Predicting the Future of Tech: Insights from Brian Wang on SpaceX, Tesla, and AI

The host of InvestAnswers introduced Brian Wang, setting the stage for a discussion centered on the pursuit of financial freedom through accurate foresight. The host stated that predicting the future is the most straightforward path to making money and achieving financial independence. However, the host also noted that the real challenge lies in steadfastly adhering to one's predictions, even when widespread doubt and opposition arise.

The host further elaborated on the concept, highlighting that a significant portion of wealth is generated from a small percentage of assets—specifically, the top 0.3% of assets, which are often the most unpopular or 'hated.' This implies that identifying such assets early and having the resilience to hold onto them through various market turbulences is crucial for substantial wealth creation.

Brian Wang discussed his forecasting accuracy using the Brier score, a statistical measure. He explained that a forecaster with a 50/50 chance, akin to flipping a coin and always calling 'heads,' would achieve a Brier score of approximately 0.25. In contrast, Wang claimed a much lower Brier score, around 0.71 in his framing, which he interprets as roughly 90-95% of his predictions being correct.

Wang’s accuracy has been independently verified, with him ranking eighth out of 34,000 forecasters. He noted that his personal trading target for correct predictions is 80%. He maintains highly detailed logs of every prediction and movement to track his performance and refine his models. This extensive record-keeping, spanning 35,000 publicly accessible articles over 20 years, also serves as a valuable training resource for AI models.

His approach to forecasting is rooted in experience and evaluating past successes and failures, prioritizing concrete, physical, and quantifiable data over gut feelings or opinions. He seeks to identify clear physical evidence that offers a predictive advantage. For instance, in predicting SpaceX's superior performance over other rocket companies, the focus is on tangible metrics rather than subjective assessments.

The aim is to base predictions on solid evidence, rather than intuition. This method helps to avoid biases and ensure that forecasts are grounded in observable and measurable realities.

Brian Wang reflected on early forecasting failures from his youth in the 1970s, which shaped his understanding of prediction. One significant miscalculation involved the Space Shuttle's launch costs. As a teenager, he projected a cost of $5 million per launch, a figure that ultimately escalated to around $1 billion.

He attributed this discrepancy to a misunderstanding of NASA's political and economic motivations. Wang concluded that NASA's decisions were driven more by political agendas and government job creation programs than by purely technical objectives. This realization highlighted that technical feasibility alone is insufficient for a prediction to materialize, as broader political and economic factors can profoundly influence outcomes.

Another example cited was the unfulfilled plan for nuclear reactor construction in the 1970s. Despite the technical feasibility of building 1,000 reactors instead of the eventual 100, these plans never came to fruition. These experiences taught Wang that even when a technology is entirely viable, external factors like the absence of a driving force, such as an "Elon Musk" or a dedicated company, can prevent its realization.

The discussion touched upon recent events, including a record $6.3 billion crypto hack, rumored to involve North Korea, and the impact of Elon Musk's online activities. According to the speaker, Musk's 'trolling' led to SpaceX's stock price falling back to its August levels, despite significant positive information about the company emerging since then, indicating market price inefficiencies.

The speaker also addressed a threat by CZ, the founder of Binance, to sell his SpaceX shares. While not closely following the detailed exchange, the speaker noted that the threat was likely not fully carried out. An investigation by the speaker revealed that CZ's stake in SpaceX is held within a fund that requires board approval for sales, making an independent liquidation difficult and unlikely to happen.

This incident underscored how even a single tweet or online interaction can cause a 5-10% drop in a stock's value within minutes, highlighting the volatility and emotional impact on markets, especially in the crypto space.

The speaker discussed the potential for public backlash against autonomous taxi services, particularly in cities like Austin, Dallas, Houston, and Miami, where Waymo, Uber, and traditional taxis are already facing displacement. The speaker predicted that no significant backlash would occur, drawing parallels to past events.

