Synthszr Charts — die großen AI-Marken im Wettkampf ums Podium

RoboTTT

Nvidia · od 15. Juli 2026 (Paper/Projektseite veröffentlicht) · 2× · naposledy 15. 9. 2026

Momentum

RoboTTT is a research framework from NVIDIA (with Stanford and UT Austin) that integrates Test-Time Training (TTT) into vision-language-action robot policies, scaling visuomotor context to 8,000 timesteps — three orders of magnitude beyond prior state-of-the-art policies — without increasing inference latency. It is instantiated on the pretrained GR00T N1.7 model (3-billion parameters) by adding TTT layers to its Diffusion Transformer action head. On challenging real-robot manipulation tasks (bimanual assembly on a YAM setup), RoboTTT improves overall performance by 87% over a single-step baseline and is the only model to fully complete a five-minute, ten-stage assembly task. It is a research project with a published paper and project page, not a commercial product with pricing.

Vývoj momenta

18.06.16.09.

Vlastnosti

Deployment Model	Research prototype with published paper/project page and demo videos; no commercial product deployment documented
Use Case Scope	Bimanual robot manipulation/assembly (e.g. Circuit Assembly, Pup Go Car, Gear Bot) on YAM robot platform with 4 RGB cameras
Integrations	Instantiated on GR00T N1.7; training uses sequence action forcing + truncated backpropagation through time on 16 NVIDIA GB200 GPUs
Platform	Built on NVIDIA GR00T N1.7 (Vision-Language-Action model, 3B parameters), TTT layers added to its 16 DiT layers in the Diffusion Transformer action head
Release Date	July 15, 2026 (arXiv paper and project page)

Další produkty v této kategorii: Robotika a Embodied AI

Zdroje (2)

Company Analysis: Nvidia

As of 12/09/2026

HOLDSynthszr Vote

Data-driven balance: NVIDIA’s latest reported fundamentals remain exceptional (Q2 FY2027 revenue $96.2B, +106% YoY; ~75% gross margin) and capital returns are substantial (~$26B in Q2). However, management is guiding a near-term gross margin step-down (Q3 ~74%; Q4 trough ~71–72%) driven by memory cost inflation, which raises the probability of near-term earnings volatility and valuation multiple sensitivity. With current price in EUR not provided and valuation still premium on common metrics, HOLD is the most disciplined stance until there is clearer evidence that (1) margin trough is contained as guided and (2) pricing actions and supply conditions support the FY2028 margin recovery path. (nvidianews.nvidia.com)

Summary

NVIDIA is a full-stack accelerated-computing company whose business model combines (1) high-performance GPUs and networking (InfiniBand/Ethernet), (2) tightly integrated systems and reference architectures (e.g., NVL racks), and (3) a large software platform led by CUDA and an expanding set of AI/industry libraries. This integration allows NVIDIA to monetize not only silicon, but also system-level value (performance per watt, time-to-deploy, and developer productivity), which is increasingly critical as customers build “AI factories” constrained by power, memory availability, and deployment speed. (nvidianews.nvidia.com) Market position remains dominant in AI training and increasingly in inference, supported by a broad ecosystem (developers, model builders, cloud partners, OEMs) and rapid product cadence (Blackwell as the current revenue majority; Vera Rubin referenced as a next platform). Competitive advantages are primarily ecosystem lock-in (CUDA + libraries), scale of platform adoption across hyperscalers and enterprises, and the ability to deliver end-to-end infrastructure rather than discrete components. (investor.nvidia.com) In the most recent quarter (Q2 FY2027, ended July 26, 2026), NVIDIA reported revenue of $96.2B (+18% QoQ, +106% YoY) and GAAP EPS of $2.46 (non-GAAP EPS $2.22). Gross margin was ~75% (GAAP and non-GAAP). Management reiterated that its outlook assumes no Data Center compute revenue from China, highlighting a deliberate de-risking of forward guidance from that geography. (nvidianews.nvidia.com) Key recent developments include: (1) a large capital return program—about $26B returned in Q2 FY2027 (buybacks + dividends), with substantial remaining repurchase authorization; and (2) margin guidance pressure driven by memory cost inflation. Management guided Q3 gross margin around 74% and expects a trough in Q4 around 71–72%, then a recovery toward ~72–73% in FY2028 as price increases take effect. (sec.gov) Valuation metrics vary by data provider, but broadly indicate a premium multiple versus the market; recent snapshots show trailing P/E in the high-20s and forward P/E in the mid-20s on major finance aggregators. Investors should frame near-term performance around (a) supply-driven shipment capacity for leading systems, (b) customer capex pacing and data-center build constraints (land/power/shell/capital), and (c) the temporary margin compression from memory costs. Short- to medium-term outlook is supported by strong demand signals and company guidance, but with tighter sensitivity to component costs and deployment bottlenecks than in prior cycles. (ca.finance.yahoo.com)

