What are the three hardest things to buy this year?

What will be the hottest thing in 2023? Jay Chou's concert is bound to be on the list.

Fans who have just experienced the ticket rush and the grandeur of the concert have the most say: a few days ago in Shanghai, scalped tickets were sold for over ten thousand yuan, and shortly after, it became a hot topic due to the official implementation of real-name ticketing and the collective return of scalped tickets.

The four consecutive days of performances directly boosted Shanghai's cultural tourism and catering industries. In addition to the tens of thousands of audience members inside the venue, many fans even stood outside the venue in the rain.

According to data from Ele.me, on the day of Jay Chou's concert on October 12th, the number of food delivery orders surged. Takeaway orders in the Xuhui District, where the venue is located, increased by 44% compared to the previous day, and orders in the nearby Changning District increased by 56%. Balconies in residential areas near the stadium and under pedestrian bridges became "wild stands," and many consumers directly ordered foldable stools, disposable raincoats, beer, snacks, and other items to enjoy the concert experience, and some even sold houses at the entrance of the concert.

Data from Meituan and Dianping also show that during Jay Chou's concert in Shanghai, tourism consumption orders in Shanghai increased by 100% compared to the same period last year. The number of dine-in orders in the Shanghai area increased by over 150% compared to last year.

In addition to Shanghai, Jay Chou also held concerts in Hong Kong, Haikou, Hohhot, Tianjin, Taiyuan, and other cities this year, single-handedly boosting the cultural and tourism economy of several cities. The concert in Haikou attracted 154,600 attendees and generated 976 million yuan in tourism revenue, which is three times the tourism revenue during the Dragon Boat Festival holiday in Haikou.

The influence of the "Jay Chou economy" is evident.

The only two popular products that can compete with Jay Chou concert tickets are NVIDIA's GPUs and Novonordisk's weight-loss drugs.

In High Demand! GPUs are Selling Like Hotcakes

Since the emergence of ChatGPT, the popularity of artificial intelligence has exceeded everyone's expectations. But the tech companies that made money before training large models are NVIDIA, which sells GPUs to them.

As of the end of the second quarter of this year, NVIDIA, the AI water seller with a market value of over a trillion dollars, earned $10.3 billion by selling data center hardware. This is roughly equivalent to the shipment of about 300,000 flagship GPUs H100.

Total revenue doubled to reach $13.5 billion, setting a historical record, and net profit skyrocketed by 843% to $6.2 billion. The performance far exceeded expectations, leaving a group of Wall Street analysts stunned: they knew NVIDIA was making money, but they didn't expect it to be this profitable.

What's even more crucial is that this is far from the ceiling that NVIDIA can reach.

Currently, the price of an NVIDIA H100 is between $25,000 and $35,000, and they are in high demand. Elon Musk even joked during a live broadcast once that the H100 is harder to buy than drugs. According to GPU Utils' speculation, the potential total value of NVIDIA's GPU orders may exceed $20 billion, with a supply gap of up to 430,000 units for the flagship GPU H100.

From startups to large tech companies, academic institutions, and wealthy oil-producing countries in the Middle East, everyone is eagerly waiting for NVIDIA to deliver. In August, there were reports that H100 orders were already queued up until Q1 or even Q2 of next year. OpenAI CEO Sam Altman has complained more than once that the limited supply and insufficient computing power of GPUs have affected the speed of ChatGPT. There were rumors before that OpenAI found H100 too expensive and, coupled with impatience, planned to develop its own AI chip.

It is not an exaggeration to say that NVIDIA's GPUs are like tickets to a Jay Chou concert, both in high demand and in short supply. Whatever is released is far from enough, and as many as there are, they can be sold.

NVIDIA CEO Jensen Huang also stated:

"Our current shipment volume is far from meeting the demand."

The root cause is insufficient production capacity.

More precisely, it is the two key technologies that are bottlenecking NVIDIA - HBM memory and CoWoS packaging - that are in short supply, causing the leather jacket-wearing leader to be anxious.

Especially the CoWoS packaging exclusively supplied by TSMC. In addition to the explosive growth in demand from the AI sector, TSMC also needs to meet the advanced packaging needs of other customers such as Apple and Amazon. Since last year, TSMC's demand for advanced packaging capacity has almost doubled, and even with full production capacity, it is difficult to bridge the supply-demand gap.

In the past few months, TSMC has squeezed out factory space in Hsinchu Science Park, Central Taiwan Science Park, Southern Taiwan Science Park, and Longtan to increase CoWoS production capacity. The Zhunan testing and packaging plant has also increased the construction of advanced packaging production lines to meet the surging demand.

In addition, according to the information previously disclosed by NVIDIA, the shipment volume of H100 next year may be more than 1.5 million units, about three times the annual shipment volume of this year. According to NVIDIA's CFO Colette Kress, regarding the demand for CoWoS packaging, NVIDIA has started to seek other suppliers besides TSMC, and it is expected that the supply will gradually increase in the coming quarters.

Bernstein analysts previously stated that with the improvement of TSMC's packaging capabilities next year, NVIDIA may generate $75 billion to $90 billion in data center and AI chip revenue in 2024, while the consensus expectation of Wall Street analysts is $42 billion.

Bernstein predicts that NVIDIA's performance will continue to grow and may continue to rise for at least the next 12-18 months.

However, it is worth mentioning that tech blogger Pete Warden believes that NVIDIA's GPU dominance is only temporary. The current achievements are actually based on the current industry's mainstream demand for LLM training. If inference becomes more mainstream in the future, the winner will shift from NVIDIA to CPU manufacturers. Warden pointed out that only a few tech giants have the capability to deploy large language models in practical applications. Most companies are still in the early stages of development and require a large amount of data, hardware, and talent to complete the training of LLM.