The reasoning was that if taxi drivers in major cities like New York and London previously resisted Uber but ultimately lost that battle, and subsequently did not mount a significant opposition to Waymo's deployment of a thousand autonomous cars, they are unlikely to react differently to a thousand Tesla robotaxis. From a driver's perspective, whether it's a Waymo or a Tesla, the impact of autonomous vehicles is the same.

This perspective suggests that the critical period for resistance to autonomous ride-sharing has already passed, and the market has largely adapted to the introduction of new, disruptive transportation technologies.

The speaker drew an analogy between Tesla's integration of Optimus robots and Costco's strategic adoption of forklifts. Just as Costco redesigned its entire store layout to maximize efficiency around forklifts, achieving five to ten times higher revenue per employee, Tesla is reportedly re-engineering its factories around Optimus.

This comprehensive redesign is not just about the robot itself but about optimizing the entire operational environment to facilitate its use. The goal is to enable Optimus to perform tasks such as lifting pallets, thereby eliminating the need for manual movement of goods to shelves, mirroring the efficiency gains observed in Costco's model. This holistic approach focuses on integrating Optimus not just as a standalone robot but as a central element of the factory's architecture.

The redesign extends to creating a 40-volt architecture and adapting the 'Unboxed' process, indicating a fundamental shift in how Tesla approaches manufacturing. This involves not only the cost of the robot but also the extensive re-engineering required to make the entire system more efficient. The speaker also highlighted the importance of the AI chip within Optimus, suggesting that reducing the cost of this 'brain' from $20,000 to $10,000, with $5,000 allocated for memory, is crucial for cost reduction and scaling.

Tesla is actively changing its infrastructure to accommodate Optimus robots, effectively 'changing the facts on the ground' to create a new operational playing field. In an interview with Joe from Munro Live, it was revealed that Tesla is modifying the charging systems for cybertaxis and their maintenance procedures to be compatible with Optimus.

This strategic adaptation aims to enable Optimus not only to service cybertaxis but also to handle delivery tasks. The expectation is that these robots will emerge from trucks and other vehicles to make deliveries, expanding their utility beyond passenger services like Uber or Uber Eats. This move signifies Tesla's ambition to integrate Optimus into a broader range of logistical and service functions.

The speaker emphasized that this approach represents a significant competitive advantage for Tesla, built upon three pillars: advanced AI 'brains,' sophisticated 'hands' (robotics), and large-scale manufacturing capabilities. These three elements collectively position Tesla ahead of competitors, with the large-scale production capability being particularly difficult for others to replicate quickly. The speaker suggested that only a large Chinese conglomerate with chip-making and manufacturing prowess could potentially compete, but even then, they would be years behind due to their current lag in AI data centers.

Tesla, under Elon Musk's direction, is dramatically accelerating the construction of its data centers, building them in approximately five to six months. This timeframe stands in stark contrast to the industry norm, where data center construction typically takes two years or even longer. Elon Musk expects the 'mini-hardware' build, which will house 220,000 chips, to be completed by November, marking a five-month construction window.

The speaker highlighted that this speed is unprecedented, citing Jensen's praise for Tesla completing 'Colossus 1' and 'Colossus 2' in about a year. Only 20% of other industry players manage to complete similar projects in under two years. This rapid deployment extends to multiple data centers, with 'Macro Heart' being completed in September and 'Mini Heart' by November, alongside another in December, all moving at an astonishing pace.

This rapid build-out also translates into a significantly faster payback period. The speaker compared Tesla's projected six-month payback for its data centers to CoreWeave's target of approximately two and a half years. CoreWeave's model involves securing five-year leases with major clients like Meta or Microsoft, adding a 20-30% markup, and then arranging financing for Special Purpose Vehicles to achieve their payback goal.

Tesla's ability to compress both construction timelines and investment recovery periods positions it as a leader in deploying critical AI infrastructure with remarkable efficiency.

