PodBrowser
Moonshots - Peter Diamandis

NVIDIA's $96.2B Quarter, China's 200,000 Fake Accounts, & OpenAI's New Chip | EP #284

Saturday, 29 August 2026 · 4 min read · Listen to the episode ↗

NVIDIA reported $96.2 billion in quarterly revenue and forecast $108 billion, while OpenAI’s reported custom “Jalapeno” chip with Broadcom could shift some workloads from NVIDIA; its performance advantages remain company claims. X said it found about 200,000 suspected China-linked inauthentic accounts, including 200 posting about U.S. AI and energy policy, but the foreign-operation attribution was not established.

NVIDIA reported quarterly revenue of $96.2 billion, up 106% year over year, and forecast $108 billion for the current quarter. Jensen Huang predicted 70% growth in 2028, versus Wall Street’s 44% consensus. Dave Blunden said NVIDIA could remain sold out for the foreseeable future and put its gross margins at 80–85%, while describing the company as holding 90% of revenue and 95% of profits in the relevant market. He identified dependence on TSMC as its largest vulnerability, noting that roughly a third of TSMC manufacturing goes to NVIDIA and another third to Apple. The speakers questioned whether NVIDIA-backed customer financing supports some demand, but did not establish that it amounts to a wash trade; they said investors need more transparency and allowed that any resulting downturn might be a “mini winter.”

OpenAI’s reported “Jalapeno,” developed with Broadcom, marks an effort to move some workloads from purchased NVIDIA chips to custom silicon. OpenAI reportedly claimed 1.5–1.9 times more AI work per watt and up to 3.6 times lower end-to-end latency than NVIDIA’s GB200 and GB300, as well as nearly 54 times the GPT OSS throughput per second per user of what it called the existing best. Those are company claims, not independently established results. The speakers estimated inference accounts for about 90% of compute and argued it is easier to shift among vendors than training. Training next-generation models still benefits from tightly connected GPU clusters: NVIDIA’s CUDA software ecosystem is difficult to port and its interconnects help link large systems, while inference can be distributed and may favor different chips.

X’s safety team said it had identified a suspected China-linked bot farm of about 200,000 inauthentic accounts, including 200 that posted about U.S. AI and energy policy. The posts argued that data centers raise household electricity bills and strain the grid, sometimes using AI-generated cartoons. Peter Diamandis called the activity a foreign influence operation intended to suppress U.S. data-center deployment, but the discussion did not establish that attribution, and the host questioned whether X’s figures were fully certain. Quincy, Washington, was offered as a counterexample to claims that data centers only burden communities: a CNN report said its poverty rate fell from 29.4% to 6.2% as data-center tax revenue helped fund a high school, hospital, library, and public-safety facilities, while residents’ property-tax rates decreased. Speakers suggested operators could also provide or subsidize community power and local services.

The debate over AI and jobs turned on the difference between automating tasks and eliminating occupations. A Washington Post analysis argued that past technology shifts created more jobs than they destroyed, while Goldman Sachs warned professional-services firms could face existential risk if AI advances faster than senior partners adapt. Salim disputed that forecast, arguing that greater business volatility could increase demand for services as firms move from selling hours to outcomes. A Principal Financial Group survey he cited found 4% of companies expected AI to reduce both staffing and wages, compared with 31% expecting both to increase. Among companies that cut staff, 1.4% attributed the cuts to AI automation; over the preceding three months, 52% increased staffing, 30% maintained it, and 12% reduced it.

Flock Safety’s systems reportedly serve more than 6,000 communities in 49 states and generate over 20 billion vehicle scans a month. Its expansion from license-plate cameras into video, audio detection, drones, and related software illustrates how surveillance can become a network rather than a single camera. Liam warned that connected systems can track movement across public spaces and make following people increasingly cheap; reported abuses include officers using camera access to track current or former partners and deaths after license plates were misread. The speakers called for governance that can keep pace with the technology and constitutional-level public debate. They also raised risks from AI companions, which can acquire intimate knowledge and may cause lasting harm if addictive use begins during childhood, while recognizing potential benefits for lonely or disabled people and for elder companionship, therapy, and social-skills coaching.

Other discussions highlighted early-stage opportunities and unresolved constraints. The FDA approved Deraxaniracib, described as the first RAS inhibitor for metastatic pancreatic adenocarcinoma; in previously treated patients, median survival rose from 6.7 to 13.2 months and tumor response rates from 11% to 32%, though the drug is not a cure. A fluoride-free biomimetic gel restored enamel-like properties in tests on extracted human teeth, but human trials had not begun. The speakers also argued that China’s rapid solar deployment reflects manufacturing, supply chains, permitting, and execution, while U.S. grid interconnection, transmission, and permitting remain bottlenecks.

This summary was generated from the episode transcript and can contain mistakes.