FUNDA

We're excited to announce that two popular FUNDA Plays, GPU Rental Price Tracker and Thematic Sector Rotation, are now available to all our followers, with no login required. GPU Rental Price Tracker Compare GPU rental prices across providers and lease types, including spot, on-demand, reserved, and capacity block pricing. The dashboard brings cross-vendor price indices and historical trends together to provide a consistent reference for monitoring changes in compute costs. Start here: funda.ai/play/gpu-spot-price… Thematic Sector Rotation Track relative strength and momentum across thematic sectors, compare daily rankings, and monitor shifts in market leadership. The dashboard brings rotation charts, leadership maps, and market-style indicators together to help assess how sector trends are evolving. Start here: funda.ai/play/sector-rotatio… If you find these Plays valuable, subscribe to FUNDA to unlock the full capabilities of our research platform.

Preview|NVDA FY27Q3: HBM Delays May Slow the Rubin Ramp; Near-Term Demand Remains Solid Earlier HBM Delays May Slow the FY27Q4 Rubin Ramp. We expect FY27Q4 Rubin GPU shipments to fall slightly below earlier expectations because of shipment delays caused by prior HBM delays. Customer demand remains solid, and B300 demand should support near-term results. We expect the delayed Rubin shipments to move into 2027. Detailed Report fundaai.substack.com/p/previ…

$TSM reported October revenue of NT$587bn, up 59.7% YoY. Factoring in working days and TSMC’s Q4 revenue guidance, the solid October revenue performance was well within expectations. Separately, regarding rumors that Nvidia is considering order cuts, we view that as pure speculation. Even though Rubin faces slight delays due to HBM technical issues, we do not believe Nvidia has reduced its orders with TSMC.

This past weekend, we got a ton of inquiries from Taiwanese investors asking about space compute. That honestly caught me off guard because this seems to be the first time this has happened

Weekly|Thoughts on Short/Mid/Long Terms, Texas Permitting, Cybersecurity, APP & Unity, Space Compute, Optical Demand After one of the most violent market selloffs in recent years, we want to take a step back and consider the AI sector across different time horizons. Over the long term, we remain excited about AI’s potential. Frontier labs continue to make meaningful progress toward AGI, with larger pretrained models delivering substantial capability gains. As reinforcement learning scales and data accumulates, general-purpose models are also expanding into areas previously served by specialized models, including robotics and video and audio generation. In our conversations with leading researchers, optimism about reaching AGI in 2028 or beyond is higher than we have previously encountered. The medium-term picture, through roughly 2027, is harder to assess. Much of the leading labs’ revenue still comes from AI startups and large technology companies. For startups dependent on venture funding, how sustainable is their spending if macro conditions and funding availability deteriorate? Large technology companies are already optimizing token consumption: where does their spending settle, and what would drive renewed acceleration? Meanwhile, our enterprise surveys have consistently pointed to slower adoption than the market had hoped for. Taken together, these trends raise the possibility of a meaningful gap between the spending growth investors expect and what customers ultimately deliver. On the semiconductor and hardware side, we are hearing more reports from the Asian supply chain of slower deliveries and procurement schedules. These warrant attention, though they do not yet establish order cuts. Are they mainly the result of product transitions and deployment timing, or early signs of tighter spending discipline? And how resilient would infrastructure investment prove if financing conditions tightened further, particularly for neoclouds? These medium-term questions can coexist with a near-term rebound. Following such a sharp selloff, even modest improvements in fourth-quarter revenue expectations could matter if share prices already discount a much weaker spending outlook. A strong model release could provide a more credible catalyst, particularly if better capabilities translate into stronger paid demand and more durable pricing. We are not making a market-direction call. We do think it is important to remain open-minded as expectations reset and new evidence emerges. There are also new areas worth watching closely. Our latest space compute work is one bright spot: progress in thermal design, supply-chain orders and launch planning is making orbital compute a more concrete engineering proposition. We see a path to roughly 0.5 GW of on-orbit capacity by the end of 2028 in our base case, and approximately 1 GW in an optimistic scenario, with thermal validation and launch cadence still critical. Cybersecurity remains another area where we see substantial opportunity. AI is accelerating vulnerability discovery and increasing the demands on defense, remediation and governance. The need is growing, but the timing of monetization matters: more attacks will not necessarily translate into an immediate revenue uplift for listed security vendors. Beyond the AI infrastructure trade, AppLovin and Unity are a reminder that opportunities remain elsewhere. We have followed both companies closely through several quarters, and our channel work has captured important shifts in their competitive dynamics and operating momentum. Their strong results and post-earnings rallies reinforce the value of that sustained research. We will continue looking for and tracking opportunities across our coverage, including those outside the market’s dominant AI narrative. This Week’s Reports LLM 26Q4: Capabilities advance faster than commercial returns. Our deep dive examines frontier competition, falling task costs and AI-assisted research, with roughly 70% of leading labs’ compute spending still going toward uses other than paid inference. fundaai.substack.com/p/deepl… Space compute: Engineering progress makes orbital capacity more tangible. Our supply-chain work identifies 6,000 planned racks and a path to roughly 0.5 GW of on-orbit compute by end-2028 in the base case, or about 1 GW in the optimistic case, subject to launch cadence and thermal validation. fundaai.substack.com/p/deeps… Cybersecurity: Customers still favor existing cloud and security platforms. Our latest checks put remediation, response and governance at the center of spending, with limited near-term budget uplift despite faster AI-driven vulnerability discovery. fundaai.substack.com/p/colum… Texas AIDC: Permitting uncertainty rises after the election. We maintain our 2027 delivery forecast for now, but a prolonged freeze could reduce deliverable supply by another 2–3 GW of IT capacity in our downside scenario. fundaai.substack.com/p/resea… Preview & Review | APP & Unity 26Q3: Both grow through the competitive dispute. Following our channel checks on bidding and data access, AppLovin returned to a beat-and-raise with 55% revenue growth, while Unity’s Vector grew 40% sequentially against guidance of roughly 20%. fundaai.substack.com/p/previ… fundaai.substack.com/p/revie… fundaai.substack.com/p/revie… Review | DDOG 26Q3: Pricing and commitments complicate the guide. Revenue grew 35.2%, with non-AI growth accelerating for a fifth quarter, but minimum commitments and lower renewal pricing held the Q4 growth guide to 25–26%. fundaai.substack.com/p/revie… Review | PLTR FY26Q3: Growth stays strong as bookings soften. Revenue rose 96% and Q4 guidance implies roughly 88% growth; shorter contracts and repricing flexibility offer a possible explanation for softer bookings without establishing weaker underlying demand. fundaai.substack.com/p/revie… INIO: The installed base supports a long service tail. Our initiation examines roughly 44 GW of installed capacity, $8bn of equipment backlog and 10–15-year service relationships, alongside delivery execution and the November lock-up expiry. fundaai.substack.com/p/deepi… Premium Report Snapshot A portion of our research is reserved for Premium subscribers and is not distributed via Substack. Below is a snapshot of what Premium subscribers received this week beyond the Substack feed. Research | LLM: Step-Level Routing Could Pressure Frontier-Model Revenue Deep | BTM Power: Policy Delays and Equipment Delivery Cycles Review | LITE FY27Q1 & Marvell Acquisition: OCS and Laser Pricing Drive Growth Preview & Review | COHR FY27Q1: PhotonLink Agreement Extends Demand Visibility Preview & Review | AMD 26Q3: Management Raises 2027 Server CPU Growth Outlook to About 100% Review | NET 26Q3: Platform Cross-Selling Supports Faster Growth Review | FSLY 26Q3: Growth Slows Outside the Top Ten Customers Review | TSEM 26Q3: Long-Term Orders Extend Through 2029 Review | Kioxia FY27Q2: Stronger Mobile Demand Meets a Softer Guide Review | Ibiden FY27Q2: CPU Demand Drives the Beat Preview | AMAT FY26Q4: Orders Remain Visible Into 2028 Detailed Report fundaai.substack.com/p/weekl…