The exceptional speed of Tesla's data center construction is attributed to the engineering talent from SpaceX. Elon Musk reportedly deployed a 'SWAT team' of SpaceX engineers, accustomed to the monumental task of building rockets and launching them into orbit, to this project. Musk purportedly described data center construction as a "piece of cake" compared to Starship development.

This cadre of top-tier engineers is credited with enabling data centers to be built in six months, a stark contrast to the two-and-a-half years typically taken by rival companies. The presenter highlighted that these SpaceX engineers not only deliver projects rapidly but also achieve investment recoupment within six months, with plans to scale production to one data center per month.

Additionally, SpaceX has reportedly improved energy efficiency by 15-20% through design changes, following the 'Wright learning curve.' This expertise also extends to HVAC and cooling systems, contributing to the overall efficiency of the data centers. The current stock price of SpaceX reached $135, aligning with its pre-IPO valuation, while the day-of-IPO price was $152. These engineering advantages are expected to drive the company's growth, with projections of 2.7 GW by year-end and potentially 10-20 GW next year, contingent on GPU and memory availability to meet the infrastructure's energy capacity.

The power setup for the data centers is planned as a diversified mix, moving away from full reliance on the traditional grid. For a 10 GW capacity, an example split includes 4 GW from the grid, 4 GW from natural gas, 2 GW from solar, and 2 GW from battery storage. The battery and solar components work in conjunction to provide greater stability to the power supply.

This multi-faceted approach helps to mitigate grid reliance and manage peak demand times, such as between 6 AM and 9 AM. The presenter also noted that about 100 GW of additional capacity could be integrated into the U.S. grid if data centers were willing to accept being the first to be shut off. This would circumvent the need for stringent Service Level Agreement (SLA) uptime requirements for certain temporary projects.

Furthermore, the presenter stated that Megapacks for 2027 are already sold out, with an anticipation that half of them might be allocated to SpaceX sites. These Megapacks would play a crucial role in stabilizing electricity and protecting GPUs from potential damage due to power fluctuations or outages, thereby enhancing the reliability of the data center operations.

The presenter projected significant upside for SpaceX, with an additional 100 to 200 billion dollars potentially generated from leasing extra data centers without chips. This leads to an overall valuation estimate reaching or exceeding a thousand dollars, based on conservative price-to-earnings ratios and excluding other business segments.

However, the presenter highlighted the sensitivity of these projections to timing. A delay of merely two or three months in deployment or operational readiness would push the projected endpoint to the first quarter of 2028. Despite this, markets are expected to look forward, and consistent performance over three or four consecutive quarters could prompt Wall Street to adjust valuations accordingly.

The current growth trajectory is compared to Nvidia's early post-GPT phase, characterized by near-doubling revenues quarter after quarter. This predictability stems from continuous data center construction, chip production, and deals secured months in advance, creating a rapidly growing and highly foreseeable revenue stream for SpaceX, mirroring Nvidia's past success.

The guest discussed the exponential growth expected in compute demand, illustrating it with a scaling example. Moving from 10 million Muse customers to a billion users, a target Meta aims for, represents a 100-fold increase in user base. If each user's engagement also increases by a factor of 10, the overall compute power required would multiply by 1,000.

The argument concludes that the entity capable of meeting this immense and escalating demand for compute power will capture a significant market share. This growth is also anticipated to precede a substantial increase in profit margins, with potential rises from 20-30% to 60-70%, surpassing even Nvidia's margins.

Such margin expansion, coupled with high demand, would lead to accelerated profit growth, potentially doubling or tripling profits rapidly. The guest also touched upon the broader economic impact, suggesting that a condensed series of industrial revolutions, fueled by such technological advancements, could lead to a 100-fold growth in the entire economy within a short span.

The guest referenced Elon Musk's long-term predictions of quadrillion-dollar market capitalizations, asserting that these figures start to make sense in the context of this projected exponential economic growth, driven by massive energy production for AI data centers.

