nvidia is watching to see where AI compute will generate the highest value returns.
This is a key question.
The entire market is now discussing AI capex: Microsoft, Amazon, Google, and Meta spending hundreds of billions on data centers, Nvidia selling GPUs, TSMC expanding advanced nodes, SK Hynix and Micron expanding HBM, Coherent and Corning expanding optical links. Someone must eventually foot the bill. In the short run, it’s LLM training and inference; in the mid-term, it’s enterprise agents, AI search, AI office, AI customer service, AI coding. In the long run, fields that can truly consume massive compute are biopharma, materials science, robotics, industrial simulation, and autonomous driving—high value industries.
AI pharma is a typical example.
It is less likely than AI chat apps to scale quickly, and its revenue is not as transparent as AI cloud, but, if successful, it could change the cost curve for drug discovery. Biopharma is by nature a high-data, high-risk, long-cycle industry. Traditional drug R&D goes through target discovery, candidate screening, preclinical and three clinical phases—failure eats massive time and capital at any stage. If AI can improve protein design, candidate filtering, and preclinical validation, it’s not merely boosting office efficiency—it’s rewriting a high-margin industry’s R&D function.
But this is also the area most easily misunderstood by the market.
AI pharma is not like optical modules—sign contract today, ramp up tomorrow, recognize revenue the next day. Its key is not how many proteins the model generates, but whether candidates can enter trials, pass safety/efficacy, secure regulatory approval, and ultimately commercialize. Capital can speculate on the theme, but true value realization must pass through clinical data.
Thus, GENB is more of a “signal for future demand” than a short-term trading clue.
Nvidia’s challenge today is not selling GPUs, but proving this round of AI capex won’t become next cycle’s glut. That proof can’t rely on model companies burning cash or cloud vendors hiking capex alone. AI has to penetrate industries that can deliver real returns. Pharma, materials, robotics, and industrial simulation are such directions.
GENB’s small position is a wake-up call: the last mile for the AI value chain is not in server rooms, but on the real-world P&L statement.
If AI remains stuck at the infrastructure layer, it will eventually have to answer for its return on capital. Only when AI enters drug discovery, industrial design, energy optimization, robotics, and enterprise decision-making does compute demand move from “training fever” to “industrial necessity.”
This is the outcome Nvidia most wants to see.
Because Nvidia sells more than GPUs—it sells “intelligent means of production.” The value of the means of production depends on whether downstream users can create more revenue from them. The significance of AI pharma is that it pushes compute from the internet and cloud computing into vast physical industries and scientific research.
So what’s truly worth tracking isn’t the codes in the 13F, but the questions Nvidia poses with its capital and orders:
After GPUs, what is the AI factory most lacking?
Who can bring compute online more quickly?
Who can transfer data faster?
Who can get this compute power into higher-value industries?
These three questions could be the roadmap for the next phase of AI investment.
0
0
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
Understand the market, then trade.
Bitget offers one-stop trading for cryptocurrencies, stocks, and gold.
Trade now！
You may also like
VOOI fluctuates 800.7% within 24 hours: Extreme price swings in low liquidity markets
Bitget Pulse•2026/05/18 15:19
BGSC fluctuates 89.3% in 24 hours: Token burn effect drives severe price volatility
Bitget Pulse•2026/05/18 14:12
HIVE stock rallies 35% after unveiling 320MW 'AI gigafactory' project in Toronto
The Block•2026/05/18 14:09
Trending news
MoreCrypto prices
More
Bitcoin
BTC
$76,052.24
-2.45%
Ethereum
ETH
$2,097.52
-3.94%
Tether USDt
USDT
$0.9994
-0.02%
BNB
BNB
$634.87
-2.57%
XRP
XRP
$1.37
-2.78%
USDC
USDC
$0.9998
-0.00%
Solana
SOL
$83.7
-2.98%
TRON
TRX
$0.3542
-0.87%
Dogecoin
DOGE
$0.1037
-6.07%
Hyperliquid
HYPE
$44.61
+2.07%
How to buy BTC
Bitget lists BTC – Buy or sell BTC quickly on Bitget!
Trade now
Become a trader now?A welcome pack worth 6200 USDT for new users!
Sign up now