Our SpaceX update published just before this reported a initial order for 6,000 space-specific racks, and we lay out a bull case of getting 1GW of compute into space by end 2028.

Deep|Space Compute: SPCX Could Reach 1 GW of On-Orbit Compute in 2028 A thermal design is now in place: chips run hotter, and each satellite carries one NVL72 rack. SpaceX has disclosed AI1 at 250 kW peak and 175 kW average with a 160 m² radiator, but not the operating temperature. Working backward from our June radiator table, that implies a radiator surface of ~65–75°C and chip junction temperatures of 80–90°C. Our industry checks corroborate this: the space version runs its chips at junction temperatures of 80–90°C. That is feasible from an engineering standpoint, since chips in terrestrial racks often run above 80°C at full load, and the temperature itself is not a problem for the chips; the binding constraint is memory. In orbit, running hot is worth far more for heat rejection: in a vacuum, heat can only be shed by radiation, and radiative capacity rises with the fourth power of temperature. Rejecting heat at ~40°C, as is customary on the ground, would require 240–290 m² of radiator; by running hotter, a ~2 t satellite carries exactly one NVL72, a Starship carries ~40 satellites, and no on-orbit assembly is needed. The compute payload is locked: a space version of NVL72, codename Kessel. In August, SpaceX announced it is co-designing AI1’s compute payload with NVIDIA, using Vera CPUs and Rubin GPUs. the space rack architecture was finalized in October: it keeps the 72-GPU NVLink domain, compresses the 48U upright terrestrial rack into a ~9U flat enclosure, and cuts the weight to a few hundred kilograms. The first generation targets inference; large-model training stays on the ground. The chips and HBM are standard parts. Radiation is addressed through an external shield, HBM’s built-in ECC and system-level fault tolerance, and heat through a single liquid loop running straight to the external radiator, concentrating demand for advanced materials and lightweight alloys in the shielding, radiator and structural components. Supply-chain orders are in: 6,000 racks for 2027, equal to 1.05 GW. SpaceX has ordered 6,000 space-specific racks for 2027 from a leading server ODM, the sole JDM (joint design manufacturer) on the program, which is due to complete NPI in May 2027, in time for the test satellite, and finish production in December. The 6,000 racks equate to 432,000 GPUs and ~$42–45 billion of hardware value at terrestrial rack pricing; at one rack per satellite, they equal 1.05 GW. Lifetime volume for the program is ~30,000 racks, matching the plan for a first phase of ~30,000 satellites, or ~2.16 million GPUs at 72 per rack, equal to 5.25 GW (7.5 GW peak); the first 6,000 racks are a fifth of that. Constraints on the ground are pushing compute into orbit. Democrats won both houses of Congress in the November 3 midterms, and Texas’s incoming governor has pledged to pause new data centers until the legislature sets rules. Permitting, grid-connection and cost-allocation pressure on terrestrial data centers is rising, and the impact falls mainly on capacity added from 2028, just as Starmind scales. Orbital compute needs no land, grid connection or water, and no state or local data-center permits. As constraints on terrestrial capacity tighten, orbital compute becomes more attractive. Bull case: ~1 GW on orbit by end-2028. In our bull case, Starship has ~220 launches available in 2028 (Starbase expanded to 100, plus 44 from LC-39A and 76 from SLC-37), of which ~70%, or ~154, go to compute satellites. At 40 satellites per launch, that is enough to launch all 6,000 racks, so we estimate end-2028 on-orbit compute of ~1 GW in our bull case (average compute-payload power at 175 kW per satellite). Two checkpoints ahead. Flight 15, expected no earlier than late November, will attempt to catch both stages with the tower; a successful catch and reflight are what support 30–40 launches in 2027. The test satellite launching on Starship in May 2027 will deliver the first on-orbit thermal data and show whether the 175 kW average per satellite is achievable in orbit, which is what our GW estimates rest on. Detailed Report fundaai.substack.com/p/deeps…