The guest expressed a high degree of confidence, placing it at 98%, that SpaceX will achieve 13 gigawatts of capacity by the end of 2027. This projection is based on observable construction activities in Texas and Mississippi, including land acquisitions and natural gas pipeline development.

Semi Analysis, an industry firm, has also corroborated this, stating that adding 10 GW on top of the 3 GW already expected online by the end of this year is easily achievable for SpaceX. The guest believes this 10 + 3 GW estimate is highly probable.

Furthermore, the guest anticipates that additional buildings, driven by the demand from 'Dark GPUs' of companies like Microsoft and Google, could add another 4-5 GW to SpaceX's capacity. These additional structures would initially stand empty, awaiting future deployment.

The increased confidence in capacity led to a shift in the stock target from $850 to $1105 after the 3 GW addition. When combining the guest's 98% certainty with Brian's 92% accuracy, the overall probability remains very high, suggesting a robust outlook for SpaceX's compute infrastructure.

The discussion shifted to the challenges faced by AI firms like Anthropic and OpenAI, despite their efforts to commission their own chips from manufacturers such as Broadcom and AMD. The host questioned where these chips would ultimately be deployed, asking, "Where are they going to use those chips? Are they just going to put them in Oracle's data centers?"

This highlights a critical bottleneck: even with proprietary chips, these companies still require infrastructure to host and operate them. This necessitates new partnerships for building and deploying data centers, and each additional partner in the chain inevitably reduces the profit margin for the parent company. The presenter noted that even major cloud providers like Amazon and Google are reaching their limits in compute capacity by 2027-2028, leading them to ration resources even for themselves.

This situation presents a significant opportunity for SpaceX. While other companies might eventually adopt modular, rapid construction methods or even hire former SpaceX rocket scientists to build faster, such a pivot would be challenging and take time. The more difficulties other providers encounter in building out their infrastructure, the longer SpaceX maintains its superior solution and pricing power in the market.

The guest provided a detailed comparison between the Rubin and GB300 chips, highlighting Rubin's superior efficiency and potential market impact. While a single Rubin chip consumes approximately 1.5 times more power than a GB300, it yields a significantly higher inference capacity—three to six times more, and sometimes up to ten times, depending on the use case. To provide a conservative estimate, the guest projected a three to six-fold increase in inference.

This enhanced efficiency translates directly to revenue potential. The guest estimated that revenue per gigawatt could surge from an average of $50 billion to around $150 billion if the Rubin's inference capabilities are fully monetized. The "Rubin Ultra" is expected to offer even greater improvements, substantially boosting the economics of data centers.

Moreover, the Rubin chip is not merely about increased inference; it is also crucial for training more advanced "frontier models." This necessity means that frontier developers will have no alternative but to switch to Rubin, driving its adoption. Since SpaceX is building data centers rapidly, it is positioned to be among the first to deploy Rubin on a large scale, giving it a significant competitive edge in the evolving AI landscape.

Anthropic's compute spending with SpaceX has reportedly seen a significant increase, escalating from $1.2 billion to $3.1 billion per month, according to information from their S-1 filing covering through the end of the year. This substantial rise indicates a growing reliance on SpaceX's infrastructure, and the presenter anticipates that major AI players like Google and others will become significant customers of 'SpaceX Ultra' in the coming year.

The discussion also introduced the concept of 'sovereign AI' as a critical future customer segment. Firms like Palantir are reportedly advising Fortune 500 companies against using public AI services like Anthropic or OpenAI due to intellectual property risks. This advice pushes demand towards private, sovereign AI solutions, where data remains within a company's control.

This trend suggests a future where companies may require dedicated, secure AI compute slots. The presenter envisioned a scenario where figures like Karp might approach Elon Musk to secure "50 slots for 50 sovereign AIs for 50 Fortune 500 companies," highlighting the immense potential of this niche market. This move towards sovereign AI positions SpaceX, with its full-stack approach, as a more stable and attractive partner for businesses concerned with data security and intellectual property.