NVIDIA's Huang says AI equipment can last more than 8 years

$50 billion of equipment was described as "the lowest-risk asset"

Huang described the computing equipment that runs AI as data centers (AI factories). He said a gigawatt-class facility costs about $50 billion to build. A one gawatt data center, one gawatt AI factory costs about $50 billion.── Jensen Huang (2026-10-21 11:17 UTC · YouTube "NVIDIA GTC Berlin 2026 Keynote With CEO Jensen Huang" (NVIDIA) at 0:14:32, source)

This figure becomes the premise when banks and investors decide on loans or investments. [Council inference] Huang's statement is a claim that it is safe to use this huge equipment as collateral.

Huang went on to say that the risk of taking on NVIDIA's compute assets is lower than that of any other asset. The risk involved in underwriting NVIDIA's asset in our NVIDIA computing is the lowest of any asset out there.── Jensen Huang (2026-10-21 18:35 UTC · YouTube "Jensen Huang Exclusive: AI Financing Fears & Rising GPU Rents | GTC Berlin" (Deirdre Bosa and Yahoo Finance) at 0:07:48, source)

This statement bears directly on how buyers of the equipment estimate its value several years from now. If the equipment is expected to retain value, funding is easier to raise. But Huang is also a party on the selling side. He said NVIDIA has invested about $10 billion, not $1 billion, in companies so far, and announced investments in 14 more companies this week. [Council inference] Because the seller is also involved in the buyers' flow of funds, this statement needs to be read with a discount.

The assumed equipment lifespan has stretched from 4 years to 8 years or more

On the lifespan of computers, Huang said that they were once thought to last 4 years, but in practice will likely last 8 years or more. people are discovering that what people thought used to be a four-year life cycle really is probably eight to maybe more── Jensen Huang (2026-10-21 18:35 UTC · YouTube "Jensen Huang Exclusive: AI Financing Fears & Rising GPU Rents | GTC Berlin" (Deirdre Bosa and Yahoo Finance) at 0:06:28, source)

If the lifespan doubles, the period over which the equipment's cost is recovered also doubles. For companies investing, the annual burden gets lighter and profitability becomes easier to achieve. Huang says this because he sees strong demand and believes work will keep coming even for older equipment. In the same appearance, he also said the residual value of NVIDIA's computers, that is, their worth after use, is very high.

Yet within the same remarks there is a point that looks contradictory. Blackwell, the current generation, is about 30 times faster than the previous generation, Hopper. The next one, Vera Rubin, is said to be another 30 times faster and to arrive in two years. [Council inference] If each new generation is an order of magnitude faster every two years, it is unclear whether old equipment can hold its value for 8 years. Experts are divided here as well. The indicator of how split the council is on this topic is 0.966, close to the maximum.

On the funding side, a cooler response is also being reported

The same week, another report appeared. OpenAI is seeking $30 billion in funding and has lowered its valuation to $1.25 trillion, but investor reaction is reportedly cool. One report quotes a voice saying "this time is a bit disappointing."

No actual statement within the council touches on this report. [Council inference] The premise that equipment will retain its value rests on AI companies being able to keep raising funds. If funding thins out, the premise wobbles too.

There are also developments on the side of workers. American Express is cutting 5% of its employees, saying it is "preparing for AI." Goldman Sachs estimated that about 10% of workers will be replaced by AI within a few years. This connects to the view that the huge capital investment will be recouped through replacing that much work. However, no actual statement within the council touches on this point, so nothing can be said with certainty.

No one has yet verified that "open models have reached the frontier"

Huang said that of the roughly $600 billion in venture funding that flowed to AI startups, 80% of the companies use open models (AI whose contents anyone can obtain and use). I think $600 billion worth of venture financing went into AI native startups. 80% of those AI native startups are using open models.── Jensen Huang (2026-10-21 18:35 UTC · YouTube "Jensen Huang Exclusive: AI Financing Fears & Rising GPU Rents | GTC Berlin" (Deirdre Bosa and Yahoo Finance) at 0:13:02, source)

If this is true, companies that use AI can get started at low cost without owning huge equipment themselves. On another occasion, Huang also said that open models have reached the frontier level. At the same time, he recommends renting closed models from frontier labs wherever possible. [Council inference] Placed side by side as they are, these two statements sound contradictory.

Who will verify whether open models have reached the frontier, and with which independent evaluation? On this topic only Huang spoke, and there is no other actual statement within the council. Unless a means of verification is shown, it is reasonable for readers to treat this as one company's claim.

The people who gain and the people who bear the burden are not in the same place

Huang said this is the best time in the past 100 years to invest in sustainable energy. this is the best chance to invest in sustainable energy at no better time than you know now in the last 100 years── Jensen Huang (2026-10-21 18:35 UTC · YouTube "Jensen Huang Exclusive: AI Financing Fears & Rising GPU Rents | GTC Berlin" (Deirdre Bosa and Yahoo Finance) at 0:19:31, source)

It is a tailwind for utilities and renewable energy operators. But in towns where gigawatt-class facilities are built, electricity and water use rise. In Australia, backlash against AI data centers is reported to be threatening plans worth $70 billion. The council's Gebru is said to hold the position of criticizing this as an investment that pushes environmental burdens onto the local area (inferred). If a third-party audit confirmed that measured burdens do not increase, this view would be rejected.

[Council inference] Three future benefits can be considered: longer lifespans that make profitability easier, a shorter time from installation to start of production, from 4–5 months to about 1 month, and a wider base of entrants as open models spread. Conversely, there are also three risks: investment on the scale of $50 billion and a supply shortage lasting several years, capital losses if the residual-value premise fails, and the power and environmental burden falling on local communities. The benefits rest on the seller's estimates, while the risks are borne by the people around. This asymmetry is the core of today's story.