We're excited to see what insights @SaaSletter draws from his conversations with our network. Let him know what companies you'd like him to cover, and get in touch with us if you are interested in running your own interviews.

Yesterday, OpenAI released a new cybersecurity-focused model, GPT-6 Cyber. We report on our channel checks, in particular about the delegation of cybersecurity work to frontier labs. $CRWD $PANW $NET fundaai.substack.com/p/colum…

Review|APP 26Q3: From Being Caught Off Guard to Responding Effectively After publishing our preview, we also spoke with an expert at a DSP that competes with both AppLovin and Unity. The discussion helped us check our views and better understand how competition between Unity and AppLovin is changing. The following points increased our confidence in the views set out in our preview. Detailed Report fundaai.substack.com/p/revie…

Deep|LLM 26Q4: Model Capabilities Advance Rapidly, While Near-Term Monetization Remains Under Pressure Competition has expanded from two leaders to a 4+1 field. OpenAI, Anthropic and Google are closely matched on composite benchmarks, but Opus 5.5 is clearly ahead in our hands-on use. Meta and SpaceX are stepping up investment, while the gap between open-weight and frontier models has widened again. Competition is driving price cuts as enterprises shift to optimizing token usage. List prices for the next tier below the top flagships are converging on $4/$20 per million input/output tokens. The effective unit price in Ramp’s sample is about 40% below its March peak. Implementation and ongoing support have become critical to enterprise adoption. Recursive self-improvement (RSI) is advancing model capabilities. AI is playing a larger role in its own development. The first lab to cross the RSI threshold could pull further ahead as its models accelerate subsequent development. Math benchmarks are largely saturated, and AI-generated proofs of open problems are now under review. Foundation models are becoming the basis for physical AI, video production and scientific research. Astra demonstrated robotics capabilities without specialized training on robot actions. Pure diffusion may be less suited to end-to-end video production; the workflow is likely to shift toward LLMs directing and diffusion models rendering. Cyber offense and defense are creating new demand for tokens. Our October 28 cybersecurity report estimates that security-related model revenue could grow from about US$5–7 billion to US$50–70 billion. The investment–revenue gap is widening in the near term, even as the long-term opportunity expands. Models can increasingly address white-collar and blue-collar work beyond coding. Based on the leading labs’ cost structures, we estimate that about 70% of their compute spending is not used for paid inference. We expect that split to persist through the end of 2027. Detailed Report fundaai.substack.com/p/deepl…

Publishing this report has been stressful. Many investors who read last Friday’s Edgewater check pushed back on our conclusions, arguing that we had overestimated budgets and underestimated the impact of take rate. They also asked whether AppLovin had rolled back its IAP model. You can really feel just how much of a consensus short this stock has become. We spent a great deal of time explaining: we had not overestimated budgets. Take rate has recovered from its June low. AppLovin has not rolled back its IAP model; instead, we believe it has updated its IAA model. $APP

We will have an even more interesting update on the data drama coming later today.

Still need some interesting or turnaround names outside AI right? $APP

Most existing data center capacity is held by CSPs, while most new capacity in Texas is being added by neoclouds. CSPs can also add capacity outside the U.S. (see $GOOG recent 15bn investment in Malaysia) bloomberg.com/news/articles/…