In addition, for developers of large models, NVIDIA GPUs are the most concise and efficient choice. It is much easier to operate than competing products such as AMD OpenCL, Google TPU, and Cerebras. The software stack is more mature and there are more examples, documents, and other resources available.

Moreover, it is much easier to find engineers familiar with NVIDIA GPUs, and it has better integration with all major frameworks. Coupled with the synergy of the CUDA platform, NVIDIA has completely achieved a winner-takes-all situation.

However, with the development of the entire AI industry, the computational power used by enterprises to run models based on user requests will exceed the training cycle. From this trend, the focus on the hardware side will shift to reducing inference costs. For user-facing applications, the most important task is to reduce latency.

This is not the strong point of GPUs, but rather the domain of CPUs.

Warden said that although CPUs are "laughably slow" in large model training, the requirements for hardware and workloads are very different when inference dominates the entire AI budget. CPUs have more mature development tools and communities than NVIDIA, and the unit operation cost is much cheaper than GPUs. More importantly, the model weights are fixed and can be easily replicated on a large number of machines during initialization.

Enterprises focus on increasing revenue and reducing costs. More users mean more inference demands. The inference cost of CPUs is lower than that of GPUs, so its demand will inevitably surpass GPUs.

Warden predicts that the winners of this shift will be traditional CPU platforms such as Intel x86 and Arm.

Weight loss drugs are popular, Novo Nordisk advances to the top of European stocks

Although NVIDIA GPUs are popular, with the gradual decline of AI speculation in the third quarter, the attention in the capital market is gradually shifting to the new king - GLP-1 class weight loss drug manufacturers.

As a groundbreaking product that allows people to "lose weight while lying down," GLP-1 weight loss drugs can be regarded as a blessing for one billion obese patients worldwide. Specifically, in China and the United States, 15% and 40% of the population respectively suffer from obesity, and the huge market space also means enormous business opportunities.

According to a report released by Goldman Sachs, the market size of global anti-obesity drugs may grow to about $100 billion by 2030.

With the continuous release of clinical trial data, GLP-1 drug companies have received more positive news: weight loss miracle drugs may not only have the effect of weight loss, but also can treat more than ten diseases such as diabetes, kidney disease, dementia, Parkinson's disease, sleep apnea, NASH (non-alcoholic steatohepatitis), psoriasis, and alcohol addiction, showing great potential to become a "panacea."

Currently, the GLP-1 weight loss drug market is basically monopolized by Danish pharmaceutical company Novo Nordisk and American pharmaceutical company Eli Lilly. Novo Nordisk, which holds the hot-selling weight loss drugs Wegovy and Ozempic, has even surpassed LVMH Group and become the top stock in the European stock market. The rapid growth of Novo Nordisk's stock price has even directly affected the exchange rate and interest rate of the Danish krone. From last year, the trend sparked by Semaglutide, a weight-loss drug developed by Novo Nordisk, has caused a frenzy among people trying to lose weight, and soon the product started to run out of stock.

The exaggerated demand for weight-loss drugs has caused shortages of Wegovy, the weight-loss version of Semaglutide, and Ozempic, the diabetes version, both of which are under Novo Nordisk. Genuine diabetes patients can only receive injections once a week instead of once a day as before.

It is worth noting that Wegovy has not yet been covered by insurance in the United States, and the price for a year's treatment is as high as $16,000, but it is still highly sought after by countless obese patients. With the increase in subsequent clinical trials and the inclusion of Wegovy in medical insurance by the FDA, the demand will only become more insane.

Currently, Novo Nordisk's diabetes products are facing a shortage across the board, and the shortage has even spread to the previous generation GLP-1 product, Saxenda. The drug that diabetes patients rely on for their lives has been snapped up by people trying to lose weight, leaving them in a "no medication available" situation.

According to the latest drug shortage list from the FDA, the supply of Saxenda will be limited until the end of 2023 due to the surge in demand. The problem of drug shortages for diabetes patients will become increasingly serious.

Although Novo Nordisk is ramping up production, the supply of weight-loss drugs is still far from meeting the needs of those who hope to lose weight and get rid of the troubles of obesity by taking medication.

Overseas leading companies have achieved impressive results, and Chinese pharmaceutical companies, which have huge market space, will undoubtedly also make their mark in this field.

Currently, domestic pharmaceutical companies are actively laying out GLP-1 weight-loss drugs: Huadong Medicine has obtained approval for the domestic versions of Saxenda for diabetes and weight-loss indications, and Qilu, Huadong, Union, and Lizhu Medicine have entered Phase 3 clinical trials for Semaglutide. In terms of multi-target layout, Xinda Biotechnology and Hengrui Medicine are leading the way in domestic progress, with both companies entering Phase 3 and Phase 2 clinical trials, respectively.

According to statistics from Debon Securities, there are already more than a dozen domestic GLP-1 drugs that have been approved for the market, and several of them have layouts for obesity indications. The future of domestic GLP-1 weight-loss drugs is promising.

Conclusion

From the crowded Jay Chou concert to the hard-to-buy H100 and weight-loss drugs, the price fluctuations reflect the constant law of supply and demand in the market. Whether it is popular music or technological changes, the old guard always gives way to the younger and more compelling newcomers. Trendy concepts will fade away and be replaced by a new round of concept hype.

Why not take a guess at what the market will hype after AI and weight-loss drugs?