Key Takeaways

Q2 FY2027 results showed continued hyper-growth: $96.2B revenue (+106% YoY) with ~75% gross margin, reinforcing NVIDIA’s scale advantage in AI infrastructure. (nvidianews.nvidia.com)
Management’s forward outlook explicitly assumes no China Data Center compute revenue, reducing guidance dependence on a volatile regulatory/geopolitical variable. (nvidianews.nvidia.com)
Gross margin is guided to step down (Q3 ~74%; Q4 trough ~71–72%) due to memory cost inflation, with an expected partial recovery in FY2028 as pricing actions flow through. (investor.nvidia.com)
Capital return is now material: ~$26B returned in Q2 FY2027 (buybacks + dividends) and significant remaining repurchase authorization, supporting per-share metrics. (sec.gov)
Execution risk is shifting from “demand creation” to “delivery economics”: supply availability, memory scarcity, and data-center build constraints (power/land/capex) are key swing factors for revenue timing and margins. (sec.gov)

Action Ideas

BUY

For investors with a 12–24 month horizon, NVIDIA remains a high-conviction AI infrastructure compounder based on (1) sustained revenue acceleration in the latest reported quarter, (2) platform-level competitive advantages (hardware + networking + software), and (3) explicit management commentary that demand remains strong even as guidance excludes China Data Center compute. The near-term margin dip is framed by management as cost-driven (memory) with planned pricing offsets, suggesting earnings power can normalize if execution matches guidance. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027?utm_source=openai))

Horizon: 18 mo.

HOLD

For investors already positioned, a HOLD stance is supported by strong fundamentals and shareholder returns, but balanced by near-term margin compression and elevated valuation multiples. Maintain exposure while monitoring (a) quarterly gross margin progression versus the guided Q4 trough and FY2028 recovery band, and (b) evidence that supply and deployment constraints are easing. ([investor.nvidia.com](https://investor.nvidia.com/files/content_files/TRANSCRIPT_-NVIDIA-Corp-NVDA-US-Q2-2027-Earnings-Call-26-August-2026-5_00-PM-ET.pdf?utm_source=openai))

Horizon: 9 mo.

SELL

For risk-managed portfolios with short horizons (3–6 months) or strict margin/valuation discipline, consider trimming/selling into strength if the investment case depends on continued margin expansion in the next two quarters. Management is guiding a gross margin step-down (Q3 ~74% and Q4 ~71–72%) driven by memory costs, which can pressure near-term earnings revisions and increase volatility around prints. ([investor.nvidia.com](https://investor.nvidia.com/files/content_files/TRANSCRIPT_-NVIDIA-Corp-NVDA-US-Q2-2027-Earnings-Call-26-August-2026-5_00-PM-ET.pdf?utm_source=openai))

Horizon: 6 mo.

Contrarian Insights

• Margin compression may be more “structural for a period” than the market assumes: management explicitly attributes the guided trough to memory scarcity/cost inflation and indicates a multi-quarter path (Q3 down, Q4 trough, FY2028 partial recovery). If memory remains tight, the recovery could be slower, keeping operating leverage below consensus expectations even with strong revenue. (investor.nvidia.com)
• Excluding China Data Center compute from forward outlook can be interpreted as a quality upgrade to guidance rather than a growth ceiling: it reduces reliance on a volatile region and may lower the probability of negative guidance resets, even if it caps a potential upside source. (nvidianews.nvidia.com)