Nvidia Bets on Local AI and Denser Racks While a $20 Billion Groq Deal Draws Fire
Published on 10/07/2026 at 15:01 | Editorial boerse-global.de
Nvidia used a single day to push in two directions at once: onto the desktop and deeper into the data center. At a San Francisco event, chief executive Jensen Huang, joined by Microsoft's Satya Nadella, unveiled the Surface Laptop Ultra, a machine built around Nvidia's RTX Spark technology. The pitch is straightforward — run complex AI agents on the device itself rather than tying up permanent capacity in Microsoft's Azure cloud.
Shifting that workload to local hardware is meant to trim the running costs of cloud infrastructure and cut response times for users. In its top configuration, RTX Spark delivers one petaflop at FP4 precision and up to 128 gigabytes of shared memory, enough to run speech and coding models with as many as 120 billion parameters without a cloud connection, according to Microsoft. The laptop carries a 15-inch mini-LED display and is billed as lasting a full day on a charge despite the silicon inside.
The economics are less settled. Memory components are a major cost driver, and market watchers see the resulting sticker prices above $2,000 as a real obstacle to rapid mass adoption of local AI hardware.
Racks, tokens and a legal headache
At a San Jose gathering the same day, Nvidia rolled out an expanded data center architecture. Its new LPX rack pairs the Groq 3 LPU compute unit with Nvidia's Rubin graphics processors. The company says the system generates more tokens per second while drawing less power, with the Bluefield 4 STX and Spectrum-6-SPX networking systems rounding out the lineup.
The deal underpinning that hardware has landed Nvidia in court. Former Groq engineers have sued over the transaction, valued at more than $20 billion, alleging that those in charge sold assets below value and skipped a standard valuation process. Of the total, $17 billion covered technology licenses and $3 billion took the form of Nvidia stock for transferring employees. Groq has called the claims unfounded.
Should investors sell immediately? Or is it worth buying Nvidia?
Morgan Stanley: grid strain is a 2027 problem, not an Nvidia problem
Investors weighing the stock face a shifting backdrop. For several quarters, raw demand for AI graphics processors drove the share price; now a more fundamental constraint is moving into view — the load limits of the U.S. power grid. Morgan Stanley weighed in on Monday, arguing that Nvidia is relatively well insulated from data center power shortages compared with other market participants, and that these infrastructure hurdles do not put its 2027 forecasts at risk.
That confidence lands in a stretch when major customers keep mobilizing enormous sums for hardware. Reuters, citing the Financial Times, reported yesterday that SpaceX is seeking $40 billion in financing earmarked for buying Nvidia chips — $10 billion in bank loans and $30 billion in bonds, with Apollo Global Management leading the transaction.
The two threads together capture the market's mood: the tech industry's appetite to spend is unbroken, while pressure builds on operators to actually power the processors they have ordered. The question that matters for the share price is not order intake but how fast data center operators can bring new grid connections online. Without reliable feeders and substations, even the most advanced accelerator cards cannot be switched on. Should grid buildout delays mount, customers may push delivery dates back, and Nvidia would book revenue later than planned. Valuation models have assumed seamless scaling of global compute; available megawatts are becoming the primary pace-setter for when future revenue is actually realized.
Buybacks as ballast
In the bullish case, the wave of multibillion-dollar hardware investment simply continues. The willingness of large customers to take on substantial debt to buy Nvidia processors underscores how strategically important AI infrastructure has become. As long as companies press ahead with the buildout, the order book for coming years stays intact.
Management adds stability of its own through decisive capital allocation. In late September, Nvidia authorized a $150 billion increase to its existing share buyback program, leaving $235 billion available to be deployed through fiscal 2028. A repurchase of that size underpins the stock and counters dilution, while demonstrating that the company can return large amounts of free cash to shareholders even as it funds enormous research spending.
Where the thesis could crack
The counterargument is tangible. Physical limits on energy infrastructure cannot be accelerated at will. In many U.S. regions, lead times for specialty transformers run to several years, and permitting for high-voltage lines often drags. If grid capacity noticeably slows data center expansion, growth rates risk being compressed. Hardware that cannot be plugged in is either not shipped or ties up dead capital at the buyer. Add the financing risk on the customer side: a high share of debt raises vulnerability to delays, and if expected returns fail to materialize on time, planned follow-on orders could be cut or scaled back.
Nvidia at a turning point? This analysis reveals what investors need to know now.
As long as demand holds up as it has in recent initiatives and power supply does not strangle deliveries through 2027, the upward move stays intact, with the chipmaker's market position and its scheduled buybacks providing a solid foundation. But if sentiment among infrastructure operators turns and grid bottlenecks force whole project phases to slip, investors should brace for a meaningful correction — rich valuation multiples leave little room for delays in revenue recognition.
The next concrete catalyst is progress on SpaceX's planned $40 billion financing; a successful placement of the loans and bonds would confirm that liquidity in the tech sector remains abundant. Meanwhile, the gradual execution of the authorized $235 billion buyback through fiscal 2028 serves as a lasting gauge of the company's financial strength.
The tape
The market has so far shrugged off the legal dispute. The stock traded at EUR 212.05 on the day, a modest decline of 0.2%, keeping it within touching distance of its records — just 1.9% below its 52-week high. In premarket trading the shares changed hands at EUR 213.70, leaving only a sliver to the 52-week peak of EUR 216.10. What investors are rewarding is the persistent demand for accelerator cards and Nvidia's push to open new hardware markets beyond traditional hyperscale data centers.
Ad
Nvidia Stock: New Analysis - 7 October
Fresh Nvidia information released. What's the impact for investors? Our latest independent report examines recent figures and market trends.
